Ad Format Requirements That Separate True In-Conversation Placements From Repackaged Search Inventory

A conversational AI advertising platform must match ads to what the live conversation means, place them within that conversation, support creative built for a single exchange, and have native publisher infrastructure. This article explains why those four requirements separate genuine in-conversation placements from repackaged search inventory. Appearing on an AI surface alone does not establish a new advertising channel.

Why the label "AI advertising" now covers two structurally different things

Most spending labeled “AI advertising” today buys traditional paid search inventory beside AI-generated results. The label obscures a basic architectural divide. A broad match, Standard Shopping, or Performance Max campaign can surface in both Google search results and Google AI Mode without a separate advertiser choice. Buyers may therefore compare placements as though they share targeting mechanics, creative specifications, and publisher infrastructure when they do not. The useful distinction is not whether AI appears on the page, but whether the advertising system was built for the exchange taking place there.

What makes a placement genuinely in-conversation: four requirements

A genuine in-conversation placement satisfies four requirements simultaneously. Contextual resolution matches the ad to the meaning of the live exchange. Placement position puts it inside or directly adjacent to the conversational turn. Creative structure defines a single contextual unit, rather than an asset competing across a multi-ad results page. Publisher infrastructure supplies a native auction and payment rail designed for conversational inventory, rather than a search auction reassigned to a different surface.

These requirements form a diagnostic. Each deals with something that changes when a person goes from scanning results to talking through a problem. The targeting system receives a fuller account of intent, while the placement enters an exchange with its own context and constraints. The format and the infrastructure must support that relationship together. Satisfying one or two requirements does not compensate for missing the others, particularly when the missing capability requires new infrastructure.

Contextual resolution: matching to conversation meaning, not to a keyword string

Contextual resolution reads the arc of the live conversation, including what the person asked, what has already been established, and what they are trying to accomplish. That is a different input from query vocabulary. A three-word search query requires an ad system to reconstruct intent from limited signals, whereas a conversation may already contain a budget, a constraint, a shortlist, and rejected alternatives. Keyword expansion can broaden the vocabulary used to find a match. It cannot substitute for resolving the relationship between those details in an ongoing exchange.

ChatGPT illustrates the difference through its targeting control. Its primary control is a context hint, a natural-language description of the conversations where an ad belongs, capped at 280 characters. There is no keyword layer in this model. The platform matches that hint against the live conversation, and a relevance-weighted second-price auction applies relevance weighting before price. A tightly relevant ad at a lower bid can therefore outrank a generic ad at a higher bid, making conversational fit part of allocation.

For a project management software advertiser, a context hint can describe a team that has outgrown spreadsheets, is comparing tools, and has mentioned a headcount constraint. Bidding on “project management software” describes the product category but does not express that combination of circumstances. Google AI Mode, by contrast, extends the keyword and audience signals of existing broad match and Performance Max campaigns to an AI surface. For buyers, the distinction is whether the system resolves the conversation itself or extends signals from an existing campaign.

Placement position: why beside-the-answer and inside-the-answer are not the same channel

Position determines how an ad enters the experience. An ad on a separate layer beside an AI-generated result still sits where a search ad would, even though the content next to it has changed. A sponsored card directly below a conversational turn instead meets the user at the end of an exchange about a problem or decision. Both placements can offer value, but they establish different expectations about the relationship between the ad and the surrounding content. Their classification should follow that relationship.

The Answer Independence principle places an important boundary around the conversational format. The paid placement buys visibility adjacent to the AI’s answer. Consequently, “inside a live AI chatbot conversation” should not be treated as a promise that payment inserts a brand into the assistant’s reasoning or recommendation. A placement can belong to the conversational experience while remaining separate from the answer itself.

Google AI Mode’s Conversational Discovery ads and Highlighted Answers appear inside AI-generated results and recommendation lists, but the user’s frame of reference remains a results-page structure. Microsoft Copilot occupies a middle position because its assistant interface is conversational while its ads are assembled from existing campaign assets. The ad has moved into an assistant without necessarily becoming purpose-built for that exchange. Neither case establishes a performance hierarchy, but both show why interface design and advertising architecture must be evaluated separately.

Creative structure: what a single contextual unit requires that a search creative cannot provide

A conversational creative specification describes a single contextual unit that has one conversation, one exchange, and one sponsored card. Search creative serves a different reading pattern. On a results page, multiple ads appear together, and brevity helps users scan competing options. In a conversational placement, the same brevity must preserve coherence with a specific moment in the exchange. A short headline is therefore not evidence that the two formats are interchangeable.

The launch chat_card specification includes a character-limited title, body copy of up to 100 characters, an image, a favicon, and a destination URL. Additional formats have since been introduced, but this specification makes the underlying constraint clear. The unit must fit a narrow conversational opening. With one or two placements per session, each ad stands out more than it would among the multiple sponsored results on a typical search results page. A generic value proposition and broad call to action do not express the same relationship to context as a unit built for the constraint under discussion.

