AI chatbot advertising is no longer a forecast. OpenAI has introduced ChatGPT ads, Google is testing sponsored answers inside Gemini conversations, and on 16 September 2026 four marketing and management scholars published an analysis in The Conversation arguing that this shift is structurally different from anything the web has done before. Their case is not that ads are bad. It is that AI chatbot advertising puts the ad somewhere it has never been before.
The argument opens with a quotation that is almost thirty years old. When Sergey Brin and Larry Page were still university researchers, the landmark paper presenting Google warned that “the goals of the advertising business model do not always correspond to providing quality search to users” and that “advertising-funded search engines will be inherently biased toward the advertisers and away from the needs of the consumers.” They wrote that about ranked lists of blue links. The same incentive is now being applied to a system that returns one answer.
This article works through the authors’ case: why AI chatbot advertising differs from banner and sponsored-link advertising, what OpenAI and Google have actually shipped, the four risks the research identifies, why fact-checking and media literacy were built for a different information environment, and what a policy response would need to cover. The piece is written for anyone whose organisation depends on being findable — if that includes you, our AEO services page covers the visibility side of the same shift.
Table of contents
- Why AI Chatbot Advertising Is Not Just Online Advertising Again
- What OpenAI, Google and Others Have Actually Shipped
- The Adoption Numbers That Make AI Chatbot Advertising Inevitable
- The Four Risks the Research Identifies
- Why AI Chatbot Advertising Is So Hard to Audit
- Why Existing Defences Do Not Fit
- The Engagement Problem AI Chatbot Advertising Inherits
- What a Serious Policy Response Would Cover
- How AI Chatbot Advertising Changes What Publishers and Brands Do
- What to Watch as AI Chatbot Advertising Scales
- Frequently Asked Questions About AI Chatbot Advertising
- References
Why AI Chatbot Advertising Is Not Just Online Advertising Again
The distinction the authors draw between AI chatbot advertising and its predecessors is precise, and everything else follows from it.
Ads used to sit beside the information
Online ads generally appear alongside information, as banners or sponsored links. They occupy their own space. A reader can see where the editorial content stops and the paid content starts, even when the labelling is poor.
AI chatbot advertising is inside the conversation
AI advertising is being integrated into the conversation itself, potentially changing the answers you receive. There is no beside in AI chatbot advertising. The sponsored material and the generated material arrive in the same voice, in the same paragraph, in response to the same question.
Why that matters more than labelling
Even with clear labels, users may struggle to distinguish advertising from organically generated information when both appear within an AI conversation. In AI chatbot advertising, a label tells you that money is present. It does not tell you which clause it bought.
The commercial pressure behind it
AI companies are commercialising to offset the cash they are burning through: OpenAI reportedly burned $37 billion in the first quarter of 2026 alone. Against a number like that, AI chatbot advertising is not an optional revenue experiment. It is the business model that the technology is being fitted to.
Chesbrough’s point about business models
The organisational theorist Henry Chesbrough has argued that “technology by itself has no inherent value; that value only arises when it is commercialized through a business model.” Applied to AI chatbot advertising, the insight is that you cannot evaluate a chatbot separately from how it is paid for.
| Property | Web advertising | AI chatbot advertising |
|---|---|---|
| Where it sits | Alongside the content | Inside the generated answer |
| What the user sees | Ranked sources plus ads | One synthesised response |
| Persistence | Public and repeatable | Personalised and ephemeral |
| Third-party auditability | Difficult | Harder still |
| Where the click goes | To an external website | Increasingly nowhere |
| Regulatory fit | Covered by platform rules | Largely unaddressed |
What OpenAI, Google and Others Have Actually Shipped
The AI chatbot advertising market has moved further than most readers assume, and in more than one direction.
OpenAI
OpenAI has introduced ChatGPT ads designed around the way people interact with AI assistants. The company also states publicly that advertising will not change its answers.
Google is testing sponsored answers within Gemini conversations. The company that grew from the paper warning about advertiser bias is now running the experiment that paper described.
