Are you ready for ChatGPT Ads? The new rules of SI advertising

A user may no longer stop at typing the name of the product they want to buy. In the same conversation they can explain where they will use it, how much they can spend, which features they cannot do without and why they were unhappy last time.

A search for “desk” can become a detailed need: “I work in a small room; I’ll be using two monitors, I need to be able to move the desk when necessary and I have no one to help me assemble it.” A conversation like this can create an advertising context that carries not only the product category but also the conditions that will decide the purchase. Using that context for matching does not mean the private conversation is opened up to the advertiser.

This is where the strategic importance of ChatGPT Ads begins. Brands can appear with an offer that helps the user at the very moment they are working through a decision. The commercial value, however, emerges after the ad is shown: is the product genuinely a fit? Is the information up to date? Can the user move on to a trustworthy next step? Does the customer acquired actually contribute to the business?

Preparing for SI advertising means building four capabilities together: understanding customer need, providing accurate product information, creating a trustworthy experience and measuring the outcome.

At Webtures, we see this shift as a growth area that requires marketing to work more closely with data, technology and commercial operations. This guide starts with how ChatGPT ads work today and covers how brands can prepare, which investments they should prioritise and where the link to agentic commerce is made.

VerifiedThe platform claims in this guide were checked one by one against primary English-language sources on 21 September 2026, reviewed again following OpenAI’s product update on 30 September 2026, and once more on 6 October 2026 following OpenAI’s 5 October announcement. The evidence boxes beneath the relevant sections show the finding and its source. Verified means the claim is written in a primary source. Live, Limited indicates that the feature is real but running in a limited scope. Partly Available shows that the approach exists but may not be usable in Türkiye. Changed Course is used where the source reveals a situation different from the common assumption.Verification date: 21 September 2026; updates: 30 September and 6 October 2026. Platform documentation changes frequently; conditions should be re-verified when planning campaigns.

What is ChatGPT Ads and what does it offer?

ChatGPT Ads is the advertising system that serves sponsored ads in eligible user experiences. Ads appear in areas that are visually separated from ChatGPT’s answer and clearly labelled as sponsored.

For a brand, this channel can build visibility at different decision moments: product discovery, weighing alternatives, researching services and getting ready to buy. If a user’s conversation carries a suitable commercial context, a relevant product or service can become a candidate for an ad.

Ad matching and organic answer generation are separate processes. By allocating budget, an advertiser does not buy the right to change ChatGPT’s independent answer or to place its brand in a particular position within that answer. Sponsored visibility does not mean OpenAI endorses the business either.

This distinction matters for setting the right campaign objective. What you should expect from the ad is reaching a suitable customer through paid placement and measuring the next commercial step. Organic SI visibility, brand reputation and compatibility with agents that can transact must be managed separately.

Where do ChatGPT ads stand today?

OpenAI announced its approach to advertising on 16 January 2026; the first US test began on 9 February 2026. As of September 2026, the product has become a beta advertising platform with campaign setup, budget management, conversion measurement, catalogue ads and automation capabilities.

VerifiedOpenAI announced its approach to advertising on 16 January 2026 and launched its first US test on 9 February 2026. Ads are shown on the Free and Go plans; Plus, Pro, Business, Enterprise and Edu accounts are ad-free. Ads are not shown to accounts identified as belonging to users under 18, or in temporary chats.Sources: OpenAI, Our approach to advertising and expanding access (16 January 2026); Ads in ChatGPT help page; TechCrunch (9 February 2026).

Beta does not mean advertising is just an idea. But the feature scope, access and delivery behaviour continue to evolve. A screenshot or campaign experience from an earlier period may not explain all of today’s conditions.

That is why three questions need to be asked separately when assessing the channel: Is the feature documented? Is it available in our account? Does it produce enough volume and economic return in our campaign?

What changed in ChatGPT Ads at the end of September 2026?

With an update announced to advertisers on 30 September 2026, OpenAI made it easier to scale product feed campaigns, read conversion results and manage account names; the markets where ads are live also expanded. Five changes directly affect some of the recommendations in this guide.

• Product feed campaigns can be set up in bulk: using the Product feed template downloaded via Campaigns > Bulk uploads > Bulk sheet in Ads Manager, multiple product feed campaigns and ad groups are created from a single spreadsheet; ad templates are generated automatically.
• Product review status is now visible: the Products tab shows where each product stands in review and the reasons for any rejection.
• Off-objective conversions can be reported via the API: the Insights API also returns attributed conversion events outside the campaign’s selected objective. Attributed purchases can be reported while optimisation targets a different event; the campaign’s objective and optimisation setting do not change.
• Account, brand and legal names are now separate: under Settings > General, the account name used in the dashboard, the brand name shown next to the ad and the legal name used for billing and business verification are managed separately.
• Southeast Asia and Taiwan have opened: ChatGPT Ads went live in Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam and Taiwan.

Changed CourseIn the first version of this guide, relying on the help centre, we wrote that bulk upload did not support creating product feed campaigns. On 30 September 2026 OpenAI announced this flow and the review status in the Products tab; on the same date, the help centre’s quickstart article still showed the old restriction.Sources: OpenAI Ads Platform product update (email to advertisers, 30 September 2026); Quickstart: Launch your first campaign (viewed 30 September 2026).

VerifiedThe Insights API conversions endpoint returns campaign-objective conversions and, optionally, attributed standard and custom events; the events do not need to be the campaign objective.Source: OpenAI Ads API, Insights.

VerifiedThe advertiser name entered at sign-up is written to all three fields and can later be edited separately. Changing the account name triggers no review; changing the brand name starts a brand review and pauses delivery until the new name is approved; changing the legal name restarts business review and may pause delivery.Source: Ads Manager Account Setup.

VerifiedOn 30 September 2026, 63 countries were marked as available on the self-serve access list; Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam and Taiwan are on the list.Source: Ads Manager Availability (viewed 30 September 2026).

For brands, the practical consequences fall under three headings: brand name changes should be planned independently of the campaign calendar; catalogue-heavy advertisers should consider building a category-based campaign structure with the spreadsheet template; business intelligence teams should start reporting business events such as purchases and applications separately from the optimisation objective.

What changed in ChatGPT Ads in October 2026?

On 5 October 2026, stating that ChatGPT reaches 1.2 billion people every week, OpenAI announced updates to the advertising platform under three headings: a new visual ad format, expanded measurement partnerships and work on brand suitability.

• Visual ad format: visual ads showing product inspiration, product use or the experience a product makes possible will first be tested while ChatGPT is generating images. The ad will be clearly labelled and kept separate from the generated image. The test begins in the US in October 2026 with an initial group of advertisers.
• Conversion data transfer: Hightouch, Tealium and LiveRamp integrations make it easier to send conversion data from an advertiser’s existing systems to ChatGPT Ads.
• Attribution partners: Conversions API, reporting and click attribution are supported with AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin; Fospha, Measured and INCRMNTAL are full-funnel measurement partners.
• Incrementality: OpenAI notes that this work is at an early stage; geo-based experiments are being explored with Haus, Measured and WorkMagic.
• Brand safety: eligible advertisers can use excluded phrases (Negative Phrases) for narrow, brand-specific placement restrictions; brand suitability assessment pilots are being developed with DoubleVerify and Integral Ad Science.

