What In-Conversation Ad Inventory Actually Is and Why a Site Chatbot Is Not a Supply Source

In-conversation ad inventory is paid placement on an open AI surface, defined by readable conversational signals and infrastructure that connects advertisers to placements. ChatGPT, Google AI Overviews, and Microsoft Copilot offer buying paths, while Perplexity discontinued its ad program in February 2026. A brand’s site chatbot is not a supply source because its closed customer dialogue has no external advertising surface and lacks the signal access and transactional infrastructure needed to sell placements.

Why "in-conversation inventory" needs a precise definition before any buyer or seller can act on it

A media buyer approving an in-conversation line item needs to know whether it purchases an actual placement. A publisher describing a chatbot as inventory needs to know whether an outside advertiser can buy access to it. Neither question is answered by the presence of a conversational interface. Yet the phrase is circulating without a shared definition, allowing budgets and supply claims to rest on the appearance of a product.

The vocabulary adds another source of confusion. BCG’s January 2026 analysis identifies in-answer ads, in-conversation ads alongside dialogue, and agentic ads inside task completion as distinct emerging forms. Those labels describe where advertising appears, but they do not establish whether a particular surface has sellable supply. An answer, a dialogue, and a completed task can all contain commercial content without offering an advertising opportunity to an external buyer.

The distinction therefore requires a structural definition. The relevant tests concern the surface, the signals it reads, and the infrastructure required to transact a placement. Together, they separate a conversational AI advertising platform from a tool that merely uses conversation. That separation gives buyers a basis for approving spend and publishers a basis for describing what they can offer.

What platforms let brands place ads inside AI conversations, and what makes them qualify?

A qualifying surface is open to third-party advertising against interactions between the platform and its users. “Open” describes advertiser access. The platform controls the interface and has the standing to monetize it. A brand-owned support dialogue exists within the brand’s own customer relationship.

Advertisers have been able to buy live inventory on Google AI Overviews through Google Ads since 2024. Eligible campaigns using broad match, AI Max, Performance Max, or Shopping with smart bidding are already serving, with no separate opt-in. Google’s help page identifies Text and Shopping ads from existing campaigns as eligible to appear within AI Overviews. This is an established buying path with no chatbot installation that an advertiser must build.

Microsoft Copilot also uses the infrastructure already in place for campaigns. Eligible campaigns are automatically opted in, and ads are assembled from campaign assets. Microsoft states that display is not guaranteed. Google Gemini AI Mode represents a different deployment status, with Conversational Discovery ads and Highlighted Answers shown at Google Marketing Live 2026 but not yet buyable as of the sources’ review dates.

ChatGPT offers paid placements through its advertising infrastructure. OpenAI’s May 5, 2026 announcement introduced a beta self-serve Ads Manager where advertisers could sign up and buy ChatGPT ads directly. Perplexity, however, discontinued its program in February 2026 after launching sponsored follow-ups in late 2024. Meta AI has announced no general inventory, and Anthropic Claude has none because its policy rules it out.

These examples distinguish deployment statuses. Where supply is available, the defining property is that an advertiser can purchase a placement within someone else’s AI interface. Owning a chatbot does not provide that same access.

The signal layer: how conversational context replaces the keyword and why that changes what counts as inventory

In-conversation inventory reads the live meaning of an interaction rather than relying on a keyword string, a cookie-assembled profile, or a page category. The conversation itself supplies the targeting context. This changes the unit an advertiser competes for from an isolated expression of interest to intent developing through an exchange.

ChatGPT makes that difference concrete through the context hint. Advertisers write natural-language descriptions at the ad group level, which are matched against the live conversation in a relevance-weighted, second-price auction. A more relevant ad at a lower bid can outrank a generic ad at a higher bid. OpenAI documents no official character cap for these hints. The widely cited 280-character figure is a practitioner convention.

Conversation depth adds a further bid dimension. Early turns can express awareness-level intent, while deeper turns can signal comparison or a decision. This measures how far the user has moved through the exchange, which keyword match types do not track. In this contextual model, the signal exists while the conversation is active, no durable user profile is created or synced, and the signal disappears when the conversation ends.

A site chatbot also gathers useful conversation data. Its data, however, is proprietary information about the brand’s customers and prospects, enclosed within the brand’s owned environment. It is not readable by a third-party auction or available for matching against other advertisers’ context hints. Signal readability is therefore an architectural property.

The infrastructure layer: what an LLM SSP actually does and why a site chatbot has none of it

Real inventory requires infrastructure that turns an appropriate moment in a conversation into an ad placement, then records what was delivered. This work includes evaluating context, running the auction, rendering the creative, and handling attribution.