Google’s Conversational Discovery ads complicate any simple contrast between static search assets and conversational creative. They use Gemini to generate creative for a specific conversational query in real time. That is a substantive change in creative structure. However, generating the creative per query does not by itself replace the targeting and auction beneath it. The format has a generative layer between allocation and impression, while the other requirements still need to be checked separately.

Format specifications are architectural evidence. Creative written for a results page assumes a different setting from creative entering an exchange where the platform knows a specific constraint, creating a structural mismatch. A system may address it through a native card specification or a purpose-built generative layer, but adapting the words alone does not establish that the full placement is conversational.

Publisher infrastructure: why the payment rail underneath matters as much as the format above it

Publisher infrastructure determines what the system can sell, price, and measure. Repackaged search inventory gets to an AI surface through an auction, targeting primitive, payment rail, and reporting model that were built for results pages. A native conversational surface requires those functions to be designed for its own inventory. That distinction is less visible than the ad card. It is nevertheless central to whether the buyer receives conversational targeting or a familiar campaign delivered somewhere new.

OpenAI built a native Ads Manager, auction, and format. Its self-serve opening in May 2026 had no minimum spend. The buying model has also expanded as the infrastructure develops. Three objectives are available, including CPM-priced Reach, CPC-priced Clicks, and Conversions. The Conversions objective supports oCPC, billed per click, or oCPM, billed per impression, while bidding toward a tracked conversion event.

Measurement also needs to change to fit a conversation. The path from an impression to a conversion may not include the trackable click associated with search. OpenAI’s Measure Results help article specifies a “1 Day” or “0 Day (disabled)” view-through reporting window and says advertisers must manually add the “View-through conversions” column. The reporting column is not on by default. OpenAI’s Ads Blog also describes incrementality measurement partnerships as “still in its early stages,” distinguishing a developing measurement layer from a finished solution.

Microsoft Copilot shows a different infrastructure choice. Its ads are assembled automatically from assets in existing campaigns and served without advertiser opt-in, demonstrating infrastructure portability by extending Microsoft’s search and display stack into an assistant interface. Portability can be useful even when it falls short of the native-infrastructure requirement. Buyers need that distinction because the same visible assistant surface can sit above very different systems for matching and allocating ads.

The publisher-side payment rail is still forming more broadly, with revenue-share models emerging but thin. Calling a placement conversational does not prove that it has this foundation. Without a mature native rail, inventory may go through infrastructure that cannot read or price the conversational context being sold. For classification, the relevant issue is not merely whether a publisher receives payment, but whether the auction and payment system were designed for conversational inventory. Measurement must also account for the gap between exposure within an exchange and a later conversion.

How the four requirements interact: a placement that fails one tends to fail the others

The four requirements share an architectural foundation. A search auction often comes with keyword-oriented targeting and creative formats made for results pages because all of them were built for the same buying model. ChatGPT’s native targeting mechanism, auction, format, and measurement layer all serve the same conversational surface. Their relationship explains why a missing infrastructure capability often accompanies gaps in contextual resolution or creative structure.

Generative creative is an important exception to that pattern. Google’s Conversational Discovery ads generate creative per query, addressing the creative requirement in a way that static search assets cannot. Yet the targeting and auction remain rooted in search infrastructure, so that advance does not establish full contextual resolution. Mixed answers to the diagnostic point to infrastructure that has been extended or adapted, with parts of the stack built for other uses. Buyers who mistake that mix for native conversation targeting risk miscalibrating creative, bidding strategy, and performance expectations before the campaign begins.

What should buyers ask before classifying a placement as genuinely in-conversation

Before committing budget, buyers should apply the framework to the current placement. Does the platform read the live conversation to make the match or extend keyword and audience signals from an existing campaign? The answer identifies the actual targeting primitive. If matching depends on those inherited signals rather than conversation meaning, the contextual resolution requirement is not met.

Does the ad appear inside or directly adjacent to the conversational turn, or on a results layer encountered as a list? The user’s frame of reference is decisive here. Does the creative specification define one exchange, one placement, and one contextual match, or repurpose a headline and description built for search? Buyers should distinguish per-query generation from static asset reuse without assuming that either establishes how the platform targets ads.

Does the surface have a native auction and payment rail for conversational inventory, with measurement adapted to a path that may not include a tracked click? An extended search or display stack does not meet that requirement simply because its destination is an assistant. These questions classify a placement in its current form. Infrastructure is still being built, and a system extended from other uses today may have a purpose-built version in the future. The buying decision should rest on what the current placement actually resolves, displays, generates, and measures.