Google AI Mode: We Tracked 120 Queries for 90 Days. These Are the Exact Content Structures That Earned Citations vs Blue-Link Rankings.

Google AI Mode SEO 2026 planning starts with one finding from RankFast’s tracking: ranking and being cited are separate outcomes. In a 120-query, 90-day study, 41 percent of position 1 pages were not cited in AI Mode for the same query, and cited pages shared two format choices: a direct answer first, and question-form H2s.

Disclosure: RankFast sells AEO services and published this study, so weigh it accordingly. Figures labelled as RankFast’s come from RankFast’s own tracking. Every outside figure is linked to its source so you can check it.

Google AI Mode Different From AI Overviews and Standard Search?

AI Mode is a separate tab inside Search, open to all US users since 20 May 2025, and it answers in a conversational format (Engadget). AI Overviews sit inside the standard results and appear only when Google’s systems judge them additive (Google Search Central). The ai mode vs ai overviews difference matters for tracking, because the two features cite different pages.

Factor	Standard search	AI Overviews	AI Mode
Where it appears	Main results page	Top of standard results, only when Google judges it additive	Separate Search tab
Answer format	Ranked links and features	Short summary with supporting links	Conversational, multi-turn answer
How sources are found	Ranking systems	May use query fan-out	May use query fan-out, and Google says the two features may use different models
Responses with no citation (Ahrefs, Sept 2025, US)	Not applicable	11 percent	3 percent
Cited URLs that appear in both features for the same query (Ahrefs)	Not applicable	13.7 percent	13.7 percent
Citations per query (seoClarity sample)	Not applicable	Not reported	About 12 (12,011 citations across 1,000 queries, my division)

Google says both features may use “query fan-out”, which means issuing several related searches across subtopics and pulling supporting pages from all of them (Google Search Central). That is the main reason a page ranking first for one query can be missing from the answer: the answer was built from several searches, and your page may not rank for any sub-question. seoClarity’s example for a single query showed six searches and 61 sites (seoClarity).

Do Pages Ranking in Positions 1 to 3 Automatically Get Cited in AI Mode?

No. In RankFast’s 120-query tracking, 41 percent of position 1 pages were not cited in AI Mode for the same query, and 34 percent of AI Mode citations came from pages outside positions 1 to 10 in standard search. In 29 of the 120 queries, a position 4 page with strong structure and named entities was cited more often than a position 1 page with weak structure. The denominators for these figures are [insert].

Independent data points the same direction but at a different size. For google ai search citations 2026, the best outside comparison is seoClarity’s study of 1,000 US transactional queries: 81 percent of queries had at least one AI Mode citation from the top 20 organic results, only 28 percent had one from the top 10, and only 19 percent of all citations came from the top 20. Position 1 pages were cited 25 percent of the time, position 2 pages 21 percent, and position 3 pages 16 percent (seoClarity). The author co-founded seoClarity, which sells AI tracking, and the page describes the method only briefly.

Here is the uncomfortable part for RankFast’s own number. If 34 percent of citations come from outside the top 10, then about 66 percent come from inside it. seoClarity’s figures imply a much smaller share inside the top 10. Query mix, collection date, and how rankings were matched to citations could each explain the gap, and RankFast’s query categories are [insert]. Until that is explained, read 34 percent as the result for one query set, not a market figure.

AI Overviews behave differently again. Ahrefs found that 76.1 percent of pages cited in AI Overviews ranked in the top 10, in a sample of 1.9 million citations from 1 million AI Overviews published in July 2025 (Ahrefs). A strategy built on one feature’s behaviour will misjudge the other.

What Content Structure Earned AI Mode Citations in the 90-Day Study?

Two format choices stood out. The first is a direct answer in the first 60 words of the page, before any introduction, context, or author paragraph. The second is H2 headings written as complete questions. This is the content structure ai mode finding at the centre of the study, and RankFast reports that these two choices account for 78 percent of the variance between pages that earned citations and pages that ranked in positions 1 to 3 without one. The method behind that 78 percent is [insert: how variance was measured].

Format choice 1: answer first

A page that spent about 200 words establishing context before answering earned 73 percent fewer citations than an equivalent page that answered first, in RankFast’s tracking. How pages were matched as equivalent is [insert]. The practical rule is a 40 to 60 word answer directly under the H1 that stands alone if a system lifts only that passage.

Format choice 2: question-form H2s

A heading phrased as the question a user would ask tended to be cited more often than a topical label for the same content. Take a law firm page on advertising cost. “What is the average cost per lead for law firm Google Ads?” matches how a query is phrased. “Law Firm Google Ads CPL” does not. The study supports a difference in citation rate between the two forms. It does not support a rule that one is always cited and the other never is.

Two supporting signals

• Named entity density. The second strongest predictor in the study was at least 2 proper nouns per section that an AI engine can verify independently: a named organisation, tool, regulator, or study, ideally linked.
• Domain authority. It was the weakest predictor of the group. This sits uneasily with experience on competitive queries, and it comes from 120 queries, so treat it as a lead to test on your own site.

What Does the 120-Query Tracking Data Show?

The table below is the structure the full dataset will be published in. The RankFast figures in the first table come from the brief. The second table is a template and every cell marked [insert] needs the real number before this page goes live.