Live, LimitedThe visual format will be tested only in the US and with an initial group of advertisers; Negative Phrases is open only to eligible advertisers; incrementality experiments and the DoubleVerify and IAS pilots are at an early stage. The announcement contains no timeline for Türkiye.Source: OpenAI, Building advertising for the way people use AI (5 October 2026).

The practical consequence for brands: teams that already manage conversion data through Hightouch, Tealium or LiveRamp can add the integration to their measurement plan; visually driven categories can prepare a library of inspiration, usage and experience images; corporate procurement teams should add the results of the independent verification pilots to their watch list.

Is it possible to advertise on ChatGPT from Türkiye?

As of 21 September 2026, Türkiye is among the countries with direct access to the ad management dashboard. Being on the country list, however, does not mean every business or every product can automatically run ads.

VerifiedTürkiye is on OpenAI’s self-serve country list. On 30 September 2026, 63 countries were marked as available on the list (56 on 21 September). The criterion is the country of the legal entity that places and is billed for the ads.Source: OpenAI Ads Manager Availability (viewed 30 September 2026).

Readiness should be assessed at four different levels:

1. Advertiser eligibility: is the legal entity that places and is billed for the ads in a supported country?
2. Account eligibility: are business details, verification, user permissions and payment complete?
3. Product eligibility: does the product being promoted comply with advertising policies in the target market?
4. Campaign eligibility: is the chosen targeting supported in the account, and can the ad be delivered?

For some new accounts, international targeting may depend on conditions such as identity verification and local campaign spend. The country, billing currency and time zone selected when opening the account should also be checked carefully.

VerifiedThe country or region, billing currency and time zone selected when creating the account cannot be changed later. In addition, some new self-serve accounts can initially advertise only in their own country.Sources: Ads Manager Beta Account Setup; Create Campaigns.

When assessing the potential in Türkiye, decisions should rest on country-level ad inventory, product category and real campaign data. ChatGPT holding a high share of SI-referred web traffic in a country does not show that the population uses ChatGPT at the same rate, or that ads will convert at the same rate.

For a brand in Türkiye, the first concrete step is to verify access and measurement conditions with a suitable product group. Large-scale budget decisions should follow that verification.

Who sees the ads, and how should reach be planned?

The addressable audience for ChatGPT ads is not a single pool covering every user. Subscription plan, age eligibility, country, user settings and display conditions can all affect reach. Ads are not shown to accounts identified as belonging to users under 18. Temporary chats are also among the experiences where no ads are shown.

VerifiedThe distinction between experiences with and without ads is defined in OpenAI’s own documentation: ads can appear on the Free and Go plans and do not appear in temporary chats.Source: Ads in ChatGPT.

Multiplying the global user count by a percentage is not enough to estimate ad reach. What share of eligible users in the relevant market have conversations related to your product? How much delivery can your campaign budget and bid secure? What share of incoming users actually fall within your service area? These questions should be answered in the pilot.

There is another important distinction on the B2B side. The presence of professional users does not prove that you are reaching the audience you want in terms of job title or purchasing authority. That has to be checked against CRM outcomes such as qualified leads, company fit and sales opportunities.

SI advertising is not converging on a single model

Different platforms place SI at different points in the advertising experience. Some serve ads inside the conversation with the user; some answer product questions; others use SI interactions as a personalisation signal within their existing ad systems.

This table compares the direction each platform is heading in; it does not mean every feature is available in Türkiye or in every account.

Partly AvailableAvailability in Türkiye differs across the approaches in the table. Google’s ads placed inside AI Overviews are live in 12 countries, and Türkiye is not on that list; ads shown above and below AI Overviews, however, run in more than 200 markets. Amazon’s sponsored prompt formats are available only in the US market. Meta has used interactions with its SI assistant as a recommendation and ad signal since 16 December 2025; the European Union, the UK and South Korea are out of scope, Türkiye is in scope. Perplexity has dropped its ad trials entirely.Sources: Google Ads Help, ads in AI Overviews; Amazon Ads, sponsored prompts; Meta Newsroom (1 October 2025); Financial Times report, as relayed (18 February 2026).

It is valuable for brands to build a shared foundation of product and customer information. But carrying the same creative, budget logic and success metric across every SI surface does not make a sound media strategy. Each channel’s user behaviour and commercial role must be tested separately.

Trust is the core design question in SI advertising

While talking to an SI assistant, a user may share their budget, their concerns and the points they cannot decide on. In this setting, it matters that the ad shown is not confused with an independent recommendation.

ChatGPT’s approach to advertising is defined around the principles of answer independence, conversation privacy, user control and long-term trust. For the advertiser, the practical translation is to present the commercial message clearly and verifiably.

Phrases such as “ChatGPT’s pick”, “Approved by SI” or “Definitely the best option for you” can create the impression of a recommendation or endorsement that does not exist. The brand voice must be consistent with the sponsored nature of the ad.

VerifiedAnswer independence is a written OpenAI principle. According to the documentation, ads do not influence ChatGPT’s answers, and advertisers cannot shape, rank or change them. A sponsored result appearing does not mean OpenAI recommends that business either.Sources: Ads in ChatGPT; OpenAI, Our approach to advertising and expanding access.

The setting in which an ad appears is also part of trust. OpenAI’s placement rules assess whether a conversation is suitable for advertising and do not show ads in emotionally sensitive or inappropriate contexts. For narrow, brand-specific restrictions, eligible advertisers can define excluded phrases (Negative Phrases).

Live, LimitedOpenAI is developing brand suitability assessment pilots with DoubleVerify and Integral Ad Science. The independent partners assess how safety standards are applied in a controlled test environment, without access to real user conversations. The pilots are not yet a verification report made available to advertisers.Source: OpenAI, Building advertising for the way people use AI (5 October 2026).

Another dimension of trust is being able to state a product’s limits. An accessory that is not compatible with a particular device, a delivery that will not arrive by a given date or a service that carries an extra charge should be stated clearly. Pulling users towards an offer that does not suit them may raise click-through rate, but it can lower commercial quality.

New rule: define customer need in its context

ChatGPT Ads draws on signals such as conversational intent, ad copy, the landing page and context hints added by the advertiser. Context hints are additional information that explains who the product is useful for and in which situation.

Rather than filling this field with a long keyword list, you need to describe real customer needs clearly. Hints are not hard delivery commands. Restrictions such as “show only in this city” or “exclude these users” must be applied through the supported targeting controls.

VerifiedContext hints are defined at ad group level and describe the conversations, topics and terms to which the product or service may be relevant. In matching, signals such as the context and intent of the conversation, the landing page, the headline and the copy are assessed together.Sources: Ads in ChatGPT: The Basics; Create Ad Groups.

