Configuring a Cookie-Free AI Chat Buy Around Live Prompt Signals Instead of Audience Segments

Buying ads inside AI chat interfaces requires configuration around live conversational intent. ChatGPT Ads Manager, Google’s eligible campaigns, and Microsoft Advertising are ways to buy these ads, and each surface gives buyers different controls. A cross-surface DSP must do more than aggregate inventory. It must evaluate conversational context and coordinate creative, destinations, and attribution without treating a prompt as a user profile.

Why buying in AI chat is a configuration problem, not a settings problem

A practitioner who starts an AI chat buy by importing a Google Search campaign’s keyword lists, match types, and audience segments starts with the wrong targeting model. Those controls do not describe the live conversational signal, creating a structural mismatch. In the cookie-free model described here, each conversation is a discrete targeting context, rather than an extension of a profile assembled from earlier browsing.

That distinction changes what the buyer must configure. Buyers should focus on what the person is trying to accomplish now. What they did last week is outside that context. Writing, bidding, and measurement therefore need a different foundation, even when an existing search campaign supplies access to the inventory.

How the live conversation becomes the only targeting signal

Conversational context includes the current question, earlier exchanges, the AI’s responses, and the follow-up questions that reveal where the discussion is heading. “What about vegetarian options?” becomes legible through the preceding restaurant discussion. The targeting system reads that relationship.

This differs from matching an ad to a static page. Transformer architecture maintains coherence across turns, and conversational targeting draws on the state that supports that coherence. The context must be reassessed as dialogue progresses. A restaurant recommendation can become trip planning, then a budget discussion, then a question about transportation, with a different advertising opportunity at each stage.

Semantic clustering also changes what counts as a signal. Someone talking about team coordination, deadline tracking, and remote collaboration may show intent that fits project management software without naming that category. The meaning supplies the connection that keywords miss. Similarly, budget consciousness can be visible in constraints and trade-offs without the word “price” appearing.

What a context hint is, and the craft of writing one that matches well

A context hint is a freeform, natural-language description of the conversations where an ad belongs. Written at the ad group level, it is matched against the live conversation. It is the buyer’s primary targeting control in this model. A useful hint describes the user’s goal, decision stage, and circumstances under which the offer becomes helpful.

An illustrative strong hint would describe a team discussing missed deadlines and remote coordination, now comparing ways to track responsibilities. An illustrative weak hint would string together project management software, remote collaboration, and deadline tracking as product attributes. The first describes a situation with an emerging need. The second names a category without explaining why an ad belongs in the exchange.

The landing page must support the same scenario because it is also a relevance input. A hint about coordination problems paired with a page addressing a different use case creates a gap. Exclusions need to be as clear as descriptions of where the ad belongs. They should express out-of-scope situations and intent stages semantically, rather than rely on imported negative keywords, subject to the platform’s available controls.

How the relevance-weighted auction works and what it means for bidding strategy

The relevance-weighted, second-price auction has two distinct components. Relevance helps determine which ad can win, while second-price pricing sets what the winner pays. These should not be treated as interchangeable levers. A precisely matched ad with a modest budget can outcompete a vague ad with a larger one.

Jingyi Qiu and Qiaozhu Mei’s May 2026 paper argues that extensions of classical Generalized Second-Price mechanisms, built around position-based allocation, are less suited to fluid, context-dependent LLM responses. That distinction limits the usefulness of importing search bidding logic unchanged. Buyers should base bids on the conversational moments they value.

Early buyers therefore gain more from precise hints than from treating budget as a substitute for fit. The field is still learning this writing discipline. The bidding objective is to win the exchange in which the offer helps.

Structuring the campaign around conversational moments rather than audience tiers

Early exploration calls for creative that helps the user understand options, while narrowing calls for a response to a particular constraint. Decision adjacency warrants a more direct offer. Each stage needs its own hint and creative because the ad enters a different conversation.

A travel brand could illustrate this with three ad groups. The first would address conversations exploring possible trips, the second discussions narrowing plans around budget or transportation constraints, and the third exchanges approaching a specific travel action. The groups would not represent three kinds of people. They would represent three moments when the same person might need different help.

The destination must preserve that distinction after the click. An exploratory ad should not send someone to a page framed for a decision they have not reached. ChatGPT gives buyers direct control over hints and creative per ad group. On surfaces served through existing campaigns, however, earlier campaign decisions may already be shaping AI placements before a team explicitly plans for them.

What the ad unit itself looks like and the creative constraints it creates

The central format is a labeled sponsored card below the AI’s answer, visually separated from the response text. The answer itself is not for sale. Qiu and Mei’s paper describes this separation across ChatGPT, Microsoft Copilot, and Google AI Mode as of mid-2026.

That placement creates a specific creative brief. With low ad load per session, each ad needs to feel like a useful continuation of the exchange. Generic audience copy is a poor fit here. A user who is exploring needs help making sense of the options, while one who is deciding needs a clear next step.

