Migrating Your AI Ad Budget to a DSP That Actually Holds Conversational Inventory

Buying ads inside AI chat interfaces requires a DSP with direct conversational supply. Criteo and StackAdapt are publicly named technology partners for ChatGPT inventory, but that designation alone does not establish a buyer’s supply path. Migrating an AI ad budget succeeds when the DSP can demonstrate live conversational access, timely bidding, and prompt-level reporting before any budget moves.

Why conversational AI inventory is structurally different from other digital inventory

A conversational placement is a moment inside a live dialogue. Traditional digital buying is built around pages, keywords, cookie profiles, and clicks. An assistant conversation does not provide that same structure. Its commercial opportunity depends on what the user is asking and whether an ad fits the exchange while it is unfolding.

The targeting signal is also more explicit. A user can state a budget, describe constraints, compare a shortlist, and explain what has already been rejected, all in natural language. Those details describe an active decision. However, their usefulness depends on whether the buying system can read them before deciding to bid. Access to a conversational placement does not automatically mean access to its context.

Timing makes that distinction consequential. The ad decision must arrive during a short mid-answer window while the assistant is responding. Format imposes another constraint because a poor-fit placement can interrupt the conversation. Together, these mechanics make the migration a supply-architecture decision.

Which platforms actually sell conversational inventory, and what they sell

Google AI Overviews and AI Mode make placements available through eligible campaigns automatically, without an opt-in or opt-out control. Google announced that it was testing Conversational Discovery Ads and Highlighted Answers at Google Marketing Live 2026. Both are formats built with Gemini that run in AI Mode through AI Max for Search campaigns. Performance Max is the recommended way to buy related formats such as AI-powered Shopping ads. The Gemini app itself has no ads.

ChatGPT presents a different set of buying routes. OpenAI has begun rolling out a beta self-serve Ads Manager through which advertisers can purchase ChatGPT ads directly. Criteo and StackAdapt are also publicly named technology partners for its inventory. A technology partnership does not, by itself, establish what conversational signal, reporting depth, or publisher connection reaches a particular buyer.

Anthropic Claude offers no purchasable inventory at all. Anthropic’s policy is explicitly ad-free, and no inventory is currently expected, although the company has reserved the right to revisit that stance transparently. Anthropic reinforced the policy with a Super Bowl advertisement after OpenAI announced plans to introduce ChatGPT ads. This makes the boundary between an AI audience and an advertising surface especially clear. A widely used assistant is not necessarily an available media channel.

Infrastructure built for conversational ads lets buyers buy ads and connects directly to publishers, while a generalist DSP may rely on another platform’s connection for access. Neither category label proves access to a named assistant. Google’s campaign-controlled placements, ChatGPT’s buying routes, and Claude’s absence of inventory describe a fragmented market.

How most DSPs actually connect to this inventory — and where the chain breaks

Most DSPs claiming conversational inventory access sit one or more intermediary hops away from the supply. That distance can introduce latency, obscure the placement’s origin, and reduce the context available for targeting. The access claim may still be accurate. What it does not establish is whether the channel’s distinctive signal survives the buying chain.

Some supply is not independently available through third-party DSPs. Google and Microsoft inventory flows through their own advertising platforms, so buyers with eligible Google Ads or Microsoft Advertising campaigns may already reach those surfaces without an additional DSP. Those routes do not provide granular AI-specific placement reporting. Moving budget to a DSP cannot, by itself, turn this platform-controlled inventory into an independently selectable conversational buy.

Each additional system sits between the conversational opportunity and the returned decision. During a mid-answer window, an extra hop can disqualify a bid. Generic buying infrastructure’s ability to respond within the assistant’s window is a separate question from its breadth of reach.

Context can disappear through the same chain. Generalist DSPs can reach many surfaces without reading live conversations, while single-surface AI ad networks can read context but reach only their own surface. Combining those capabilities requires direct supply relationships. When an intermediary returns only impression and click aggregates, the buyer also loses visibility into what users were asking when the ads appeared.

What "holding inventory" actually means and why it changes what a DSP can deliver

Holding conversational inventory means running an SSP or exchange that has contracts with AI publishers. The defining feature is pre-bid access to the conversational signal. A direct system sees the context needed to evaluate the opportunity before the auction, rather than receiving a sanitized impression record afterward. This is an operational distinction that a buyer can audit.

The publisher-side component is an LLM SSP. It identifies monetizable moments within chat sessions, evaluates context, runs auctions, and applies publisher-defined controls. A DSP with its own SSP layer participates directly in that architecture. A DSP without it accesses the output of someone else’s auction.

With pre-bid context, a buyer can match an ad to a user’s stated budget, constraints, and intent. Direct integration also removes intermediary steps that can prevent a decision from reaching the mid-answer window. Reporting then completes the connection. With direct supply relationships, buyers can get conversation- or prompt-level data, while access through intermediaries usually gives them only impression and click aggregates. The buyer can therefore evaluate contextual fit.