The machinery underneath
Both approaches appear to follow closely how advertising technologies already work, including real-time bidding and micro-targeting based on behavioural data. AI chatbot advertising is not a new pipeline; it is the existing one pointed at a new surface.
Perplexity’s retreat
Not everyone has continued. Perplexity, an AI-powered answer engine, tested sponsored conversations and then discontinued the programme — a data point worth more than a press release, because it is a company choosing not to take the revenue.
The zero-click destination
Commercial AI systems are heading towards a “zero-click internet” in which users make purchases inside a corporate-controlled AI environment rather than on independent websites. Attention and commercial opportunity stay with the AI provider. That, rather than the ad unit itself, is the strategic prize.
| Provider | Status | Public position |
|---|---|---|
| OpenAI | ChatGPT ads introduced | Advertising will not change answers |
| Sponsored answers in testing | Framed as a marketing product | |
| Perplexity | Tested, then discontinued | Cited concerns about trust |
| SpaceXAI (Grok) | Prior output-steering incidents | Blamed an unauthorised code change |
The Adoption Numbers That Make AI Chatbot Advertising Inevitable
AI chatbot advertising follows attention, and the attention has already moved.
Half of consumers are already there
Around half of consumers report using AI conversations for search. That is the threshold at which an advertising market becomes worth building, and it has been crossed.
Conversations are becoming relationships
Conversations with AI are now common enough that some users develop dependencies with AI companions. An advertising surface that people return to daily and confide in is qualitatively different from a search box.
Why the 1998 warning took 28 years to arrive here
Brin and Page published their warning in 1998. ChatGPT ads and Gemini sponsored answers are shipping in 2026 — 28 years later, against a system that gives one answer rather than ten ranked links, which is precisely the condition that makes the original warning sharper rather than weaker.
The Four Risks the Research Identifies
The authors’ peer-reviewed work sets out the risks AI chatbot advertising poses to free speech and to free markets. They are worth taking separately.
Compressed perspectives
AI-generated answers compress competing perspectives and may make alternative viewpoints less visible. Where a ranked list showed ten sources disagreeing, an answer shaped by AI chatbot advertising shows one position. Minority voices and dissenting perspectives are the first casualties of compression.
Pay-to-play concentration
If large corporations can purchase preferential visibility within AI systems, smaller firms face a pay-to-play barrier. The predictable outcome of AI chatbot advertising under those conditions is increased market concentration, not increased choice.
Amplified disinformation
The United Nations has warned that embedding micro-targeted adverts into AI-generated content could amplify misinformation and polarising material. Advertisers are not limited to commercial brands; anyone willing to pay can attempt to increase their reach.
Private, unscrutinised messages
Existing ad technology targets messages at individuals using personal data. Applied to AI conversations, misleading or manipulative messages could be delivered privately, without the public scrutiny that accompanies traditional political advertising — the “dark money” problem that AdTech researchers have documented for years.
The precedent already on the record
Steering is not hypothetical. SpaceXAI’s Grok chatbot generated false “white genocide” claims in response to unrelated prompts and produced other extremist content, reportedly because of an unauthorised code change made by an employee. The controlling corporation can change what the model says; that capability exists whether or not it is sold.
Why AI Chatbot Advertising Is So Hard to Audit
The AI chatbot advertising oversight problem is not a failure of will. It is a design property of the medium.
The data is not available to outsiders
Advertising technology is difficult to audit because outsiders often lack access to the data. Researchers who have tried to audit dark advertising on conventional platforms have had to reconstruct it from the outside.
Complexity is used as a defence
AdTech companies weaponise complexity to prevent oversight. A system nobody outside can describe is a system nobody outside can hold to account, and AI chatbot advertising adds a model’s inscrutability on top of an ad stack’s.
Compliance is voluntary
Preventing paid influence from shaping AI-generated answers remains largely voluntary and depends on AI companies honouring their promises. That is not a criticism of any particular company. It is a description of the enforcement architecture, which is currently a set of blog posts.