Trust is therefore a constraint on execution. The user has just received an answer they may trust more than traditional search results. An abrupt change in tone or quality can undermine that confidence. Creative should match the conversational stage without implying that the assistant’s answer endorses the advertiser.

Which platforms let buyers advertise inside AI chat interfaces?

As of late 2026, the buyable surfaces described here are ChatGPT, Google AI Mode and AI Overviews, and Microsoft Copilot. They do not offer one interchangeable buying workflow. ChatGPT Ads Manager provides direct control over context hints, creative per ad group, and the landing page.

Google serves these placements through eligible Search campaigns that use broad match or AI Max, as well as through Performance Max and Shopping campaigns. Google recommends smart bidding, with AI Max requiring it. Google launched shopping ads with Direct Offers inside AI Mode in February 2026. Asset quality and feed completeness are key configuration levers. Existing eligibility means an audit can precede any decision to add budget.

Microsoft has served conversational ads since Bing Chat in 2023 and expanded into Copilot Checkout in January 2026. Eligible campaigns and ad types are automatically opted in, with no opt-out and no guaranteed display. Buyers should therefore also check how their existing accounts are set up. Copilot’s position within Microsoft workflows also makes the surface relevant to business audiences.

These routes show how to buy on each platform but do not verify which third-party DSP integrations are available. Meta AI is not directly buyable in this landscape. Advertising elsewhere in a platform’s ecosystem should not be confused with buying inside its assistant.

What a cross-surface buying program needs that no single platform's tools provide

Native tooling optimizes for its own inventory. ChatGPT can draft creative from a website, prefill copy from metadata, and support measurement through partners, but those capabilities serve ChatGPT. Google’s AI Max builds creative per query, while Microsoft assembles Copilot ads from existing account assets.

The useful distinction is between contextual capability and reach. Generalist DSPs offer cross-surface buying but, in the model described here, are not built to evaluate live prompts against ad candidates. AI ad networks that serve a single surface can read context but cannot reach the other surfaces. Access alone is not evidence that a DSP can target what someone is actively asking.

The buyer still owns the program’s common rules. These include one brand voice, product claims that stay the same, different creative angles, and landing pages that fit each moment. Trying different creative should show which need leads to a conversion, beyond comparing the versions that platforms generate. A shared attribution rule must also count conversions once across surfaces, or the apparent performance of each configuration remains disconnected from the others.

Measurement and attribution in a channel where the standard playbook does not apply

Prompt-level attribution remains an unresolved part of AI chat buying. Advertisers get aggregated performance data, but the conversations stay private. Privacy limits what comes back to the buyer. ChatGPT advertisers have asked to see more about prompt-matching and to have more control over negative targeting, but the available information shows that those controls are still limited.

Measurement design must therefore start from observable signals. Buyers should define one attribution rule across ChatGPT, Google AI Mode, and Copilot before launch, rather than reconcile competing rules after spend begins. ChatGPT supports Pixel and Conversions API implementation. OpenAI’s measurement partnerships with Haus, Measured, and WorkMagic explore geo-based experiments to assess causal impact.

Until lower-funnel evidence matures, the channel should be treated as upper- and mid-funnel by default. Incrementality experiments offer a more honest assessment of contribution than relying on last-click alone. Hint performance can still guide configuration changes. It is a proxy for relevance quality.

Brand safety and responsible configuration inside AI conversations

A conversation can change suitability without changing sessions. A product discussion may drift into medical, legal, or financial advice, distress, or misinformation. Safety therefore belongs in the initial configuration.

The practical response is to describe excluded situations semantically. Within the controls available, hints should spell out the sensitive adjacencies, distress signals, and high-stakes decisions where the offer does not belong. A keyword list cannot cover every way to say the same thing. Landing pages also need scrutiny so that they do not introduce claims unsupported by the conversational framing.

Qiu and Mei frame the broader challenge as making commercial influence attributable, measurable, contestable, and aligned with user welfare. Their paper also cites research showing that advertising woven into LLM outputs often goes unnoticed. Visible separation protects the credibility of the interaction.

A working configuration sequence for a first AI chat campaign

The first campaign should begin with two to four conversational moments, described as scenarios. Exploring, narrowing, and deciding provide a starting structure. Those moments become the ad groups where direct control exists. For inherited placements, the corresponding first move is to audit what eligible Google and Copilot campaigns are already serving.

Next, write a hint for each moment, including out-of-scope situations. Read it aloud as a brief to a colleague. A clear scenario should identify the need and stage without sounding like a tag cloud.

Then build distinct copy and imagery for each group, followed by a landing page that continues the same framing. If using ChatGPT’s opt-in text customization, give it inputs clear enough to help it adapt the text in useful ways. The page is part of the targeting logic. Its claims and intent stage must align with the hint as well as the ad.

With that chain in place, set bids around the moments worth winning and complete measurement before spend. Define the shared attribution rule, implement ChatGPT’s Pixel and Conversions API, and consider incrementality testing from the first meaningful spend level. After launch, examine hint quality before raising bids. Repeated improvements in precision build an advantage that compounds while other buyers are still learning to configure the channel.