A Google AI Mode Ad Does Not Buy a Citation: 50,032 Ad-Eligible Keywords, Tested

The ad slot and the source list are different doors

A performance marketer bids on an expensive commercial term, wins the placement, and watches her ad render inside a Google AI Mode answer. Then she scrolls to the bottom of that same answer, to the list of sources the model says it read. Her domain is not on it. Neither is her landing page.

That is the normal outcome, not the exception. SE Ranking analyzed 50,032 commercial keywords across 20 niches, roughly 2,500 each, and every keyword in the sample was chosen in advance because it was already known to trigger a website text ad. On 30 June 2026, in the United States, 14,733 of them, or 29.45%, returned a text ad inside AI Mode. Read that number with its condition attached. It is a hit rate inside a set curated to hit, not the share of commercial searches that carry an ad, and it comes from a vendor that sells AI Mode and competitor ad tracking, from one day, in one country, with no independent replication.

The finding that does not lean on any of that is the one worth your morning. Within the keywords that did show an ad, only 11.53% of advertiser domains appeared among the sources AI Mode cited, and just 1.95% at the exact URL. SE Ranking's own section heading puts it bluntly: "88% of brands aren't cited for the keyword they advertise on." Paying for the slot bought the slot. It did not buy a mention in the answer above it.

What the study actually measured

Before you quote any of these numbers in a planning doc, it is worth being precise about what was counted, because the headline figure and the durable figure have very different exposure to the method.

A keyword set built to trigger ads

The methodology is stated in the study without hedging: 50,032 commercial keywords across 20 niches, roughly 2,500 each, "all selected to trigger a website (text) ad." That is a deliberate design choice, and a reasonable one if your question is what the ad layer looks like where ads are possible. It is not a random sample of commercial search. So 29.45% answers "how often does an ad-eligible commercial keyword actually surface an ad in AI Mode," and it does not answer "how much of commercial search now has ads in it." Search Engine Journal made the same point in its writeup, adding that SE Ranking sells the tracking products this research showcases.

One day, one country, one vendor

Data was collected on 30 June 2026 and reflects AI Mode in the US. Four limitations travel with every number below:

• One collection date, on a surface the study itself calls volatile.
• One country, no international sample.
• Vendor research, from a company selling the trackers it showcases.
• No independent replication by anyone.

The study's own disclaimer is unusually honest for vendor research: the results "describe patterns in this specific dataset and may not hold for every keyword, niche, region, or point in time." The large keyword studies published this year by other tool vendors measured AI Overviews, a different surface from AI Mode, so they can neither confirm nor contradict any of this.

Why the caveat cuts toward understatement

Here is the part most summaries drop. SE Ranking argues 29.45% should be read as a floor rather than a ceiling, because AI Mode is volatile and the same query can return an ad in one session and nothing in the next. A single-snapshot measurement of an unstable surface undercounts. So the preselection caveat and the volatility caveat push in opposite directions, and the honest reading is that nobody currently knows the true prevalence. That uncertainty is exactly why you should not build a budget on the 29% and should build one on the structural finding instead.

The citation gap survives every caveat

The 11.53% figure is a within-dataset comparison. Both sides of it, the ads and the citations, come from the same 14,733 keywords, measured in the same sessions, so the preselection has nothing to distort. This is the number to carry into a strategy meeting.

11.53% at the domain, 1.95% at the URL

For every keyword that returned a text ad, SE Ranking checked whether the advertised page or its site appeared in the source list AI Mode generated for that same query. The domain showed up 11.53% of the time. The exact advertised URL showed up 1.95% of the time. Search Engine Journal's writeup reports the same split and spells out the denominator: these percentages are computed over the keywords that showed an ad, not over all 50,032 tested. Worth being clear that the trade coverage is a summary of this one release rather than a second measurement, so the study still stands alone.

Matched on brand strength, still not cited

The obvious objection is that big brands both advertise and get cited, so any correlation would be brand strength rather than ad spend. SE Ranking tested it. It compared advertisers against non-advertising domains of the same strength for the same query, matched on Domain Trust, backlinks, referring domains and organic standing, and reports that "when we controlled for brand strength, buying an ad gave no citation advantage." The methodological detail behind that control is thin in the published writeup, so hold it as a strong directional finding rather than a proven law. It still points the same way as the raw numbers.