These examples must be backed by features that are genuinely offered. Adding capabilities the product does not have in order to match more conversations creates false expectations.

When preparing the needs map, sales calls, support tickets, on-site searches, return reasons and consented customer research can be assessed together. The aim is not to guess people’s private conversations but to make clear which problems the business solves well.

New rule: build the campaign in a structure that can learn

Putting every product into a single ad group at the start can make it hard to see which need is delivering results. An overly fragmented structure, on the other hand, reduces the data each group receives. We map where Google Ads habits break in this channel, point by point, in ChatGPT Ads vs Google Ads.

Three levels can be used to organise a campaign:

• Campaign: business objective, market and budget.
• Ad group: related products or a shared customer need.
• Ad: different value propositions and creative executions for that need.

A furniture brand, for example, can address small-space living, ease of moving and long-term home working with separate messages. These groups may contain the same products; what matters is that each tests a different reason to decide.

Choosing a small number of meaningful hypotheses for the first pilot makes the results easier to read. The assumption “our customers care about this feature” must be kept apart from the finding “the ad that explains this feature brings in more qualified demand”.

New rule: creative should make the decision easier

In a conversational setting, an ad’s value can grow with its contribution to the task the user is trying to complete at that moment. Explaining what is offered, who it suits and what concrete benefit it delivers is a stronger test hypothesis than an abstract brand promise. We explain how headline, image and context hints work together in creative assets for conversational ads.

The following structure can be used when preparing an ad:

1. Define the customer’s need clearly.
2. Show the feature that answers that need.
3. Explain the feature’s practical benefit.
4. State the important condition that affects the decision.
5. Offer a meaningful next step.

Alongside “Inspire your living space”, a concrete message such as “Extendable table for small dining areas” can be tested. The description should complete the headline with dimensions, assembly or delivery information rather than repeating it.

The image must support the same offer. A crowded design that does not show the product’s scale, use or distinguishing feature can make the decision harder. In SI-generated images, the real product’s dimensions, material and appearance must not be altered.

Attracting attention is not the only measure of success here. The fit between the expectation the ad creates and the reality the customer sees on the product page and after the sale must be maintained.

How should ad formats and sponsored agents be assessed?

The basic ad card consists of elements such as the business name, logo, headline, description, image and landing page. Product feed campaigns make it possible to use the products in a catalogue to generate ads. Sponsored Agents, meanwhile, tests a different experience in which the user moves from the ad into a conversation with a business’s SI representative.

As of the research date, Sponsored Agents is in a limited alpha in the US with selected advertisers. It is not generally available and early access requests are not being accepted. The conversation with the sponsored agent is kept separate from the user’s original, independent ChatGPT conversation.

Live, LimitedSponsored Agents exists as a real ad format; the user can move from the ad into a conversation with the business’s SI representative. OpenAI states that this is currently a limited alpha test run only with selected advertisers and that it is not accepting early access requests. In addition, product carousels showing multiple products in a single ad unit were introduced in August 2026.Sources: Sponsored Agents in ChatGPT Ads; Digiday (6 August 2026).

What this development means for readiness: a brand’s product information must be able to produce consistent answers to different questions. The dimensions, compatibility, service scope, delivery and return conditions a sales adviser would be expected to know must be organised systematically.

Not every brand needs to build its own agent straight away. The first piece of work is to establish how reliable the existing information is and where the gaps are. That information foundation can be reused for the website, customer service, ad production and future agent experiences.

How does SI-assisted campaign management change responsibility?

The updates announced on 16 September 2026 include creating, updating and analysing campaigns in natural language; ad copy and image suggestions; and optional copy adaptation and translation. A HubSpot integration and a Shopify advertising app also connect to this workflow. The Shopify app was announced for US-based stores, with expansion to supported international markets planned for 23 September.

VerifiedThe 16 September 2026 announcement is verified: setting up, updating and reviewing campaigns in natural language; copy and image suggestions generated from the landing page; optional copy adaptation and translation. HubSpot was announced as OpenAI’s first CRM partner for ChatGPT Ads. The Shopify app opened to US merchants on 3 September 2026; expansion to international markets was announced for 23 September, meaning it had not yet happened when this article was published.Sources: HubSpot investor relations release (16 September 2026); OpenAI help page.

These developments can speed up operational work. Turning a brief into an ad, or preparing the first analysis of a report, may require less manual effort. Management’s job is to decide which suggestion is commercially right and to set the limits on what can change.

Automatic translation, for example, can produce a warranty statement that does not carry the same legal or commercial meaning. Copy adaptation can spread a feature that belongs to one product across a whole category. Image generation can show an accessory as included with a product when it is not.

That is why approved claims, product information that must be protected, fields that may be changed and review owners should be defined before automation. The organisation’s approval structure must be maintained for publishing automated suggestions, increasing budgets and changing accounts.

The landing page is part of ad performance

If a user clicks an ad while discussing a specific need, the landing page must quickly show how that need is met. Searching for the product again on a generic home page, or piecing together offer conditions from different pages, adds new friction to the decision.

A good landing page answers these questions:

• What exactly is being offered?
• Which users and conditions of use does this offer suit?
• What is the price, or the pricing logic?
• Are there extra items that change the total cost?
• How do delivery, installation or the start of service work?
• What are the return, cancellation, warranty and support terms?
• How much time and information does the next step require?

In e-commerce, landing on the right product variant matters; for service businesses, showing the right scope matters. A “free assessment” ad must not turn into paid consultancy on the page. A service not offered in a city must not be left vague in order to collect applications from users in that region.

Mobile usability, page speed, form errors and problems in the checkout steps should also be included in the assessment. Ad optimisation and conversion optimisation must work around the same business objective.

How do you prepare bot access and technical visibility?

It may not be enough for the user to be able to open the ad’s destination page. The platform’s review systems must also be able to reach it. For ChatGPT Ads, OAI-AdsBot must be able to access the destination page; OAI-SearchBot is assessed separately in terms of search and content understandability.

Access problems do not come only from the robots.txt file. A firewall, CDN, bot protection, CAPTCHA, mandatory login, geographic restriction or rate limit can also prevent the page from being reviewed. If product images are hosted on a separate server, access to them must be checked too.

A bot access policy should be built around function. Crawlers used for ad review, search visibility and model training do not serve the same purpose. It is stated that data collected by OAI-AdsBot is not used to train foundation models.

VerifiedOpenAI’s crawler guidance for advertisers makes allowing OAI-AdsBot mandatory, recommends allowing OAI-SearchBot and asks brands that use a product feed to open their product image URLs to OAI-SearchBot as well. An image server returning a 403 can stop feed processing. The guidance also states that OAI-AdsBot is recognised by Cloudflare as a verified bot.Source: Advertiser Guidance for Allowing OpenAI Web Crawlers.

The technical team must confirm from logs that the relevant page returns a successful response, that the necessary content is visible and that protection rules do not block legitimate access. Rather than temporarily removing every security layer, access should be controlled and verifiable.

