Advertising for AI agents.
A category that didn't exist in 2024 and now has more than $100 million of funding behind it. Here's what it actually means, what it looks like, and what Surfacedd is building.
What “advertising for AI agents” means
An AI agent is any software that produces output for a user on behalf of an AI model. That includes chatbots, AI image generators, voice assistants, code assistants, and autonomous task agents that complete work across multiple surfaces. The agent is the intermediary. The output is what the user reads, sees, hears, or runs.
Advertising for AI agents is sponsored content placed inside, around, or attached to the output an agent produces. It is distinct from advertising about AI — marketing AI products to buyers — and from advertising with AI, which is using AI tools to generate creative assets for traditional ad channels. Neither of those categories speaks to what users see when they talk to an assistant. This one does.
Contrast the shape of the category with traditional web advertising. There are no cookies to target with. Page views are a weak unit because an agent’s session might produce one long output or a hundred small ones. Ad slots in HTML don’t exist because there is no page. Relevance is decided by the model’s context — the user’s prompt, the conversation history, the surface the answer ships on — not by a third-party identifier. Disclosure is structural: the ad has to be labeled as part of the surface itself, because there is no chrome around the answer to host a disclaimer.
Three properties define the category. First, the ad lives inside an AI surface, not next to it. Second, relevance comes from context, not tracking. Third, disclosure is built into the unit, not added as an option. If a placement misses any of those three, it is not advertising for AI agents — it is something else wearing the label.
What it looks like in practice
The Surfacedd SDK exposes four Surface types today. Each one is a concrete placement with a concrete disclosure. None of them is a banner retrofitted into an AI app.
Text Surface
A labeled sponsored message appended to a chatbot response, or a disclosed brand recommendation attached to a how-to answer. A user asks “best running shoes under $150,” the agent returns its organic answer, and a separate Text Surface appears below with one sponsor, clearly marked. The Surface does not edit what the agent said; it sits after it.
Image Surface
An AI image generator places a named product inside a scene, with a credit line identifying it as sponsored. A user prompts “kitchen counter with coffee setup,” the generator produces the image, and the output includes a disclosed sponsor product — one item, one scene, one credit line. No watermark across the frame, no popup interrupting the download.
Voice Surface
A short, spoken, disclosed sponsor segment before or after a voice assistant’s answer. “This recipe is brought to you by…” is structurally similar to a podcast mid-roll, but shorter — 5 to 10 seconds — and always bracketed by an explicit sponsor cue. The assistant does not slip the brand into its organic answer.
Code Surface
A disclosed sponsor line in the comment layer of a code completion, or a disclosed suggestion for a paid tool inside a developer agent’s recommendation. The ad never goes in the generated code itself; it sits in comments or in a sidecar suggestion that the developer can accept, dismiss, or ignore.
Every example above is how Surfacedd’s Surfaces work today. Nothing implied, nothing imagined.
Who builds it today
The category has real players as of early 2026. None of them looks exactly like the next one, and the spread of approaches is the point — this is a market being defined in public.
Koah Labsraised $26 million for an “AdSense for GenAI” thesis and ships SDKs for JavaScript, React, React Native, Flutter, iOS, and Android. Text-first inventory, with expansion signaled but not shipped at the time of writing. See how Surfacedd differs from Koah for the head-to-head.
Nexad closed a $6 million seed from a16z Speedrun for chatbot advertising, with a hybrid pay-per-click and pay-per-purchase model. Heavy agency tilt; less self-serve.
ProRata.ai has raised more than $75 million and runs publisher-tilted Gist Ads inside AI answer engines. The gravitational center of that product is the publisher side of the market, not the AI app side.
Aryel partners with Criteo on in-chat ads. Dappier works with Sovrn and LiveRamp and lists $5–15 CPM rate cards. Kontext raised $10 million for contextual text ads priced at $2–4 CPM. ChatAds runs a 100%-retention commission model aimed at publishers. Imprezia (YC S25) ships inline contextual mentions. Adgentic exposes an MCP server for agent frameworks.
Those are the ones with public funding and shipping products. The adjacent layer — Carbon Ads, EthicalAds, BuySellAds — built networks for developer blogs and doc sites in the pre-AI web and are being asked whether they fit the AI era. For most of those, the answer is “partially.” See all comparisons for the matrix.
How Surfacedd is different
Three differentiators, each defensible with a specific claim.
Cross-surface by default
Text, image, voice, and code ship in a single SDK. Most competitors today ship one Surface and hint at more. If the AI layer is going to host advertising, the network that matters is the one that can place across modalities without forcing developers to integrate four different vendors. The Surfacedd SDK is a single install.
60/40 revenue share, published.
Developers keep 60%. Surfacedd keeps 40% and covers demand, serving, billing, fraud protection, and disclosure compliance. The number is on the pricing page. Networks that don’t publish their share are making a different bet; that bet is not a better one for an indie AI app.
Removability by design
Any developer can remove Surfacedd from their app in one line of code, at any time, with no penalty and no cooldown. This is a design promise, not a feature. The best signal of an honest ad network is how easy it is to turn off.
See Surfacedd’s side-by-side comparisons for the specifics against each competitor.
Why disclosure is the only defensible position
AI outputs read as authoritative. Users are conditioned to treat a chatbot’s answer as a synthesis of what is true, not a sponsored arrangement. Undisclosed sponsored content inside an AI output is therefore materially worse than undisclosed sponsored content inside a blog post. The trust gradient is steeper, and the failure mode is bigger.
Regulatory direction is clear. The US FTC endorsement guides were updated in 2023 and are being interpreted against AI-generated recommendations. The EU Digital Services Act treats very-large-online-platforms with stricter disclosure obligations, and AI apps that reach scale fall into that bucket. The UK CAP code applies to in-app recommendations regardless of whether the recommendation is generated by a model. Every direction of regulatory travel points at mandatory labeling for sponsored content in AI outputs.
Retrofitting disclosure is harder than retrofitting revenue. Networks built without disclosure as a structural invariant end up with an architecture where every new placement, every new surface, every new advertiser has to relitigate whether and how to label. The networks that survive the next five years will be the ones that built for disclosure on day one.
Frequently asked questions.
What is advertising for AI agents?
How is it different from traditional advertising?
Is advertising in AI outputs legal?
How does Surfacedd compare to Koah or Nexad?
Advertising for AI agents, built to be disclosed.
Join the waitlist. We are onboarding developers and advertisers in the order they sign up.