Publisher controls require the same conversational granularity. An AI publisher of financial news may need to block pharmaceutical promotions in conversations on health topics. A domain-level rule cannot express that distinction because the sensitive context exists within the session. A direct publisher relationship allows controls to operate at the conversation level. Holding inventory changes both what the DSP can target and what the publisher can prevent.

The five questions that separate direct inventory from intermediary access before you migrate a dollar

The buyer’s audit should begin with a supply question. “Which AI surfaces does your inventory actually come from, and can you show direct publisher agreements?” A direct supplier can name the surfaces, explain how it connects to them, and point to contracts or relationships with exchanges. An intermediary may name only the exchange it accesses. For a ChatGPT claim, ask whether the route passes through Criteo or StackAdapt or uses the DSP’s own connection.

“Does your platform read the live conversational signal before the auction, or receive an impression record after it?” Ask which fields the bid request contains. A full-context request should be distinguishable from a standard OpenRTB display request. Post-auction records do not provide the pre-bid understanding needed to target what someone is actively asking.

“What is your decision latency, and how does it compare to the mid-answer response window?” The answer needs a quantified decision speed in this environment. Ask how many systems the request traverses before a decision returns. Unexplained hops leave the central timing constraint unresolved.

“What level of reporting is available — impression aggregate, placement-level, or prompt-level?” Prompt-level reporting exposes the conversational context in which an ad appeared. Aggregated impressions show delivery without establishing that the intended conversational moment was reached.

“How are unsafe or brand-unsuitable conversations identified and filtered, and at what point in the auction?” A topic can shift into sensitive territory mid-session, so pre-bid conversational access is necessary for real-time filtering. OpenAI’s ad principles require clear labeling and independence from answers, making enforcement another concrete question for claimed ChatGPT access. None of these questions requires technical expertise to ask, but the answers separate direct conversational supply from intermediary access.

What the IAB standards framework means for how direct inventory relationships will be verified going forward

IAB Tech Lab’s agentic standards framework extends existing programmatic infrastructure to support agent interactions and contextual signals. Familiar deal types, auctions, and measurement still have a place as the infrastructure grows. For buyers, the significance is that conversational access should become easier to inspect through defined technical requirements.

ARTF, the Agentic RTB Framework, was released for public comment in November 2025. It addresses real-time advertising transactions through a containerized architecture, with support for LLMs and orchestrators. Agentic Audiences, also released in November 2025, defines how agents exchange identity, contextual, and reinforcement signals representing real-time consumer intent. The latter is especially relevant to inventory quality. A DSP cannot demonstrate that it reads those signals if it cannot receive or process them.

AAMP, the Agentic Advertising Management Protocols, was released in January 2026. It organizes buyer and seller agent interactions across foundations, protocols, and trust, with its 3.0 release targeting the process from requesting proposals to buying. CoMP, the Content Monetization Protocols, is still a working group that deals with transactions between publishers and AI companies for content used by LLMs. Its work is relevant to verifiable publisher relationships. These initiatives are at different stages and should not be treated as a completed certification system.

As ARTF and Agentic Audiences become baseline requirements, buyers will be able to request demonstrations of compliance as a proxy for genuine conversational access. Signal support and publisher agreements cover linked but different parts of the supply path. Until the framework offers a more settled way to verify access, buyers still need proof of how each proposed surface is connected, how long decisions take, and what reporting they can get.

How to structure the migration itself — phasing budget, measuring results, and knowing when to scale

The practical migration begins with existing spend. Eligible Google AI Overviews and Microsoft Copilot campaigns already serve automatically, without an opt-out, so a buyer may have conversational exposure without a deliberate allocation or AI-specific reporting. Audit that exposure before funding another route. Then put each prospective DSP through the five questions. A platform unable to answer the questions on signal access, latency, and reporting is not ready, regardless of the surfaces it lists.

The first phase should use a defined test budget on one surface with confirmed direct access. Measure assisted conversions and prompt-level context. Users can go several prompts deep before leaving for the open web to convert, so an influential placement may receive no direct click. The test needs to reveal where the ad appeared within the conversation and whether that context matched the intended purchase intent.

Time-to-conversion provides another part of that assessment. High-intent categories such as flights and electronics have a documented conversion window of 48 hours after high-intent conversations, while other categories can extend up to two weeks. Measurement should account for those differences. Brand safety also belongs in this first phase. Incidents need to be caught at the conversation level.

Cross-surface expansion should follow evidence of contextual match quality. Once the first test returns that evidence at prompt level, additional surfaces can be added through a DSP maintaining direct relationships with multiple AI publishers. An owned SSP and exchange provide the architecture for preserving the signal across those boundaries. Expansion through intermediary aggregation does not offer the same assurance. More reachable impressions are not sufficient grounds to scale.

Scaling is warranted when prompt-level reporting demonstrates contextual fit, safety filtering operates before the bid, and direct supply supports cross-surface reach. Those conditions tie budget growth to the mechanism that makes the channel useful. The choice of DSP should ultimately be judged by whether its supply architecture delivers conversational context matched to purchase intent at the moment the user expresses it.