Why is the product catalogue moving to the centre of ad strategy?

In catalogue-based advertising, a single error in product information can be carried into a large number of ads. Wrong stock, a missing variant or an outdated price can damage the user experience and the commercial outcome regardless of creative quality.

A product feed makes it possible to use the products in a catalogue to generate ads. The accepted transfer method and how long a record is considered valid should be verified in the dashboard’s current feed documentation; these details may change throughout the beta. If price and stock change often, the business’s update frequency should be much shorter than the maximum the platform allows.

This table is an operational check; it does not replace the platform’s technical feed schema. Each channel’s mandatory fields must be applied separately.

VerifiedTwo operational constraints on product feed campaigns are documented: in new product feed campaigns, geographic targeting and exclusion can be set only at country level, and the Products tab in the dashboard is a reporting view, not the full feed inventory (from 30 September 2026 it also shows products’ review status). The third constraint, that bulk upload did not support creating product feed campaigns, was lifted on 30 September 2026.Sources: Create Campaigns; Measure Results; Troubleshooting.

To improve catalogue quality, a limited product group can be chosen at the start. Once problems are fixed in this group, which has sufficient stock, sustainable margin and clear product information, the scope can be widened.

Especially when generating descriptions with SI, missing fields must not be filled in by guesswork. If it cannot be verified that a product is waterproof, compatible with a device or meets a particular standard, those attributes must not be added to get a better match.

How do organic SI visibility and advertising work together?

GEO refers to optimisation for generative SI engines; AEO refers to optimisation for answer engines. Organic SI visibility, as addressed by these approaches, is concerned with how a brand is represented in answers, comparisons and the sources used. Advertising, by contrast, delivers reach through sponsored placement.

The two can draw on a shared information foundation: clear product definitions, consistent company information, genuine expertise, understandable content and current commercial terms. Even so, their budgets, objectives and success metrics must be kept separate.

Organic visibility is not a mandatory prerequisite for advertising. A new brand with a suitable product and working measurement can run an ad test. Equally, a brand that appears often organically does not automatically succeed in advertising.

In the same way, organic visibility is not a permanent guarantee. As sources, models and competitors change, the way a brand is represented can change too. Both areas require regular monitoring and improvement.

Uploading a product feed to Ads Manager does not guarantee being listed in organic ChatGPT answers either. The data infrastructure used may be shared; platform acceptance and results must be assessed separately.

What can be done, in concrete terms, for organic visibility?

The first step is to organise the information you want customers to learn correctly about your brand. The company’s name, expertise, products, service regions and commercial terms must not contradict each other across different pages. A feature offered on one page and described as discontinued on another creates ambiguity for users and for the systems that access that information.

The second step is to produce content that answers decision questions. You can explain not only who the product suits but also in which situations it does not. Use cases, comparison criteria, implementation steps and common problems should be addressed with genuine expertise. If original research or customer examples are used, the method and scope must be stated openly; generalisations the evidence does not support must be avoided.

On the technical side, it is valuable for important content to be accessible, for the page structure to be understandable and for appropriate structured data to be used consistently with the visible content. But adding a schema type or implementing a file standard does not guarantee inclusion in an SI answer. Technical implementation is no substitute for presenting accurate information.

A brand’s representation outside its own website should also be monitored. Independent reviews, industry publications, partner profiles and real customer experiences are all part of the information environment around a brand. The goal should be to have verifiable work represented clearly, not to manufacture fabricated praise or untrue reviews.

In measurement, a few screenshots of favourable answers are not enough. A question set can be built from target customer needs, and repeated checks can be run while recording the model, date, language and test conditions. The brand appearing, being described accurately, being matched with the relevant product and bringing in qualified traffic are separate outcomes. This test set does not represent every user conversation; it helps track change over time under the same conditions.

This is how organic visibility and advertising work together in practice: content work improves the information the customer needs to decide, advertising tests that offer in suitable contexts, and sales and support data guide the next round of content and product improvements.

Where do ChatGPT Ads and agentic commerce meet?

Agentic commerce is when SI agents carry out commercial tasks such as product research, comparing options, preparing a basket and, under suitable conditions, making a purchase, in line with the user’s request and the authority they have granted.

In this approach, not every step has to be autonomous. The agent can narrow down products, build a suitable basket and come back to the user before payment. Which part of the transaction can be carried out depends on the platform, the merchant, the payment infrastructure and the user’s authorisation.

Advertising can create a discovery or evaluation point on this journey. Afterwards the user can examine the product in more detail, ask the business’s representative a question or move into a suitable commerce flow. But opening an ad account, talking to an agent and completing payment through an agent are different capabilities.

Adding a chat box to a website does not build this whole chain. Likewise, the existence of an ad format such as Sponsored Agents does not mean every advertiser has an automatic purchasing capability.

Changed CourseThis distinction became concrete in 2026. OpenAI withdrew Instant Checkout, which completed transactions inside ChatGPT, in March 2026 and repositioned ChatGPT as a surface for discovery and comparison; the purchase is completed in the merchant’s own store or through the merchant’s ChatGPT app. Shopify merchants’ products are discoverable in ChatGPT, but payment does not finish inside the chat.Source: TechCrunch (24 March 2026).

What do ACP, UCP and MCP do?

The acronyms in SI commerce infrastructure are technical approaches that define different responsibilities. None of them, on its own, provides acceptance into a sales channel or guarantees a commercial result.

These frameworks can be used together. A commerce capability, for example, can be defined within a suitable protocol framework and exposed through MCP tools. Even so, implementing a protocol does not mean being listed automatically on any SI platform or opening sales to every user.

VerifiedThe owner and status of all three protocols are clear. OpenAI and Stripe published ACP on 29 September 2025 under the Apache 2.0 licence; the current version is dated 17 April 2026. Google announced UCP on 11 January 2026; UCP-based checkout is currently in early access with selected merchants in the US, Canada and Australia. Anthropic published MCP in November 2024 and transferred it to the Agentic AI Foundation on 9 December 2025. The warning about listing guarantees is written in OpenAI’s own merchant documentation: eligibility does not guarantee display.Sources: Agentic Commerce Protocol; Google, UCP announcement and Merchant Center; Model Context Protocol; OpenAI merchant documentation.

The technical choice should be made by first defining the target task. The integration needed for product discovery and the integration needed to complete payment carry different responsibilities. The capabilities supported by the store and payment infrastructure must be assessed together with the access conditions of the target channel.

How can you tell whether agents can actually complete shopping tasks?

Consider a sample user task: “Find me a waterproof shoe in size 40. The total, including delivery, must not exceed 3,000 TL. Get my approval before payment.” This is a scenario created for a readiness test, not a real product or price commitment.

In a successful flow, the product attribute must be verified, the correct size must be in stock, the total including delivery must be calculated and the user’s payment approval must be preserved. An agent producing a positive sentence at the end of the task is not, on its own, proof of success.

