Ads Inside AI Assistants: A First Look at Paid Placements in AI Answers

Ads are now inside roughly a third of commercial AI answers. The number that should actually change your planning, though, is a different one: in 88% of cases, the advertiser paying for that slot was not cited anywhere in the answer sitting above it.

Most coverage of this topic so far has been sightings — screenshots of ads appearing under AI Mode responses, confirmation that Google is testing formats, speculation about what it means. Useful for knowing it's happening. Not useful for deciding what to do.

Two large studies published this summer changed that, and reading them side by side produces a picture that contradicts how most PPC teams are currently thinking about this channel.

Where this actually stands

The timeline is shorter than it feels. Ads began surfacing inside AI Mode responses in late 2025. OpenAI, whose chief executive had described advertising alongside AI as unsettling and a last resort, announced in January 2026 that ads were coming to ChatGPT's free and new lower-priced tiers. By May, self-serve advertising tools were in beta. By mid-2026 Google was building formats specifically to sell these placements.

From zero to a routine feature of commercial answers in under a year.

The two platforms landing within four percentage points of each other is itself notable — this isn't Google doing something unusual, it's the shape the whole category is taking.

One methodological caveat worth carrying through the rest of this article: the AI Mode keywords were pre-selected because they were known to be capable of triggering a text ad, so the sample represents ad-eligible commercial terms rather than all commercial search. And both figures are single-day snapshots of surfaces that vary between sessions. The researchers themselves suggest treating these as floors.

The finding that breaks the standard mental model

Here is where it gets genuinely surprising.

For every keyword that returned a text ad, the researchers checked whether the advertiser also appeared in the list of sources AI Mode cited for that same query. Only 11.53% of advertiser domains showed up among the cited sources. At the exact URL level, just 1.95%.

So for roughly 88% of ad keywords, a brand paid to sit beside an answer that did not mention them.

The obvious objection is that advertisers are simply weaker domains. The researchers tested it — comparing advertisers against non-advertising domains matched on domain trust, backlinks, referring domains and organic standing for the same queries. Even controlling for strength, advertisers were cited no more often than anyone else.

The separation, stated plainly The ad slot is a paid-media decision. Being cited in the answer is an authority and content decision. Neither one moves the other, and budget cannot buy its way across the gap.

If your working assumption was that a presence in AI Mode ads would warm up your visibility inside the answer itself, the data says no. Getting cited remains the work described in generative engine optimisation and in the push to be cited by machines — and it needs its own budget line, not a hoped-for spillover from paid. If nobody currently owns that side of the work, it is an SEO and content authority brief rather than a media one, and the surfaces it targets are the ones covered in what's working in AI Overviews.

Featured Recommendation AD · AFFILIATE

4.3 / 5.0

Adwisely

AI ad automation for Shopify stores running Google and Meta retargeting and prospecting campaigns.

Best for: Shopify stores automating Google & Meta ads

Visit Website

View details Compare

And the advertisers aren't the ones ranking

The mirror image of that finding is just as striking.

Only 2.32% of advertised URLs also ranked organically for the same keyword. At domain level it reached 15.35%. Which means that for around 85% of ad keywords, the advertiser's site did not appear in organic results at all — on any page. That figure held whether the researchers looked at the top ten, top twenty or top hundred.

Part of that is expected: advertisers build dedicated landing pages that were never meant to rank. But the domain-level check catches that, and the number is still 85%.

Read the two findings together and a picture forms. Two largely separate populations are operating on the same commercial queries — one earning citations organically, one buying slots — with remarkably little overlap between them. That's an unusual market structure, and it won't necessarily persist, but it describes right now.

CPC is the only reliable predictor

If you want to know where these ads will appear in your own account, one metric does most of the work.

Splitting keywords by cost per click produced a clean gradient in ad presence: 24.33% under $2, 32.45% between $2 and $10, and 53.56% at $10 and above. Search volume and keyword difficulty showed no direct correlation at all.

That makes intuitive sense. A high CPC is the market already signalling that advertisers will pay hard for that click, and the placement logic appears to be following that signal rather than inventing its own.

The practical instruction is to stop reasoning about this at the account level and start at the keyword level, in the same way you would for scaling ad budgets without losing efficiency. A modest-volume, high-CPC term now deserves more attention than a high-volume cheap one.

Your niche decides more than your bid does

Ad presence varies more by category than by anything you control.

In the AI Mode data, Pets returned an ad on 72.38% of its keywords. Healthcare returned one on 2.64%. That's roughly a seventy-point gap between top and bottom.

The pattern is legible: high-ad niches are lead-generation territory with dense advertiser demand and a clear paid path to a customer, while the low-ad ones cluster around informational and YMYL intent where commercial demand is thinner and where both advertisers and platforms behave more cautiously.

Underneath that sits a second split that matters for competitive strategy. In Entertainment and Hobbies, 49 advertisers shared 2,622 ad appearances — over fifty each. In Pets, 68 advertisers shared 3,514. But in Healthcare, 36 advertisers split just 91 appearances, and in News and Politics, 81 advertisers split 211.

