ChatGPT Ads Arrive After the Customer Has Explained What They Want

ChatGPT ads enter a buying journey after the user has already explained what they need. That gives conversational advertising unusually rich context, but it also raises the standard for accuracy, trust and brand identity. Better targeting will not compensate for vague product information or forgettable creative.

A search ad often meets a customer at the query. A social ad interrupts a feed. A conversational ad can arrive after several sentences of intent, comparison, hesitation and context, which means the system may know far more about the job the customer is trying to get done before the advertiser ever sees a click.

That difference is starting to show up in the market. Business software accounted for almost 20% of ChatGPT ad impressions in Graphite's analysis of Similarweb data from March through August, while an Admiral Media case study found visitors reaching its own agency website from ChatGPT were 4.6 times more likely to submit a contact form than its average visitor.¹² Neither number proves conversational ads will outperform search or social across categories. They do suggest that the conversation before the click may become part of the qualification process.

The click comes later

The early ChatGPT ad mix is unusually heavy on products that need explanation. Monday.com, HubSpot, Canva and Jotform were among the brands appearing frequently, and Jotform reportedly ran more than 930 ad variations in a month.¹ Business and professional services added another 8.5% of impressions in Graphite's analysis, while developer tools accounted for 4.6%.¹ Those are categories where buyers commonly arrive with constraints, comparisons and questions rather than a simple desire to see a product. The product often needs a conversation before it needs a promotion.

That makes conversational advertising structurally different from a conventional placement. The user can describe the team size, the workflow problem, the budget, the feature they dislike in an existing tool and the outcome they want before a commercial message appears. The interface has heard the brief before the brand has earned the visit. That is a powerful position for any system deciding which product deserves to be mentioned or advertised.

The Admiral Media number is interesting for the same reason, although it needs to be handled carefully. The agency said that visitors arriving from ChatGPT were 4.6 times more likely to submit a contact form than its average visitor between March and early September.² That is one company's first-party data, not a market benchmark, but it is consistent with a plausible behaviour: somebody who has already explained their problem to an assistant may arrive at a website with more intent than somebody who clicked after a broad search. The useful signal here is the sequence of the journey, not the headline multiple.

This changes what counts as useful content. A generic page designed to catch a keyword may contain far less decision-making material than a clear comparison, a detailed case study, a precise product page or an honest explanation of who a product is for. Answer engines can only work with what brands make available to them, so specificity becomes a distribution asset, not merely a copywriting preference.

Ads move into the conversation

OpenAI has been building the commercial infrastructure around that idea. Its Shopify integration connects merchant catalogue and commerce-event data with ChatGPT Ads Manager, reducing the work needed to keep product information current and campaigns connected to what the merchant actually sells.³ ChatGPT ads have also expanded to more than 60 countries, with self-serve access opening the channel to more businesses rather than only large advertisers.⁴ The important development is less the number of markets than the point in the journey where the advertising sits. That is where the economics of the channel may become genuinely different.

Tanishq offers a useful test case because jewellery is expensive, visual and high consideration. The Indian brand began running ChatGPT ads around its Gold Exchange Programme and Rivaah wedding jewellery, putting paid media into conversations where people may already be asking detailed questions about exchange value, wedding purchases or product choice.⁵ That is a very different moment from showing a banner to somebody who happens to be reading about jewellery. The commercial message enters after the shopper has begun defining the purchase.

Trust becomes more important as context gets richer. VML research cited in the week's material found that 58% of shoppers cross-check AI-generated product recommendations before buying, 49% said AI-generated product imagery reduces their trust in a brand and 48% skip content they believe was created using AI.⁶ The numbers do not say people reject AI-assisted commerce. They say a fluent recommendation does not remove the customer's instinct to verify, especially when the purchase matters.

That should make marketers wary of treating conversational advertising as a shortcut around credibility. A recommendation can be perfectly timed and still fail if the product information is thin, the images feel synthetic or the brand cannot support the claim once the customer leaves the assistant. That is why relevance gets you considered while credibility gets you believed. The two jobs are connected, but they are not interchangeable.

The website is no longer first

Princess Cruises pushes the shift one step further. Its native ChatGPT app lets travellers discover, compare and plan cruises inside the assistant while staying connected to Princess's booking system.⁷ The customer does not need to begin on the brand's homepage, navigate a menu and translate their preferences into filters. The conversation itself becomes part of the product-discovery experience.