Access permissions must also be separated. Reading product information, changing the catalogue, preparing a basket and completing payment should not fall under the same permission scope. Spending limits, validity periods and changes that require re-approval should be defined.

After a fix, the same task must be run again. The catalogue version, test conditions, tools used and outcome should be recorded to confirm that the change has resolved the problem. A general success rate must not be inferred from a single successful attempt.

This approach is also valuable for being able to handle, operationally, the interest that advertising brings in. A campaign that brings in more users must not magnify commercial losses because of a wrong price or a broken transaction flow.

How is the cost of ChatGPT ads calculated?

There is no fixed click or impression price for ChatGPT Ads that applies to every advertiser. Ads are selected through an auction influenced by factors such as eligibility, relevance and bid.

The choice of objective must be kept separate from the payment method:

A conversion-focused campaign does not mean you pay nothing when there is no sale. The optimisation objective can be a sale while billing is based on clicks or impressions.

The suggested starting maximum bid of US$3–5 for CPC campaigns is not the actual average CPC either. It would be wrong to treat this figure as a fixed price for the Türkiye market or as a sensible cost for every product.

VerifiedThe documentation gives the recommended starting maximum bid for CPC campaigns as US$3 to US$5 per click. The billing models are also defined: oCPC is billed per valid click, oCPM per thousand impressions. The objective or billing model of an existing campaign cannot be changed later; a new campaign must be created or cloned.Sources: Ads in ChatGPT: The Basics; Conversion-optimized Campaigns.

The current minimum daily budget table lists US$25 for USD and €15 for EUR. Because there is no TRY row, no Türkiye-specific TL minimum should be assumed. The account’s billing currency and payment terms should be verified during setup.

VerifiedThe minimum daily budget varies with the account’s billing currency. The published table covers 23 currencies: 25 for the US dollar, 15 for the euro and sterling. There is no Turkish lira row in the table, so it would be wrong to assume a TL-specific floor.Source: Create Campaigns, Minimum Campaign Spend table (viewed 21 September 2026).

The first pilot budget should be set by the questions the business wants answered and the risk it can bear, not by the platform minimum. A budget that is technically enough to open a campaign may not be enough for a meaningful performance assessment.

Do not confuse a daily budget with total spend control

A daily budget is an average delivery target. Under fixed budget conditions, spend on a single day can reach up to twice the set amount; within the relevant Sunday to Saturday week, the cap is seven times the daily budget. Pro-rata calculation conditions can come into play when the budget changes.

A total campaign budget gives a tighter limit to teams that want to control the total amount allocated over the life of the campaign. The payment threshold, by contrast, relates to when charges are collected; it does not replace a budget or spending limit.

Before the first pilot, finance and marketing teams should agree who owns daily monitoring, the total amount permitted and who has authority to make changes. The delay in updating spend reports must also be taken into account. Zero spend temporarily showing in the dashboard may not mean no charges have been incurred.

VerifiedThe daily budget mechanics are documented: spend on a single day can reach up to twice the selected daily budget, while spend over a seven-day period does not exceed seven times that amount. The budget week runs Sunday to Saturday in the account’s time zone, and if the weekly allocation runs out early, delivery can stop even though the campaign appears active. Billing is postpay; the payment threshold is a collection trigger, not a spending limit.Sources: Create Campaigns; Daily Budgets; Billing and Payment.

Derive your success threshold from your own economics

The ad budget discussion should start with the contribution a customer leaves to the business. A high basket value alone does not mean a high capacity to pay for advertising; product cost, logistics, payment fees, returns and service costs must also be factored in.

The calculation below is entirely hypothetical; it is not a platform performance expectation.

Suppose a business’s net revenue per order is 3,000 TL. Assume in this example that every order from advertising is a new customer’s first order. If the margin after product cost is 35% and other variable costs not yet included in that margin are 250 TL:

• Order contribution before advertising: 3,000 × 35% − 250 = 800 TL.
• If the click-to-order conversion rate is 2%, the break-even average CPC is: 800 × 2% = 16 TL.
• If the actual average CPC is 24 TL, the acquisition cost is: 24 ÷ 2% = 1,200 TL.
• First-order contribution: 800 − 1,200 = −400 TL.

This calculation shows first-order economics. Repeat purchases, customer lifetime value or cross-sell contribution should be added only if they rest on realised, reliable data. Explaining a first-order loss with uncertain future revenue is not a basis for a sustainable scaling decision.

In B2B businesses, the same logic can be built on the probability of a sale. If, for example, 20 of 100 leads are qualified and 10% of qualified leads become customers, the overall lead-to-customer conversion rate is 2%. Media-driven customer acquisition cost can be calculated by dividing the cost per lead by 0.02. Sales team and onboarding costs must also be considered separately in total acquisition cost.

Which bid strategy should you start with?

A click-focused campaign can be used in the early stage to examine traffic quality and landing page behaviour. A conversion-focused campaign requires a correct event definition, working measurement and sufficient signal.

There is no single switching threshold that applies to every business. A rule such as “switch strategy automatically after five conversions” should not be applied without considering the sales cycle, event frequency, data quality and account conditions.

Maximize results is an automated bidding approach aimed at producing more results for the set objective within budget. It does not guarantee a particular CPA, CPC or ROAS outcome. If automated bidding produces more of a wrong or overly shallow conversion event, real customer quality may not improve.

VerifiedMaximize results exists as a real bid strategy and is selected by default in eligible new ad groups. The documentation says the strategy prioritises result volume; it makes no commitment to a particular acquisition cost or ROAS. Advertisers who want to bid manually can switch to a maximum bid at ad group level.Source: Maximize Results Bid Strategy.

That is why the optimisation event should be chosen reasonably close to commercial value. A balanced choice is needed between treating page views as success because they are easy to achieve and expecting fast learning from a sales event that happens very rarely.

If the objective, billing model or selected conversion event of an existing campaign cannot be changed, a new campaign should be created. When making comparisons, the effect of this switch on results should be recorded.

Measurement architecture: the platform, analytics and CRM must answer the same question

In ChatGPT Ads, conversion events can be sent with the Pixel and the Conversions API. When the same event is sent from both the browser and the server, using a shared event ID helps prevent the same outcome from being counted twice. Preserving the oppref value from the ad click through redirects also supports matching.

VerifiedThe measurement infrastructure is documented. Conversion events can be sent with the OpenAI Pixel, the Conversions API or both. If the same conversion arrives through both channels, the same event ID must be used for deduplication. The click reference, oppref, is appended to the landing page URL, stored in a first-party cookie and should be preserved through redirects.Source: Conversion Measurement.

It is not enough for measurement to send data technically. The event sent must represent the transaction that actually took place. A click on the apply button and a successful form submission are separate events; so are reaching the checkout page and a completed order.

Organic ChatGPT referrals and paid ad traffic must be tagged separately. The UTM standard should be kept consistent at campaign, message and product group level. Naming should be simple enough for the team to read and must not contain personal data.