The “for now” qualification
AI companies promise that advertising revenue will not influence the substance of their answers. As the authors dryly note, such promises may come with a “for now” attached. Commitments that are not written into law expire when the quarter is bad.
Why Existing Defences Do Not Fit
Fact-checking, media literacy and platform regulation were all built for an environment that AI chatbot advertising does not resemble.
Fact-checking assumes a public claim
Fact-checkers assess claims with public relevance: statements from politicians, hoaxes circulating widely. AI conversations are personalised and ephemeral, so a misleading claim made to one user may never enter the public information environment at all. Imagine a chatbot repeating a fabricated allegation about a candidate in a private conversation with a voter — a personalised falsehood can stay invisible to every fact-checker while still misleading the person who heard it.
Media literacy assumes multiple sources
Users are taught to examine sources and compare evidence. Search engines supported that by presenting ranked sources. AI systems increasingly provide a single generated response: not “an” answer but “the” answer. The comparison step the advice depends on has been removed from the interface.
Source poisoning defeats both
A company can flood the web with fabricated reports, poisoning the sources an AI system retrieves. When a consumer asks about the product, the system turns those planted claims into an authoritative answer, complete with citations. The citations make it worse, not better.
The Digital Services Act gap
The European Union’s Digital Services Act is among the most advanced legislation countering disinformation, but it was designed for social media and search engines rather than chatbots. ChatGPT falls under its scope where it functions as a search engine, and the Commission has designated it under the act — yet the rules do not guarantee the accuracy of any individual chat.
| Defence | Designed for | Why it misses |
|---|---|---|
| Fact-checking | Public, persistent claims | Chats are private and ephemeral |
| Media literacy | Comparing ranked sources | One answer, nothing to compare |
| Digital Services Act | Social media and search | Does not cover chat accuracy |
| Ad transparency registers | Publicly visible campaigns | Nothing public to register |
| Sponsored-content labels | Separable ad units | The ad is inside the prose |
The Engagement Problem AI Chatbot Advertising Inherits
AI chatbot advertising arrives with a documented behavioural history attached.
Conflict pays
Advertising-funded social media is addictive and often rewards conflict. Because controversy attracts attention, platforms can convert it into the engagement metrics advertisers value. The cost of that model has been litigated: Meta reached $1.8 billion in settlements over children’s social media addiction in August 2026.
The same incentive, a more intimate surface
A feed optimises what it shows you. A conversational assistant optimises what it says to you, in the second person, after you have told it about your life. The AI chatbot advertising incentive is identical and the leverage is greater.
Why this is a marketing problem as well as a policy one
Brands buying into AI chatbot advertising are buying placement inside someone’s trusted assistant. That is a powerful position and a fragile one — the trust being borrowed is not the advertiser’s, and it depletes. Organisations rethinking their marketing services mix should price that fragility in rather than treating the channel as another line item.
What a Serious Policy Response Would Cover
The authors’ AI chatbot advertising recommendation is short, and deliberately harder than a disclosure rule.
Boundaries, not just labels
Policymakers should establish clear boundaries preventing sponsors from shaping the evidence and substance of AI-generated conversations. A label alone is unlikely to be enough, because the problem is not that the user cannot see the ad. It is that the ad is load-bearing in the answer.
Separation of the retrieval layer
The practical version of that boundary is a rule that paid relationships cannot influence which sources a system retrieves or how it weighs them. Whether that is auditable in a large model is an open technical question, which is itself an argument for writing the requirement before the architecture hardens.
Independent access for researchers
Every AI chatbot advertising audit problem described above reduces to access. Guidelines from bodies such as UNESCO push towards platform governance, but AI chatbot advertising will need a specific right of inspection if any external party is ever to verify the promises being made.
Why “for now” is the whole issue
Voluntary AI chatbot advertising commitments hold while they are cheap. The point of a rule is that it still binds in the quarter when the revenue is needed most — which, given the cash-burn figures, is a quarter that is not far away.