An LLM supply-side platform, or LLM SSP, begins by identifying an appropriate advertising moment from commercial intent or conversational context. It evaluates that moment against advertisers’ context hints and runs a relevance-weighted auction to select and price the placement. The placement is then rendered in a format native to the interface, such as a labeled sponsored card adjacent to the answer. Finally, the system reports the impression and, where possible, connects subsequent conversion events to it.

ChatGPT’s supply path illustrates the required infrastructure concretely. Its pilot launched with a $200,000 commitment and a premium CPM. Criteo became the first named technology partner in March 2026, connecting tens of thousands of advertiser clients. Programmatic access through StackAdapt opened May 5, 2026. These are demand connections that make advertising transactions possible, rather than features of the underlying conversation alone.

The operational stack has distinct responsibilities. Demand and auction, context and identity, creative generation, and measurement and attribution each provide part of the channel. A brand’s site chatbot has no third-party auction or external demand path like this, no publisher-side buying-platform integration, and no standardized ad format.

The resulting distinction follows from the transaction itself. An AI interaction on a publisher’s property can become a monetizable event when supply-side infrastructure evaluates it and connects it to advertisers. The widget alone does not perform that conversion.

Why the free-tier constraint and the closed-surface problem matter together for buyers evaluating what they are actually reaching

A qualifying surface does not make every user reachable. ChatGPT ads reach only Free and Go tier users, while Plus, Pro, Business, Enterprise, and Education users never see them. The ad-supported audience is therefore a defined subset of the platform’s users. That is a planning constraint.

For B2B advertisers, the distinction has a specific consequence. The business buyers most likely to use ChatGPT’s paid tiers are the very people its ads cannot reach. A plan based on the platform’s general audience would therefore describe something different from the audience available through ads.

A closed site chatbot creates a different limitation. It offers no external audience to a third-party buyer at all, because the brand’s dialogue is closed to outside placements.

For buyers, the practical step is to establish both eligibility and reach before committing budget. Buyers first need to establish whether an actual placement can be purchased through an auction and demand path. They then need to identify which users can encounter it. This order keeps a constrained audience on a real platform distinct from a proposed buy on a surface that cannot accept advertising.

What publishers actually need to do to create supply, and what brands owning a chatbot have instead

AI-generated answers have sped up a zero-click pattern in which people get information without visiting its source, which cuts referral traffic. Advertising within conversational interfaces offers a monetization path that does not depend on outbound clicks. BCG’s analysis frames conversational advertising as an emerging line item separate from search and programmatic.

That opportunity requires more than deploying a widget. A publisher’s conversational surface becomes a supply source when it connects to infrastructure that identifies monetizable moments, runs an auction, and delivers advertiser demand. High interaction volume may make the surface useful, but the integration establishes its advertising function. The investment concerns a transaction system as well as an interface.

A brand-owned support or sales chatbot serves another purpose. It is owned media, handling proprietary interactions with existing customers or prospects. Those conversations are closed by design and by the competitive logic of keeping outside advertisers out of the brand’s customer relationship. The absence of advertising supply does not diminish the chatbot’s value in that role.

A site chatbot is a CRM and conversion tool that can improve service, reduce friction, and move users toward purchase. Conversational advertising inventory is what the brand buys on an external surface such as ChatGPT, AI Overviews, or Copilot to reach intent already developing there. Publishers can participate on the supply side, but the revenue opportunity becomes viable through infrastructure that connects their interactions to that external demand.

What the three-test framework means for any team deciding where to place, what to build, and what to call inventory

The open-surface, readable-signal, and transactional-infrastructure tests are all required for a surface to produce real in-conversation inventory, regardless of its traffic or conversational appearance. A chatbot’s value to a business and its status as sellable supply are separate questions.

Buyers can use the framework to decide where to spend before they commit a budget. The surface must accept third-party advertising, the targeting primitive must read conversational context rather than merely a keyword or cookie, and an actual auction and demand path must connect the purchase to the placement. Which users can see ads then sets the reach within that structure. This gives the line item a concrete meaning.

For publishers, the same framework clarifies what to build. Handling conversations alone creates an asset for content or customer interaction, while supply-side integration makes it possible to evaluate, auction, and deliver placements. The distinction helps publishers invest in what they still need to build. It also gives publishers precise language for describing an offering to buyers.

The landscape is developing at different speeds. Gemini formats have been announced without becoming buyable at the review dates, while Copilot already serves eligible campaigns automatically. As more surfaces emerge, teams can apply the same tests without treating every new assistant as a new media channel.