That matters for answer engine optimisation because the objective is no longer only to appear in a generated answer. Brands increasingly need to make their product data, availability, expertise and customer-facing information usable by systems that may help the customer act without visiting the site first. Appearing in an answer is only useful if the answer is accurate, current and recognisably connected to the brand.

Google's addition of multimodal search reporting makes the same point from another direction. Search Console can now show when pages appear after image-led searches through products such as Lens, Circle to Search and image uploads.⁸ A customer may express intent with a photograph, a screenshot, a paragraph or a conversation. The neat keyword that once organised the whole journey is becoming only one of several ways people tell a system what they want.

For marketers, this puts more pressure on the material beneath the campaign. Product feeds need to be correct. Images need to show the thing clearly. Service descriptions, pricing, availability, case studies and brand language need enough detail for a machine to use without inventing the missing parts. The same is true for Instagram content strategy: if a business wants an Instagram marketing AI system to produce recognisable posts, the system needs real brand material, real products and real preferences rather than a blank prompt.

More context, less freedom

Several launches this week point towards a useful design choice: narrow the model where the business needs reliability. Snappy AI for financial advisers analyses authorised campaign activity, audiences, performance and account configuration, but it is read-only and cannot publish a campaign or change settings.⁹ Propolis AI is grounded in more than 20 years of specialist B2B marketing data and practitioner material.¹⁰ Both products give the model a smaller, better-defined space in which to be useful. They are designed around the knowledge and permissions of a specific job rather than maximum freedom.

That is not a retreat from capable models. It is an admission that commercial usefulness depends on constraints. A general model can generate a plausible answer about almost anything, but a marketer often needs a system that knows which products exist, which claims are approved, which brand language is preferred and which actions require a human decision. A smaller freedom of action can produce a higher standard of usefulness.

Google's AI Max direction is similar. Its reporting connects the search term, creative and landing page used in an AI-driven ad journey, while AI Brief lets advertisers provide written context about the business, audience and messaging.¹¹ As the platform makes more decisions, it is also giving marketers more ways to supply context and inspect what happened. Automation does not remove the need for a brief; it makes the quality of the brief easier to see.

This principle matters well beyond paid media. An Instagram AI content management tool should not become more impressive by making it easier to publish hundreds of generic posts. It becomes more useful when it can work from a business's own media, website, product details and tone, then keep a human in the review loop. That is the logic behind AI-powered Instagram content for small businesses: use the system to reduce repetitive production work while keeping the owner responsible for what actually goes out.

Automation makes blandness cheaper

The week's creative-production stories point to the same tension from another angle. Standard Chartered's Formula 1 campaign used AI-enabled production to create visual perspectives that would have been difficult to film during the racing season.¹² The technology expanded what the team could make from an existing campaign idea rather than asking a model to invent the whole proposition. The creative concept existed before the production shortcut.

That distinction matters because production constraints are falling quickly. Image variants, video formats, ad copy and channel adaptations can be produced with less manual effort than before. When every team gains access to similar production shortcuts, the original idea carries more of the burden of differentiation.

Cheaper production also makes it easier to publish work that should have been rejected. A weak concept can now generate twenty polished assets instead of three, and automated distribution can put those assets in front of more people with less effort. Efficiency improves the consequences of good judgement, but it also magnifies the consequences of bad judgement.

This is the direction AI content marketing should take more often. Better tools can compress production time, adapt formats and remove repetitive effort, but that saved capacity should be reinvested in sharper source material and better judgement. Teams that use cheaper production only to increase volume may discover that automation makes forgettable work cheaper to distribute at scale.

The brand has to survive the handoff

Conversational advertising compresses several stages of the journey. Research, comparison, recommendation and paid placement can happen inside the same interface, and the click may arrive only after the customer has already explained what they care about. That creates an unusually efficient path to consideration, but it also gives vague brands fewer places to hide.

The most valuable response is not to flood answer engines with more material. It is to make the material they use more accurate, specific and recognisably yours. Product facts need to be current, customer evidence needs to be credible, creative needs a point of view and automation needs boundaries that stop a weak brief from becoming a thousand weak outputs.

ChatGPT can hear more of the customer's problem before the advert appears. The winning brand still has to provide an answer worth trusting after it does.

Sources