How AI Chatbot Advertising Changes What Publishers and Brands Do
The supply side of the web was built around a click that AI chatbot advertising is quietly removing.
The referral traffic assumption
Two decades of publishing economics assumed that a search result sends a reader to a page, where that page’s own advertising pays for the writing. AI chatbot advertising breaks the chain at the first link: the answer is delivered in the assistant, and the visit never happens.
What replaces the click
What replaces it, in the model the AI providers are building, is a transaction completed inside the assistant. Attention and commercial opportunity stay with the provider. For a publisher, that converts a traffic relationship into a licensing negotiation conducted from a much weaker position.
Why brands cannot simply opt out
A brand that declines to participate in AI chatbot advertising does not become invisible; it becomes describable only by whatever the model already believes about it. Declining is a decision about influence, not about presence, which is what makes the pay-to-play dynamic hard to resist.
The measurement gap
Conventional campaigns report impressions, clicks and attribution. Nobody has published a comparable measurement standard for a sponsored clause inside a generated answer, and the ephemerality that defeats fact-checkers defeats advertisers’ own verification in the same way.
What to do about it in the meantime
Publish the material an assistant would need in order to describe you correctly: specific, checkable, well-structured, and hosted where it can be retrieved. That work is useful whether or not you ever buy placement, and it is the only part of the response that does not depend on a vendor’s roadmap.
What to Watch as AI Chatbot Advertising Scales
Several near-term signals will show whether the concerns in this research are being addressed or absorbed.
Whether anyone publishes a separation policy
The meaningful commitment is not “ads are labelled” but “paid relationships do not affect retrieval or weighting”, stated precisely enough to be tested. No major provider has published one.
Whether Perplexity’s retreat holds
A company that tried sponsored conversations and stopped is the most informative actor in the market. If it returns to AI chatbot advertising under financial pressure, that tells you how durable a trust-based abstention is.
Whether regulators name chatbots explicitly
The Digital Services Act reaches ChatGPT sideways, through its search-engine function. The first instrument that names conversational assistants directly, and addresses sponsorship inside a generated answer, will set the template others copy.
Whether independent researchers get access
Every claim in this debate — including the companies’ own — is currently unfalsifiable from outside. Access for external auditors is the precondition for all of it, and it is the thing least likely to be granted voluntarily.
Frequently Asked Questions About AI Chatbot Advertising
Do ads actually change the answers a chatbot gives?
AI companies say they do not. The research position is that the claim cannot currently be verified by anyone outside those companies, which is the substance of the concern.
Which chatbots carry advertising today?
OpenAI has introduced ChatGPT ads and Google is testing sponsored answers in Gemini. Perplexity tested sponsored conversations and stopped.
How is this different from sponsored search results?
A sponsored search result sits next to organic results the user can also read. AI chatbot advertising sits inside a single synthesised answer, with no parallel organic version shown.
What is the “zero-click internet”?
A pattern in which users complete research and purchases inside the AI provider’s environment without visiting independent websites, keeping attention and transaction value with the provider.
Does the EU Digital Services Act cover this?
Only partially. It was written for social media and search engines; ChatGPT is designated under it where it acts as a search engine, but the act does not guarantee the accuracy of individual chats.
What can a business do to stay visible?
Treat retrieval, not ranking, as the objective: publish clear, verifiable, well-structured source material, and monitor how assistants describe you. Keeping track of model behaviour through resources such as our AI models and tools hub is a cheaper starting point than buying placement.
References
Advertising is coming to AI chatbots — and it could influence the answers you get
The Anatomy of a Large-Scale Hypertextual Web Search Engine
Advertising in AI Chatbots: Early Observations and Considerations
OpenAI: Testing ads in ChatGPT
Google Marketing Live: search ads
Advertising was always going to come for AI chatbots. The real question is how
Commission designates ChatGPT, Reddit and Roblox under the Digital Services Act
Code of Practice on Disinformation
AI in advertising risks fuelling misinformation crisis, UN warns