In-Chat Ads, Defined and Explained

In-chat ads are sponsored placements that appear inside the response flow of a chatbot or conversational AI product. They are distinct from banner ads, interstitials, or the paid results in a search engine. This post covers what they are, what they look like on the page, how they are priced, and the questions a developer or advertiser should ask before integrating.

What in-chat ads are.

An in-chat ad is a disclosed sponsored surface rendered inside a conversational UI. The surface can be a card below a response, a labeled line within a response, a sponsored follow-up prompt, or a sponsored tool-call suggestion in an agentic flow. What unifies them is placement context — each surface sits inside the turn-by-turn conversation rather than alongside it.

In-chat is not the same as "in-app" advertising. In-app implies fixed ad slots — banners at the top, interstitials between screens. In-chat means the ad is tied to the content of a single user turn. The sponsored card that renders under a hiking-boots question is not the same card that renders under a flight-deal question, and the placement logic depends on the model's output.

The category is sometimes called "sponsored responses" or "chatbot ads." The different names map to different formats. In-chat is the broadest label. See advertising for AI agents for the agentic subset.

Anatomy of a Surface.

A Surface is the full unit of in-chat ad inventory. It has five components.

1. Trigger. The signal that causes the ad to render. Triggers can be keyword-based (the user said "sneakers"), intent-based (the model detected a shopping intent), or agentic (the chatbot is about to call a tool). The trigger fires after the model reasons, not before, so the ad does not bias the model's output.

1. Selection. Once a trigger fires, the network selects an ad from eligible inventory. Selection runs an auction on bid, a relevance model, and policy filters. High-bid low-relevance ads should not win, and most mature networks weight relevance heavily.

1. Rendering. The ad is rendered with the required visual components: sponsored label, brand name, copy, optional image, call-to-action. The developer's UI is responsible for the visual structure; the network supplies the data.

1. Disclosure. Every surface has a disclosure element. This can be a badge, a label, a prefix, or a separator. The disclosure is non-optional. Removing it or hiding it is a compliance failure and a reputation failure.

1. Measurement. Impression, view, click, and conversion are the four common events. Networks ship pixels or server-to-server postbacks for each. Conversion windows are usually 7 days click, 1 day view.

The Surface abstraction is what lets a single creative run across multiple placements (card, line, prompt) with consistent disclosure and measurement.

The disclosure contract.

In-chat ads only work if users trust the surface. The disclosure contract is the set of rules that keep that trust intact.

The rules that production networks enforce:

1. The organic response from the model cannot contain a paid recommendation. Sponsored content lives outside the model's generated text, in a labeled slot.
2. Every sponsored surface carries a visible "sponsored" label or equivalent (paid, ad, promoted). The label must be readable at the default font size.
3. The sponsored surface must be visually distinct from organic output. This can be a card boundary, a different background color, or a separator.
4. Clicking a sponsored element must route to a disclosed destination. No cloaking.
5. The sponsored surface cannot be styled to mimic the chatbot's own voice. It is a placement, not a recommendation.

Regulation reinforces these rules. The FTC endorsement guides, the EU Digital Services Act, and the UK CAP code all apply. Read about the broader brand placement framework for how brands think about this.

Pricing models.

In-chat ads run on three common pricing models.

CPM. Cost per thousand impressions. Used for brand campaigns where the goal is visibility. CPMs on chatbot surfaces are typically higher than programmatic display because inventory is scarcer and placement context is stronger.

CPC. Cost per click. Used for performance campaigns. The buyer pays when the user taps the card or prompt. CPC on chatbot surfaces tends to be higher than social because click quality is higher — users click because the card is contextually relevant to a question they just asked.

CPA. Cost per action. Used for direct-response and agentic formats. The buyer pays when a conversion event fires. CPA is the default for sponsored tool-call suggestions because the intent is so high that brands are comfortable paying only on outcome.

Revenue share back to developers is in the 50 to 70 percent range on most networks, with the rest going to the platform for matching, policy, and measurement. This is comparable to mobile ad SDKs and better than most affiliate networks.

How it's different from in-app ads.

In-app ads and in-chat ads look similar on a phone screen but behave differently.

In-app ads are shown in predictable slots. A banner at the top of a screen, an interstitial between levels of a game, a rewarded video at a choice point. The slot is designed into the UI and the ad fills it regardless of what the user is doing.

In-chat ads are shown in context-dependent slots. The slot only exists when the conversation produces a relevant trigger. A user who never asks a shopping question will never see a shopping card. This makes the inventory more scarce but also more valuable per impression.

In-app ads are typically pre-classified — the publisher knows what screen the ad will appear on, and the brand-safety analysis is static. In-chat ads render against dynamic content, so brand safety has to run at placement time.

In-app ads have mature measurement via MMP SDKs. In-chat ads are still building a measurement stack. Attribution is weaker today, and assisted-conversion and lift studies matter more than last-click.

What you should ask before integrating.

For developers evaluating an in-chat ad network, five questions matter.

1. What formats do you support, and which one is the default? If the network leads with in-response paid text without clear separation, walk away. Disclosed card after response is the safest default.

1. What is the latency? Ad calls over 200 ms affect perceived chatbot speed. Ask for P95 latency numbers, not averages.

1. What is the revenue share? A network that takes more than 40 percent is taking too much. Ask for the effective rate after any deductions.

1. How do you handle disclosure? Ask to see the exact creative that will render in your UI. If the sponsored label is small, light, or hidden, the network is setting you up for regulatory trouble.

1. What is the fill rate, and what happens when a slot is unfilled? A network with 20 percent fill but loud house ads is a worse user experience than a network with 10 percent fill and clean blank slots.

See the ad SDK for AI apps for what an implementation looks like in practice.

In-chat ads are a narrow category by impression volume and a wide category by addressable intent. Done right, they monetize without breaking the product. Done wrong, they break the product and invite regulatory attention. The difference is in the Surface design and the disclosure contract, both of which are well-understood in 2026.