Two different markets are running inside the same product. In a concentrated niche you are competing against entrenched repeat winners who own the slots; in a fragmented one the slot is genuinely open. Which you're in should change how aggressively you enter — and it's a question your competitor analysis should now be answering.

The two surfaces behave differently

Worth separating, because the tactical implications diverge.

AI Mode puts you next to a rival. In 71.1% of ad keywords, the answer carried two competing offers in the same block. Entry into the block is the first contest; standing out inside it against someone answering the same need is the second. That's a familiar dynamic from conventional search, just compressed.

ChatGPT, for now, doesn't. Every ad observed in the ChatGPT study appeared below the generated response as a single sponsored offer with no competing advertiser alongside it. Less cluttered, and arguably a stronger position while it lasts — though "for now" is doing real work in that sentence.

ChatGPT's targeting has a measurable accuracy problem. Semantic analysis found roughly 14.35% of its ads were effectively unrelated to the prompt they appeared beside — about one in seven. And the variance by category is dramatic: as low as 2.6% mismatched in Pets, but above 50% in Relationships and in News and Politics.

The cause is structural rather than a bug being fixed. ChatGPT's system targets on natural-language context hints rather than conventional keyword matching, and advertisers cannot see the specific prompts that triggered their ads. You have a wasted-impression problem and no negative keyword equivalent to fix it with.

What you actually control right now

Less than you'd like, which is worth saying plainly rather than dressing up.

These placements currently require broad match or keywordless targeting, which hands you the least control of any setting available. There is no segmented reporting that isolates AI Mode placement performance from the rest of a campaign. And on the ChatGPT side, you can't see triggering prompts at all.

What that leaves is the two levers that were always the most durable anyway: creative and asset quality, and the page you send people to. When targeting control disappears, post-click quality becomes disproportionately important — which is the argument in post-click optimisation, and the same conclusion the industry has been reaching about automated buying generally in the state of PPC in 2026.

Also worth confirming before you spend anything here: that your conversion tracking is sound. With no placement-level reporting, blended results are all you'll get, and blended results are only interpretable if the underlying measurement is trustworthy.

What this data doesn't tell you

Three honest gaps, because a first look that oversells itself is worse than no look at all.

There is no performance data in any of this. Everything above measures presence — how often ads appear, where, and beside whom. Nobody outside the platforms currently knows what these placements cost relative to conventional search, what they convert at, or whether a click from inside an AI answer behaves differently from a click on a standard text ad. Given that visitors arriving from AI surfaces tend to be further along in their decision, there is reason to expect a difference. Reason to expect is not evidence.

Both figures are single-day snapshots. AI Mode in particular is volatile enough that the same query can return an ad in one session and nothing in the next. A number measured on 30 June describes 30 June. The direction of travel is the reliable part; the decimal places are not.

Both studies are US-only, and the AI Mode sample was pre-filtered. Those keywords were chosen because they were already known to trigger a text ad, so 29.45% describes ad-eligible commercial terms rather than commercial search as a whole. If your market is outside the US, or your terms sit in a category where paid demand is thin, your own manual check matters more than either headline number.

What none of that undermines is the structural finding. Whether the true frequency is 25% or 40%, the separation between buying a slot and being cited in the answer is a design characteristic rather than a measurement artefact — and that is the part worth planning around. It also means the cost pressure story running through rising CPMs in 2026 now has a new surface attached to it, with no benchmark to price against.

What to do this quarter

• Build the high-CPC watchlist and check manually whether ads are appearing on your most valuable terms. This costs an afternoon and tells you whether the channel is live in your market at all.
• Establish which niche market you're in — concentrated or fragmented. It determines whether entering is a fight or an opening.
• Stop treating paid and citation as one project. They are separately funded, separately measured and, on this evidence, entirely separate outcomes.
• Don't restructure your account for this. The volume doesn't justify it yet, and the formats are changing fast enough that anything you build now will need rebuilding. Keep your account structure sound for the channels that are actually spending.
• Watch the disclosure question. Sponsored content inside a conversational answer sits in genuinely unresolved territory, and the rules are moving.

Worth holding alongside all of this: AI referrals remain a very small share of total traffic, as the numbers in where AI search traffic actually lands make clear. This is a channel worth understanding before it matters, not one worth reallocating into today.

The short version

Ads are inside roughly one in three commercial AI Mode answers and one in four ChatGPT commercial prompts, having gone from nothing in under a year. CPC predicts placement; volume and difficulty don't. Your niche swings the odds by seventy points. AI Mode usually puts you beside a competitor; ChatGPT currently doesn't, but misfires on about one ad in seven. And the two findings that should reshape how you think about it: paying for the slot buys you nothing inside the answer, and the brands buying those slots are overwhelmingly not the brands ranking for the same terms.

Two markets, one query. For now.