An attributed conversion is not the same as an incremental sale

Attribution assigns an outcome to an ad interaction under a defined set of rules. Incremental impact, by contrast, tries to understand the additional outcome that would not have happened without the ad.

A user may first see a brand on social media, then compare products in ChatGPT and finally buy through a branded search. Different platforms may each credit themselves with a contribution to the same order under their own rules. That is why the sum of platform reports may not exactly equal real total sales.

In ChatGPT Ads reporting, 7, 14 or 30-day click windows and an off or one-day view window can be selected. Because conversion reporting can be delayed, a processing time of 24–48 hours should be allowed for. Modelled conversions and different reporting times can also create discrepancies between analytics systems.

VerifiedAttribution windows are selected in the dashboard: 7, 14 or 30 days for clicks; off or 1 day for views. It takes 24 to 48 hours for attributed conversions to appear in reporting.Source: Measure Results.

Live, LimitedOpenAI states that its work on incrementality measurement is at an early stage and that it is exploring geo-based experiments with Haus, Measured and WorkMagic to understand the causal impact of advertising. This is not yet a ready-made tool in the dashboard, and there is no in-platform brand lift measurement.Source: OpenAI, Building advertising for the way people use AI (5 October 2026).

In the first assessment, examining post-click and post-view results separately makes the assumptions visible. If there is sufficient volume, additional impact can be investigated with suitable control groups. Without such an experiment, “sales attributed to the platform” is a more accurate statement than a claim of “additional sales created by the ad”.

Which metrics belong in the management report?

CTR and CPC help in understanding operations; they do not explain the whole management decision. The report should be organised around the business model.

If commercial tasks are carried out with agents, additional metrics are needed: correct variant rate, offer accuracy, task completion rate, stopping correctly when required and cost per successful task.

When measuring task success, the scenario scope and number of attempts should be disclosed. Correctly stopping a transaction because the user has not granted authorisation must not be counted as a failed purchase. These cases should be measured separately.

How should early success stories be read?

In new ad channels, striking customer examples spread fast. But a high ROAS or conversion rate announced for one campaign is not an industry average.

When assessing a result, you need to know the sample size, brand awareness, product price, campaign duration, promotional effect, attribution window and share of new customers. Strong sales during a discount period in particular may not deliver the same economic result outside it.

The first partner findings OpenAI shared on 5 October 2026 should be read in this light. According to DV Rockerbox, WeightWatchers’ attributed acquisition cost on ChatGPT Ads was 15.3% lower than its blended paid search average. WorkMagic measured a statistically significant lift for Dose and reported that 67% of incremental purchases came from new customers. According to Triple Whale, 93% of Portland Leather’s ChatGPT Ads visitors were new. All three are single-advertiser results based on different measurement methods; without disclosure of sample, duration and attribution window, they should not be turned into an industry expectation.

LLM-referred traffic and ChatGPT Ads traffic are not the same data set either. Presenting the conversion rate of high-intent users arriving organically as the expected performance of a paid campaign would be misleading.

In the Webtures approach, a success story produces a hypothesis to be tested. The decision to grow a budget is made on the brand’s own data, at sufficient volume and on repeatable results. If the data is insufficient, the outcome should be recorded as “insufficient evidence”.

How can customer lists and first-party data be used?

Custom Audiences allows a business to use suitable lists of customers or prospects for inclusion, exclusion and bid adjustment in a campaign. There is a threshold of at least 25,000 matched users for inclusion and bid adjustment. Smaller audiences can be used for exclusion.

Uploading 25,000 records does not mean 25,000 matched users. Duplicates, invalid information and unmatched records can shrink the usable audience. Regional personalisation conditions must also be considered.

VerifiedThe threshold and usage rule are written in the documentation: inclusion targeting and bid multipliers require at least 25,000 matched users, and audiences below that number can be used only for exclusion. The documentation also states that the number of uploaded rows is not equal to the number of matched users and that invalid and duplicate records are not counted. For campaigns targeting the European Economic Area and Switzerland, it recommends avoiding custom audiences unless personalisation is available.Sources: Set up Custom Audiences; Create Campaigns.

This feature should not be planned on the assumption that you can target a small B2B customer database directly. First-party data has value beyond targeting too: validating lead quality, separating existing from new customers, tracking order outcomes and carrying out economic assessment.

When designing data use, the question “what information may be sent?” should be asked alongside “what information is actually needed?”. Moving unnecessary personal data into the ad infrastructure does not make for a better measurement strategy.

Privacy, KVKK (Türkiye’s Personal Data Protection Authority) and user control

ChatGPT advertisers do not have access to users’ private chat history or memory. That said, messages a user sends directly to a business and data a business sends from its own website for ad measurement are different flows.

On-site conversion measurement, cookies, customer lists and advanced matching should be assessed in terms of which data is processed, for what purpose and on what legal basis. Where required, explicit consent, the duty to inform and the conditions for transfers abroad must be addressed separately.

Hashing an email address or phone number does not automatically make the processing anonymous or free of all obligations. Sending data server-side is not a way to override user preferences and legal conditions either.

This assessment is work for legal and technical teams to do together, based on the actual data flow. Platform policy does not replace the local obligations the organisation is subject to.

Sector eligibility: the same ad route is not open to every product

The early scope of ChatGPT Ads focuses on areas such as consumer products, home and living, local services, travel, digital products and education. Sensitive and regulated categories face more restrictive conditions.

As of 21 September 2026, financial services ads are generally prohibited outside the US. In the US, approved advertisers can be permitted on a case-by-case basis in certain areas, including insurance, vehicle financing and leasing. Healthcare services also face general restrictions outside the US; legal services are prohibited outside the US.

An advertiser being approved does not mean every one of its ads, or every conversational context, is suitable. The business, creative, landing page and placement are subject to separate reviews. Sensitive user contexts such as personal health and mental health are also protected.

For an organisation in Türkiye offering banking, insurance or financial leasing services, the first step is a product-by-product eligibility assessment. In borderline cases such as car rental, classification should be based on the actual commercial offer. General financial education content must not be used as a way to promote a restricted financial product indirectly.

VerifiedThe ad policy dated 10 September 2026 draws this distinction clearly. Financial services and healthcare services ads are generally prohibited outside the US; legal services are prohibited outright outside the US. In the US, approved advertisers in these three categories can be permitted on a case-by-case basis following manual review. Auto loans and leasing, and insurance, are among the listed items in the financial services category. The policy also prohibits advertisers from misrepresenting their business location, service area or eligibility to advertise in a market.Source: OpenAI Ad Policies (10 September 2026).

Clarify responsibilities to reduce delays in large organisations

Being unable to reach the ad dashboard can stem from different problems: the corporate network, login, user role, account verification, payment or product eligibility. Producing a fix before the problem is correctly classified wastes time. We cover a five-stage transition model for large companies in enterprise transition to ChatGPT Ads.

