ChatGPT Ads: How AI Advertising Will Change Marketing

Every advertising platform before this one could sell the answer.

Google sells the top of the results page. Meta sells the feed position. In both cases the commercial placement occupies the same real estate the user came for, and the platform decides how much of it to sell.

ChatGPT cannot do that, and the reason defines the entire product. If the model’s answer can be bought, the answer stops being trustworthy, and an assistant nobody trusts has no users to advertise to. So the ads sit below the response, clearly labelled and visually separated, and OpenAI’s stated commitment is that the response itself is never influenced by advertiser payment.

That single constraint shapes the format, the inventory ceiling, and what marketers can realistically expect. Here is what has actually shipped, and what changes.

Dates and figures below reflect reporting up to mid-2026. This platform is moving quickly — treat every specific as a snapshot.

What Has Actually Shipped

Who sees them. Logged-in adult users on the Free and ChatGPT Go tiers. Paid tiers — Plus, Pro, Business, Enterprise — do not see ads.

Where they appear. As a labelled “Sponsored” placement below the answer, visually separated from it. Reporting also describes a contextual sidebar placement adjacent to the conversation, better suited to reinforcement than direct response.

Where it is live. A deliberately narrow first wave, reported as around seven markets including the US, UK, Canada, Australia, Japan, Korea and New Zealand. One analysis noted this covers a small share of global population but a large share of world GDP, which is consistent with a rollout targeted at markets with advertiser budgets rather than user volume.

How you buy. Self-serve at OpenAI’s advertiser platform, with CPM, CPC, and conversion-optimised CPC objectives.

The Targeting Is Genuinely New

This is the part worth understanding properly, because it does not map onto either incumbent.

ChatGPT ads combine the intent signal from what someone just asked with behavioural context from their chat history and saved memories, plus past ad interactions.

Search has the first without the second: Google knows what you typed, not the six months of context behind it. Social has the second without the first: Meta knows a great deal about you, but you were scrolling, not asking.

Having both at once is a new targeting position, and it is why early pricing carried a premium.

Two things it explicitly does not do, per OpenAI’s stated policy: it does not rely on third-party cross-web tracking, and advertisers do not receive conversation content.

What It Costs

Handle these figures carefully, because they moved substantially within months.

During the invite-only pilot, reporting described a substantial minimum spend — figures varied by source, generally in the tens of thousands of dollars and in one case reported considerably higher — alongside a reported CPM around $60. For context, Meta’s average CPM across its platforms is commonly cited in the $14–20 range.

With the self-serve rollout, the fixed minimum was reportedly removed, opening access to small and mid-market advertisers rather than only large brands.

If the pilot pricing was broadly accurate, advertisers were paying several times social CPM rates for the intent quality. Whether that premium holds as inventory expands is the open commercial question, and it is the number to watch.

Revenue projections circulating for the platform vary wildly between sources and are mostly unattributed. Treat any specific figure as speculation rather than disclosure.

The Inventory Ceiling Nobody Mentions

Two structural limits follow from the design, and they matter for planning.

Ads only run on free tiers. OpenAI is simultaneously selling an ad-free subscription and an advertising product. Every user who upgrades removes themselves from the addressable audience. That is an unusual tension, and it caps inventory in a way search and social do not experience.

The answer is not for sale. Because the placement sits below the response rather than inside it, there is a hard limit on how much commercial content can be inserted before the experience degrades. Compare that with a search results page, where the proportion of the screen given to ads has expanded steadily for two decades.

The practical read for advertisers: scarce inventory in a high-intent context. That supports premium pricing and limits volume.

The Creative Problem Is Different

The most important practical shift, and the one most likely to be underestimated.

Your ad reaches someone mid-thought, not mid-scroll. They asked a question and received an answer. Your placement lands in that moment as either a useful extension of the conversation or an interruption of it.

That changes the brief:

• Relevance to the immediate question beats brand awareness. An ad that does not extend the answer is noise, however good the creative.
• Formats are constrained. Reported specifications describe tight character limits on titles and body copy. There is no room for the layered messaging a feed unit allows.
• The user is closer to a decision than a social audience but has less patience than a search user, who chose to scan a page of options.
• If your product only loosely fits the questions people ask in your category, budget does not fix that. The context match is doing most of the work.

What Changes for Marketing

Six shifts worth planning for.

A third major intent channel appears. For twenty years, high-intent digital advertising meant search. This is the first credible alternative with a genuinely different signal.

Category fit becomes a spend qualifier. Products that answer a question people actually ask an assistant will perform. Products that require creating demand will struggle in a format triggered by user questions.

Measurement will lag the format. New channels arrive before attribution does. Expect a period where the results are real and difficult to evidence.

Retail gets there first. Product feed support means catalogue advertisers can scale immediately, which is why that capability shipped early.

Agencies moved before most brands. Major holding companies established partnerships early, which historically indicates a channel expected to matter.

Organic and paid become separate problems on the same surface. This is the strategically important one.

Earning the Answer vs Buying the Slot Below It

There are now two distinct ways to appear when someone asks an AI about your category.

Earn the citation. Being named inside the answer is won through association and credibility — third-party coverage, comparison content, community evidence, consistent positioning. It cannot be purchased and it takes quarters.

Buy the placement. The sponsored slot below the answer is available immediately, at a price, to anyone in a live market.

These are separate systems with separate skills, and they will increasingly be run by separate teams. The brands that do best will treat them as complementary: paid placement while the organic association is being built, and organic citation reducing dependence on paid once it exists.

The obvious risk is treating paid as a substitute for the organic work. Being the sponsored option beneath a recommendation of your competitor is a considerably weaker position than being the recommendation.

Open Questions

• Does the CPM premium survive scale? Prices set against scarce pilot inventory rarely hold as supply expands.
• How much ad load will users tolerate before the free tier degrades enough to affect usage?
• Will attribution mature? Conversion measurement in a conversational context is genuinely hard.
• Do other assistants follow? The competitive position of a single ad-supported major assistant looks different from a market where several are.
• Where does regulation land? Disclosure standards for advertising inside AI-generated experiences are not settled, and the current labelling approach may not be the eventual requirement.

Final Thoughts

The significant thing about ChatGPT ads is not that a new platform exists. It is the constraint the platform accepted: the answer is not for sale, and the advertising has to live around it.

If that holds, this becomes a high-intent, scarce-inventory channel where relevance to the immediate question does most of the work — which rewards genuine category fit rather than budget. If it erodes, the format converges on everything that came before it.

For marketers the immediate move is small: if you are in a live market and your product answers questions people actually ask, test it while inventory is scarce and competition is thin. And keep building the organic association, because the placement below the answer is worth much less than being in it.