Building creative assets for conversational ad surfaces
How should headlines, descriptions, images, context hints and landing pages be built for ads in chat interfaces like ChatGPT? Learn with practical examples.
Creative assets for conversational ads means building, as one system, the headline, description, image and landing page of the small sponsored card shown directly below an SI answer, together with the context hints that decide which conversations the card appears in. With ChatGPT Ads opening as a self-serve platform in Türkiye, marketing and brand teams are looking for ways to bring copy they have produced for search and social ads over the years into this new space. Yet the chat card resembles neither a search ad nor a social media visual. The card is small, the text space is narrow and the user is in the middle of discussing a topic with an SI. On top of that, the decision to show an ad considers not only the copy, but also the landing page and context hints together. Creative work here therefore becomes a shared task for media planning, content and product teams rather than a single copywriter. In this guide, we cover the parts of the chat card, how each one should be written and how variant tests can be structured, with examples.
Which parts of the chat card do users actually see?
The standard ad card in ChatGPT consists of the advertiser name, a small logo, a headline, a description, an image and a single destination URL. The card appears below the answer with a “Sponsored” label and offers the user an additional option without breaking the flow of the conversation. The sitelinks, callout extensions and extra headline fields familiar from search ads do not exist in this format; the card is deliberately kept simple.
The advertiser name and logo come from the verified Ads Manager account and look the same on every ad. The logo should therefore be a simple mark that stays readable at a very small size; a long wordmark loses meaning in this space. Character limits are enforced in the Ads Manager form fields for the headline and description, and different figures for these limits circulate in industry guides. The safest approach is to rely on the current character counters in Ads Manager when setting up a campaign and to preview the copy on a mobile screen to check whether it gets cut off. Writing to the very end of the limit is not always an advantage; a short headline that reads in full on mobile is stronger than a long one that gets truncated.
Where do context hints begin in the creative process?
Context hints are the invisible but decisive first part of the creative work; they determine from the start which conversation the card will appear in and therefore what the card needs to say. Advertisers define these hints at the ad group level. Hints are not exact-match keywords; they describe the situations, questions and decision moments in which a product or service is useful. When deciding whether an ad is relevant, OpenAI considers the user’s conversation, the context hints, the landing page and the ad copy together.
This approach requires a mindset shift for teams coming from search advertising. The table below shows the difference between the two approaches through a concrete example.
| Topic | Keyword Habit | Context Hint Approach |
|---|---|---|
| How it is defined | “coffee machine”, “espresso machine price” | A user who wants to make milk-foam coffee at home, is choosing their first machine and worries about maintenance |
| Ad group structure | A single group per product category | A separate group for each use case: first machine, office use, gift search |
| Starting point of the copy | Repeating the keyword in the headline | Answering the question from the conversation in one sentence |
| Landing page | Category page | A guide or filtered product list that matches the situation |
The right-hand column shows that the creative copy actually begins when the context hint is written. If the hint describes “a first-time buyer worried about maintenance”, the card’s headline should speak directly to that concern. This consistency between hint and card strengthens both the ad’s relevance assessment and the user’s willingness to click.
How do you convey one idea in a narrow character space?
Effective copy in a narrow character space conveys a single idea in a way that answers the question the user asked in the conversation. Trying to cover brand value, product features and a promotion in the card’s two lines means none of them sticks. At that moment the user is looking for an answer; the card should read as a natural continuation of that answer.
The following principles stand out in effective chat card copy:
- The headline should carry a promise or a question, not a slogan: A phrase that describes the situation, such as “Espresso Without the Limescale Hassle”, is clearer than a slogan like “The New Name for Coffee”.
- The description should name the situation and the benefit: It should combine the user’s situation and what the product delivers for it in one sentence.
- The tone should match the conversation: Instead of the compressed, keyword-heavy language of search ads, use the tone of a helpful suggestion.
- Avoid exaggerated superiority claims: Phrases such as “the best” or “the only” carry policy risk and look untrustworthy next to an SI answer.
- Prefer numbers and concrete information: Verifiable details such as “2-year warranty” or “set up in 15 minutes” are more persuasive than general adjectives.
There is an extra point of attention when writing in Turkish: the agglutinative structure of Turkish quickly lengthens words within character limits designed for English. A direct translation of an English campaign therefore often exceeds the limit or weakens the meaning. Turkish variants should be treated as rewrites, not translations.
Why does using a logo as the image become a trap?
Using a logo as the main image in the chat card means spending the card’s most valuable space repeating the brand identity that is already visible. Because the advertiser name and logo already sit in the card’s top row, the main image’s job is to show the product or the outcome it delivers. Industry guides also stress that using a logo as the main image makes the card look like a banner and fails to connect with the conversation.
Because the image is shown at a small size, busy compositions become unreadable. Images that focus on a single object, have a simple background and suit a square format work best. Placing text inside the image is also not recommended; text that cannot be read at a small size makes the card look cluttered and can cause problems in the review process. For service businesses, a scene that shows the result of the service, such as a cleaned office or an installed solar panel, can be more meaningful than a product shot.
Is the landing page the invisible part of the ad?
The landing page does not appear on the card, but because it is considered when deciding whether to show the ad, it is an inseparable part of the creative work. Since the card allows only one destination URL, choosing that URL is a creative decision, not a technical setting. An ad that goes to the homepage struggles to meet the expectation formed in the conversation and may also weaken the ad’s relevance assessment.
A good landing page fulfils the promise made in the card’s headline in its first paragraph. If the card says “an espresso machine without limescale problems”, the first section of the page should clearly explain how that problem is solved. This follows the same logic as content prepared to be cited in SI answers: an opening that answers the question directly convinces both the user and the system. As part of its GEO consulting, Webtures analyses how SI systems read pages and which passages they cite, and uses the same insight to shape the first screen of ad landing pages. A page strengthened for organic visibility thus becomes a more consistent destination for conversational ads as well.
How far should you trust the SI tools in Ads Manager?
The SI tools in Ads Manager are valuable for producing quick drafts, but every piece of copy and every image that goes live should be reviewed by a person. On 16 September 2026, OpenAI announced new SI-powered creative tools in Ads Manager and the ability to create ads with a few prompts in ChatGPT Work. Google is moving in a similar direction with Asset Studio, making it possible to generate ad creative with natural language prompts.
These tools significantly shorten production time, especially in campaigns that require many variants. However, automatically generated copy may not always reflect the brand’s voice, the legal boundaries of the sector or the product’s real features accurately. In regulated sectors such as health, finance and food in particular, an automatically generated claim can expose the brand to legal risk. The right approach is to use SI tools for first drafts and variant generation, and leave the final decision to a team member who knows the product and the brand voice. Using a GEO tool such as Brantial to read the conversational contexts where campaigns appear alongside organic visibility also gives the creative team concrete direction on which topics to focus on.
In what order should variant tests be structured?
Variant tests should start with the variable that has the greatest impact and change only one variable at a time. In conversational ads, the greatest impact usually comes from the fit between the context hint and the headline; differences in image and description are finer adjustments. The first tests should therefore compare ad groups that target different use cases for the same product.
In the first stage, a separate ad group and situation-specific headline variants are prepared for each use case. Once enough impressions have been collected, description variants are tested in the best-performing situations. In the third stage, image alternatives are tried. In the final stage, the same card is paired with different landing pages to measure the effect on conversion. When the insight from each stage feeds back into the context hints of the next campaign, the loop between creative work and media planning is complete, and campaigns gradually start appearing in more precisely matched conversations.