ChatGPT workspace or API organisation permissions do not grant automatic access to the ad account. Account roles must be defined separately on the Ads side. Account ownership should stay within the organisation; the operations team should be given access that matches its remit.

VerifiedAccording to the documentation, being a member of a workspace does not mean holding the Ads Admin role required to complete account setup; the role is granted separately from the users tab in Ads Manager. The business opens the account itself, and the team running operations is added later by invitation.Sources: Troubleshooting Common Issues; Ads Manager Beta Account Setup.

Organisational speed can be increased by building a reusable way of working rather than redoing every check from scratch for every campaign. Once approved claims, a data flow template, an access procedure and budget limits are settled, subsequent tests can be prepared faster.

Which businesses should start where?

A channel holding potential for a sector does not mean every business in that sector will get the same results. Starting priority should be set by how explainable the product is, whether the user’s need surfaces in conversation, data quality, margin and measurement capacity. The pairings below are not performance forecasts but starting suggestions for designing tests.

In every example, the ad category, target country and available campaign features must be assessed separately. A well-prepared use case does not remove a category restriction on the platform.

Three hypothetical scenarios show how this approach can be applied:

For a furniture brand: rather than starting with a broad definition such as “work desk”, it can address the need of a customer who wants to use two monitors in a small room. Desk dimensions, load capacity, cable management, assembly and delivery conditions must be consistent between the ad and the product page. Success is judged by the economics of orders placed after the click, net of cancellations and returns.

For a B2B software company: instead of a “best CRM” claim, it can pick a specific business profile that wants to unify its existing sales and support systems. Integrations must genuinely be supported, the scope of use must be clear and demo requests must be tracked in the CRM. A large number of form fills does not make a campaign successful if the sales team finds none of them suitable.

For an education brand: the expectations of a user seeking a career change can be separated from those of a user who wants to build skills in their current job. Prerequisites, course duration, assessment method and the skills that can be gained by the end of the programme are explained. Instead of unverifiable promises such as a job or income guarantee, the candidate’s fit with the programme is put first.

What these examples have in common is that they do not design the ad as a creative message alone. The accuracy of the offer, the clarity of the page and the post-sale outcome are all part of the same test.

Is your brand ready for ChatGPT Ads? A ten-question check

Assessing readiness with only the question “has the ad account been opened?” is not enough. The team should be able to answer the following questions with concrete evidence:

1. Is it clear which user need we will test, with which product or service?
2. Have access and eligibility been checked in terms of legal entity, target country and ad category?
3. Can we support the claims we will use in ads with product information, service scope or verifiable evidence?
4. Does the landing page give users the price, terms and scope information they need to decide?
5. Is it clear who is responsible for keeping product, variant, stock and delivery data up to date?
6. Have the necessary crawler access and site security been tested together?
7. Can we record conversions on the correct event and filter out duplicate records?
8. Can we compare platform data with CRM, order and finance data?
9. Have data processing, user notices, consents and access permissions been assessed?
10. Are our spending limit, economic success threshold and stop conditions written down?

This list is not a certificate or an official platform acceptance test. Its purpose is to make the readiness gap visible. If there are gaps in core areas such as category eligibility, budget authority, accurate measurement and data processing, closing them is the priority. A high number of creative variations does not make up for these gaps.

Businesses with an agentic commerce goal also need to test scenarios covering payment authorisation, price changes, stock loss, retries, cancellations and order verification. Being ready to advertise and being ready for an agent to transact are not the same level of readiness.

The first 90 days: from discovery to controlled scaling

The aim for the first 90 days should not be to use every feature but to learn under which conditions the channel works. The plan below is a working framework. For businesses with long sales cycles, seeing the final sales outcome may take longer; if there are access or eligibility restrictions, the timeline is adjusted accordingly.

The first 30 days: data, offer and measurement readiness

In the first stage, existing customer questions, sales calls, on-site searches and support tickets are reviewed. The information customers need when deciding, and the conditions under which the offer makes sense, are identified. This work grounds campaign assumptions in real customer needs.

Account and category eligibility are verified. A limited number of use cases are selected; product data and landing pages are reviewed against those use cases. Ad claims are approved, the necessary technical access is put in place and the measurement flow is tested. The test budget, contribution margin calculation and acceptable acquisition cost are agreed with the finance team.

By the end of this stage, the team should have five concrete outputs: an approved offer, campaign hypotheses, a verified data set, a tested conversion flow and a written budget limit. Organic SI visibility work can run in the same period as needed; it should not be placed in front of the paid test as a mandatory precondition.

Days 31–60: a pilot of limited scope

The pilot starts by testing a small number of needs that are meaningfully distinct from one another. Splitting the budget across many campaigns and creatives can make it hard to see which variable is driving the result. Offer, context, creative and landing page are considered together; changes are logged.

The first checks focus on data accuracy, spend, ad eligibility and user experience. The time it takes for conversion data to form and be reported is taken into account. Lead quality in the CRM, or the actual status in the order system, is compared with the platform figures. No lasting performance verdict is declared on the first few conversions.

The output of this stage should not be a performance table alone. It should also report which need drew interest, which offer was not understood, at which stage users dropped off and which leads the sales team found suitable.

Days 61–90: economic assessment and decision

In the third stage, results are assessed together with the sales cycle, return period and data lag. Where there is sufficient data, customer acquisition cost and contribution margin are calculated. Under suitable conditions, incremental sales impact is investigated using methods such as a control group or a regional test.

The budget for campaigns that look positive is increased gradually. Alongside the average result, the additional result delivered by the added budget is also tracked; it is not assumed that a campaign that works well at low volume will keep the same efficiency at higher spend. Campaigns showing data or offer problems do not move to scaling until those problems are fixed.

By the end of day 90, sensible decisions may include scaling, retesting a specific scenario, fixing the data infrastructure or stopping ad investment. A pilot’s success is not necessarily measured by whether it increases the budget. Enabling the business to make its next investment decision with better information is also a valuable outcome.

When should a pilot be stopped, and when should it be improved and scaled?

Decision conditions should be written down before the campaign starts. Changing the definition of success after the budget has been spent makes learning harder.

When a problem such as a wrong price, a misleading claim, faulty data transfer or uncontrolled spend appears, there is no need to wait for performance data to mature. The operational problem must be fixed. Economic performance decisions, on the other hand, should rest on the predefined budget, the sales cycle and sufficient data.

Frequently asked questions about ChatGPT Ads

Will advertising on ChatGPT make my brand more prominent in organic answers?

Buying ads does not give you the right to be recommended in organic answers. Ad delivery and answer generation are handled separately. Paid visibility is no substitute for organic SI visibility work.

Do I have to finish GEO work first?

No. GEO is not a mandatory prerequisite for advertising. Depending on the business’s needs, the two can run in parallel. Preparation such as accurate product information and clear pages can support both.

Can I show ads to every ChatGPT user in Türkiye?

No. Advertiser access from Türkiye does not mean every user in Türkiye will see ads on every plan and surface. User eligibility, targeting, available inventory and campaign conditions are assessed separately.