Two channels that do not feed each other

The mechanism is not mysterious once you name it. A slot in the sponsored block is allocated by an auction. A place in the source list is allocated by whatever retrieval and ranking logic assembles the answer. They read different inputs. Spending more in one has no lever on the other, in the same way that a bigger media budget never moved your position in a set of organic results. If you have been treating "we show up in AI Mode" as a single achievement, the study is telling you it is two, and you are probably only buying one of them. That is the same split we described when Google started testing AI summaries under search ads: the machine takes the framing layer, and your asset has to earn its own place.

Ads inside AI Mode are not a rumor

None of this is a prediction about where Google might go. The company has already described the formats, on its own blog, in its own words.

The two formats Google named

At Google Marketing Live, Google wrote that "Built with Gemini, we're testing new ad formats in Search and expanding our Direct Offers pilot to help brands connect with consumers," and named two of them for AI Mode.

• Conversational Discovery ads, where "your ad answers a person's specific question," with creative built for that search.
• Highlighted Answers, where AI Mode produces a list of recommendations and "highly relevant, high-quality ads are eligible to appear on that list as a Highlighted Answer."

Search Engine Journal notes both are still in testing and not widely available yet.

An independent explainer sits beside your creative

The design detail that matters most to a creative team is this one: Google says "both of these new formats will feature an independent AI explainer as part of the ad," and that its model "evaluates and synthesizes information about a product or service, and displays that context alongside the advertiser's creative." Your ad is no longer a monologue in its own box. A second, machine-written voice describes your product next to your own words, and you did not write it. Vague copy gets paraphrased into the same sentence as your competitor's vague copy.

Labeled, and still moving

Google says the formats "will also continue to be clearly labeled as" sponsored units, which lines up with the wider push toward visible AI labels on ad surfaces. Combine that with the testing status and you get a surface that is real, disclosed, and unfinished. Planning around today's exact behavior is a mistake. Planning around the separation of paid placement and citation is not, because that separation is architectural.

Where the ads actually land

If you buy search, the useful part of the dataset is not the average. It is the shape of the distribution, because it tells you which of your keywords to look at first.

CPC predicts placement, volume does not

Ad presence rose in a straight line with cost per click: 24.33% on keywords under $2, 32.45% between $2 and $10, and 53.56% at $10 and above. Search volume and keyword difficulty showed no comparable correlation. SE Ranking does not publish per-band sample sizes, so read this as a strong correlation in the data rather than a validated predictor. The practical move is still obvious: sort your commercial keywords by CPC, not by volume, when you want to know where AI Mode ads are likely to show up.

Your niche decides more than your keyword

The spread between categories is enormous. Pets keywords returned an ad 72.38% of the time. Healthcare returned one 2.64% of the time, a gap of roughly 70 percentage points. The high-ad categories cluster around lead generation with a clear paid path to a customer; the low ones sit in informational and sensitive territory where both advertisers and Google move more carefully. Scope the opportunity at the category level before you price a single keyword.

Concentration follows the same pattern. Across keywords with text ads, 25,243 ad appearances came from 2,930 unique advertisers, which would be about 8.6 appearances each if they were spread evenly. They are not.

• Entertainment and Hobbies: 49 advertisers split 2,622 appearances.
• Healthcare: 91 appearances spread across 36 advertisers.

In some niches you are fighting a small club of repeat winners. In others the slot is genuinely open, and SE Ranking warns that these advertiser-level figures are the most keyword-sensitive numbers in the study, so read them as market shape rather than a ranking.

You are almost always sharing the block

In 71.1% of ad keywords, the answer contained two ad items shown together, and only 28.9% returned a single one. The default sponsored block is built to hold a pair of competing offers. So you have two contests, not one: entering the block, then being the more convincing of the two offers inside it. That second contest is a creative problem, and it is the one most teams underfund.

Advertisers are missing from organic too

The citation gap is not an isolated quirk. The same advertisers are largely absent from the ordinary results for the terms they pay for.