Measure	RankFast figure	Denominator or method
AI Mode citations from pages outside positions 1 to 10	34 percent	[insert number of citations]
Position 1 pages not cited in AI Mode for the same query	41 percent	[insert number of queries where AI Mode appeared]
Queries where a strong-structure position 4 page out-cited a weak-structure position 1 page	29 of 120	[insert definition of out-cited]
Fewer citations when the answer starts after about 200 words	73 percent	[insert matching method]
Variance explained by the two format choices	78 percent	[insert model or measure]

Query category	Queries	AI Mode appearance rate	AI Mode citation rate	Rank of cited page	Direct answer in 60 words	Question-form H2
[insert category 1]	[insert]	[insert]	[insert]	[insert]	[yes/no share]	[yes/no share]
[insert category 2]	[insert]	[insert]	[insert]	[insert]	[yes/no share]	[yes/no share]
[insert category 3]	[insert]	[insert]	[insert]	[insert]	[yes/no share]	[yes/no share]
[insert category 4]	[insert]	[insert]	[insert]	[insert]	[yes/no share]	[yes/no share]
All queries	120	[insert]	[insert]	[insert]	[yes/no share]	[yes/no share]

Methodology

RankFast tracked 120 queries across AI Mode, AI Overviews, and standard blue-link results at the same time for 90 days after AI Mode’s wider rollout. The study period, market, device, number of runs per query, how queries were chosen, and who scored each page’s structure are [insert]. Publish these before the numbers. A reader cannot judge a result without them.

Limitations

• 120 queries from one agency is a small sample, and the queries were chosen by RankFast, not drawn at random.
• The data shows association. It cannot show that restructuring a page causes a citation.
• AI answers vary between runs and between users. Ahrefs notes that earlier research found 45 percent of AI Overview citations change between generations (Ahrefs), and I have no figure for how much AI Mode varies.
• Structure scores (direct answer yes or no, question H2 yes or no) were coded by people. Whether the coders knew the citation result is [confirm].
• AI Mode has changed since launch, so a 90-day window describes one period, not a permanent rule.

What Does Google Say About Optimising for AI Mode?

Google says there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”, and that no special schema.org markup is needed. A page only has to be indexed and eligible to show in Search with a snippet (Google Search Central). That sits in tension with a study claiming certain formats earn more citations, and the tension should not be waved away.

Danny Sullivan said on Google’s Search Off the Record podcast, published 8 January 2026, that Google does not want people turning content into bite-sized chunks for LLMs, and that he had heard the same from Google engineers. He warned against making one version for LLMs and another for search, and argued that any gain from chunking may not last (Search Engine Land).

The two format choices here are ordinary editorial practice, not chunking: answer the question first, and write headings that say what the section answers. Apply them to the one page your readers see, not to a second hidden version. Treat the result as a low-cost test on a few pages, and judge it against your own tracking, not against this study.

How Do You Track Whether Your Content Is Earning AI Mode Citations?

Use a manual protocol first. Pick 20 to 30 target queries and run each 5 times in an incognito Chrome window set to your target market. For each run, record whether AI Mode appeared, whether your URL was cited, and which URL was cited instead. Log it weekly in a shared Google Sheet. Five runs per query matter because answers vary between generations (Ahrefs).

Search Console now has some data, but not enough on its own. Google announced Generative AI performance reports on 3 June 2026, rolled out to a subset of sites. They show impressions, pages, countries, devices, and dates, and they group AI Overviews and AI Mode together. Clicks are not listed, and the data also stays in the main performance report (PPC Land).

Paid tools can speed up the logging. A vendor-published comparison says Ahrefs and Profound track Google AI Mode, though that vendor competes with both (LLM Pulse). I could not confirm what Semrush or Peec AI track. Test any tool against five queries you have checked by hand before trusting it. Fastlook is RankFast’s own AI visibility product, and whether it separates AI Mode from other assistants is [confirm].

How Do You Apply This to an Existing Page?

A workable google ai mode content strategy starts from your tracking log, not from a rewrite of the whole site. Anyone searching for how to rank in google ai mode will find advice to write quality content. The steps below are more specific, and RankFast estimates each page takes about 3 hours [confirm].

1. Pick pages that rank in the top 10 but are never cited in your tracking log. These are the cheapest test.
2. Rewrite the opening as a 40 to 60 word direct answer to the page’s main question, before any introduction or author paragraph.
3. Turn topical H2s into questions that match how a person would ask, and put a short direct answer under each.
4. Add at least 2 verifiable named entities per section, each linked to its source: a regulator, a study, a tool, or a named organisation.
5. Add one original data point that the top competing pages do not contain.
6. Re-run the 5-run protocol after Google re-indexes the page and compare with the earlier log before touching other pages.

What Should You Ask an AI SEO Agency Before Hiring One?

Search results for the best aeo agency, the best ai seo agencies, or the top ai seo agencies are mostly lists written by the agencies on them. A fair test of any aeo agency or ai seo agency is whether it can show per-query data, not screenshots. Ask the same of RankFast, which sells aeo services and has an interest in how you read this study.

1. Which queries did you track, how many runs per query, in which market and on which device?
2. How do you separate AI Mode from AI Overviews in your reporting?
3. Can you show one page where a citation appeared after a change, and one where it did not?
4. What does Google say about the tactic you are proposing, and where do you disagree?
5. What is the sample size behind each claim, and who chose the queries?

Frequently Asked Questions