Do ads appear only when the keywords I set are typed?

The guidance that describes campaign context should not be thought of in the same way as exact-match keywords in traditional search ads. The user’s need and the context of the conversation are what matter. That is why the user scenario needs to be defined clearly.

Can advertisers see users’ private conversations?

Advertisers are not given users’ private ChatGPT conversations. Information a user shares by contacting an advertiser directly is a separate interaction. Data a business collects on its own website is also subject to that business’s own notice and data processing responsibilities.

Is there a fixed cost per click for ChatGPT Ads?

No. Campaign conditions, competition, objective and bid strategy can all affect cost. The suggested bid on the platform is not a guarantee of the cost per click you will actually pay. Planning should start from the acquisition cost the business can sustain.

Can a meaningful test be run on a small budget?

That depends on the expected traffic and conversion volume, the sales cycle and the question the test is asking. Meeting the minimum daily budget does not mean you will generate enough data to make a reliable decision about business outcomes. A narrow, clear hypothesis can increase the learning value of a limited budget.

Is uploading the product catalogue enough to drive sales?

No. Catalogue accuracy, campaign eligibility, the offer, stock, the landing page, delivery terms and measurement must work together. Product data uploaded for advertising does not guarantee organic product recommendations either.

Is Sponsored Agents currently open to every advertiser?

As of the date of this update, it runs with selected advertisers and in a limited scope. It should not be planned as a generally available feature. In the meantime, brands can prepare product information and verified answers to customer questions.

Is adding a chatbot to my website enough for agentic commerce?

No. Generating chat answers is different from the capabilities of checking live stock, building a basket, authorised payment and creating verified orders. The transaction infrastructure and failure scenarios must be designed and tested separately.

Does ACP, UCP or MCP integration guarantee being listed in ChatGPT?

No. Technical protocol support does not guarantee platform acceptance, commercial eligibility, visibility or sales. Which function the integration serves and which connection methods the platform supports must be assessed together.

Can I measure success by CTR or platform ROAS?

These metrics are useful for diagnosis; they do not show the full business outcome. They must be assessed together with customer quality, realised sales, returns, contribution margin and incremental sales impact. In long sales cycles especially, CRM data is decisive.

Verification method and sources

The platform claims in this guide were verified against primary English-language sources on 21 September 2026. In the 30 September 2026 update, the product changes OpenAI announced to advertisers were compared with the help centre and developer documentation; the two features mentioned only in the announcement, bulk creation of product feed campaigns and review status in the Products tab, were flagged separately along with their source. In the 6 October 2026 update, information from OpenAI’s 5 October announcement on the visual format, measurement partnerships, first partner findings and brand safety was added; the partner findings were reported per advertiser and together with the source partner. The order of precedence was: first OpenAI’s own help centre and policy pages, then the protocols’ own documentation, and finally companies’ official announcements. No figure that appears only in a secondary source was presented as established fact.

Two points could not be verified, so no definitive statement is made about them in the text. First, which transfer methods a product feed can be submitted through and how long a product record is considered valid: these details do not appear in the help centre’s general advertising articles. Second, the amount of the temporary pre-authorisation that may be applied when a card is added: the documentation states only that a temporary authorisation hold may be visible and gives no figure.

Primary sources

• OpenAI, Ads in ChatGPT: plans that show ads, age limit, temporary chats and answer independence.
• OpenAI, Ads Manager Availability: countries where self-serve access is open, and Türkiye’s status.
• OpenAI, Ads Manager Account Setup: the separation of account, brand and legal names and the effects on review.
• OpenAI Ads API, Insights: reporting on attributed conversions outside the campaign objective.
• OpenAI, Ads in ChatGPT: The Basics: context hints and the suggested starting bid.
• OpenAI, Create Campaigns: campaign structure, minimum daily budget table, country targeting and the custom audience threshold.
• OpenAI, Create Ad Groups: writing context hints.
• OpenAI, Ads Manager Beta Account Setup: account setup, fields that cannot be changed and team invitations.
• OpenAI, Measure Results: attribution windows, reporting delay and product reporting.
• OpenAI, Billing and Payment: the postpay model, payment threshold and spending caps.
• OpenAI, Troubleshooting Common Issues: telling role and access problems apart.
• OpenAI, crawler guidance for advertisers: OAI-AdsBot and OAI-SearchBot access.
• OpenAI, Set up Custom Audiences: the 25,000 matched-user threshold and the exclusion rule.
• OpenAI, Conversion Measurement: Pixel, Conversions API, event ID and oppref.
• OpenAI, Conversion-optimized Campaigns: oCPC and oCPM billing.
• OpenAI, Daily Budgets: daily and seven-day spending caps.
• OpenAI, Maximize Results Bid Strategy: the automated bid strategy.
• OpenAI, Sponsored Agents in ChatGPT Ads: limited alpha status.
• OpenAI, Building advertising for the way people use AI (5 October 2026): the visual ad format test, measurement and incrementality partners, first partner findings, Negative Phrases and brand suitability pilots.
• OpenAI, Ad Policies (10 September 2026): category restrictions and the country distinction.
• OpenAI, Our approach to advertising and expanding access (16 January 2026): the announced principles.
• Agentic Commerce Protocol: ACP versions and licence.
• Google, UCP announcement (11 January 2026) and Merchant Center: UCP scope and early access.
• Model Context Protocol: the transfer of MCP to the Agentic AI Foundation.
• OpenAI merchant documentation: the statement that eligibility does not guarantee display.
• Google Ads Help, ads in AI Overviews: country coverage.
• Meta Newsroom (1 October 2025): use of SI interactions as an ad signal.
• Amazon Ads, sponsored prompts: US coverage.
• HubSpot investor relations (16 September 2026): the CRM partnership.
• TechCrunch (24 March 2026): the withdrawal of Instant Checkout.

The Webtures view: building SI advertising into growth strategy

Preparing for SI advertising is a broader management question than learning to use the ad dashboard. How clearly the brand explains what it offers, how current it keeps that information, how much easier it makes the customer’s decision and how profitably it can meet the demand created must all be assessed together. To run the channel with us from setup to measurement, see our ChatGPT Ads management service.

For Webtures, the starting point is matching the business’s growth objective with customer need. Organic SI visibility, paid media and agentic commerce are different parts of that match. Separate objectives and metrics should be set for each, while a shared way of working is built across product information, technical access, content quality and measurement infrastructure.

This approach does not produce a project the marketing team is expected to solve alone. Product, sales, technology, finance and data owners meet within the same decision framework. What becomes visible is not only the demand advertising brings in, but also whether the business can meet that demand with the right offer and the right experience.

The value of ChatGPT Ads for your brand will not come from simply being present in a new channel. It will come from presenting an offer that fits the right need, keeping the promise you make and validating the result against your business economics.

To assess your brand’s readiness for SI advertising across strategy, organic visibility, product data, ad operations and measurement, you can build a roadmap with Webtures that fits your business goals.