2.32% at the URL, 15.35% at the domain

Only 2.32% of advertised URLs also ranked organically for the same keyword. At the domain level it rose to 15.35%, and that figure held identical against the top 10, top 20 and top 100. So for roughly 85% of ad keywords, the advertiser's site did not appear in organic results at all, on any page.

Landing pages built not to rank

Part of the URL gap has a boring explanation, and the study says so: advertisers build dedicated campaign landing pages that were never meant to rank. That is why the domain-level check exists. The domain-level gap is the one that should bother you, because it survives that explanation and it holds all the way down to position 100.

Three scoreboards, one budget

Put the three findings side by side and the picture is a single, uncomfortable shape.

Where you can appear	How often advertisers did	What buys it
Sponsored slot in AI Mode	29.45% of ad-eligible keywords	The auction
Cited source in the answer	11.53% at domain level	Authority and content
Organic result for the same term	15.35% at domain level	Ranking work

Three doors, three keys, one budget that most teams still report as one line. If you are judging any of this by surface metrics, click-through rate will mislead you first, because it cannot tell you which door the visit came through.

What to do with this if you buy ads

The study is descriptive. Here is the operational read, in the order a small team can actually execute it.

Separate the budgets and the reports

Split AI Mode placement, organic ranking and source citation into three tracked outcomes, even if one person owns all three. The overlap in the data is small enough that crediting them jointly will produce a wrong conclusion within a quarter.

• Placement: did our ad enter the sponsored block, and did it sit beside one competitor or none?
• Citation: is our domain in the source list for the queries we care about, ad or no ad?
• Ranking: does any page on our site appear organically for the terms we bid on?

This is a creative and channel analytics discipline before it is a media-buying one, and it is the same instinct behind defending branded terms as their own line item rather than folding them into a blended number.

Feed the channel you can still control

Inside the sponsored block, targeting is increasingly automated and increasingly out of your hands, and the answer around your ad is machine-written. What remains yours is the asset:

• The face and the demonstration, which no explainer paraphrases away.
• The first two seconds, where the choice between two offers is usually made.
• The one specific, checkable claim that survives being restated in neutral language.

That is not a consolation prize. When two competing offers sit in the same block and a neutral explainer describes both, the differentiator is whichever creative gives a person a reason to care. The quality gap between AI ads that work and ones that do not has never been about the tooling.

A worked example

Take a supplement brand with 40 commercial keywords, split by cost per click. Applying the study's rates to each band:

• 8 keywords above $10, at 53.56%, gives roughly 4 with an ad.
• 12 keywords between $2 and $10, at 32.45%, gives roughly 4.
• 20 keywords under $2, at 24.33%, gives roughly 5.

That is about thirteen keywords where AI Mode ad exposure is plausible. On the citation side, at 11.53%, you would expect your domain in the source list for one or two of those thirteen. So the correct plan is not "win AI Mode." It is: bid the thirteen, build content that earns citation on the queries you care about regardless of the auction, and produce enough ad variants to win the second contest inside the block. At roughly $2 to $11 per AI-generated clip, twelve variants is a normal week rather than a production budget, which is why UGC-style ad formats end up carrying most of this load. Producing at that volume does put a second question on you, which is being able to say where every layer of the asset came from, and the AI music lawsuits are where that bill is arriving first.

How Novoads solves the creative supply problem

When placement is bought at auction and citation is earned elsewhere, the only input left that you fully control is the creative, and the constraint on creative has always been supply. Novoads turns an uploaded product image or a written script plus an AI actor into an ad-ready vertical video, so a team can put twelve angles into the block instead of one hero asset and let the results choose. Voices cover 31 languages, and access starts at $1 for 3 days before the standard plan begins. It does not buy you a citation. Nothing does. It buys you enough shots at the contest you can actually win.

Visibility is three scoreboards now, not one

For twenty years, being visible on Google meant a position, and money moved you up it. Inside an AI answer, visibility fractured into three separate things that happen to share a screen: the slot you buy, the ranking you earn, and the citation the model grants. The study's most useful sentence is the one about matched brand strength, because it says the connection between the three is not weak, it is absent. Budget them as three, measure them as three, and put your creative energy into the one place where paying more still changes the outcome.