AI Overview Presence for Marketing Queries: A Manual Monthly Sample

Key points before the detail:

• Reported AI Overview rates range from 15.69% to about 50% of queries, depending on who measured, which queries they used and when.
• No source read for this page reports a rate for marketing queries. The closest is 82% for B2B technology queries, reported by Omnibound from BrightEdge data for February 2026.
• This page publishes no Kanso sample. It gives a manual monthly method, a log template and the arithmetic for how much a sample of a given size can tell you.
• A sample of 100 queries has a margin of about 10 points at the 95% level, so a month-to-month change smaller than about 14 points is within noise.
• The method is manual because Kanso's working premise is that automated querying of Google is not permitted. Verify that against Google's current terms yourself.

How often do AI Overviews appear in Google Search?

Nobody can give you one rate, because the published figures range from 15.69% to about 50% of queries and each measures a different set. Omnibound's roundup lists Semrush at 15.69% in November 2025, Conductor at 25.11% in Q1 2026 and BrightEdge at 48% in February 2026. It also quotes Google as saying roughly 50% of US queries.

The newest figure in the sources is a 43% to 48% range for Q3 2026, which Safari Digital attributes to Similarweb and Semrush. The useful number for you is the rate across your own queries, and the rest of this page is about how to get it without scraping Google.

What does this page report, and what does it leave out?

It reports published figures and a method, and it leaves out any sample of Kanso's own. No manual check of marketing queries was run for this page, so there is no presence rate, no screenshot and no month-by-month table of Kanso results.

That matters because the title promises a sample, and a title is a promise. What you get instead is the part that decides whether a sample can be trusted: the comparison figures, the rules for collecting one by hand and the arithmetic for reading it. A real sample run to this method would slot into the log below.

One more gap is worth stating. None of the supplied sources gives an AI Overview rate for queries about marketing, SEO or growth, or for India, so a marketer in either position has no published baseline of their own kind.

What do the published prevalence figures say?

They say the share of queries with an AI Overview is somewhere between about 16% and about 50%, with the higher figures from later dates and informational-heavy query sets. The table collects every prevalence figure in the sources, with the period, the scope and the page that reports it.

The table was built by taking each prevalence figure in the three statistics pages, keeping the measurer, the period and the scope as the page states them, and leaving out any figure the page attributed inconsistently. Both reporting pages are secondary sources, and Omnibound's page is dated June 2026.

The spread is easy to state. The highest BrightEdge figure is 48% and the lowest Semrush figure is 15.69%, a gap of 48 minus 15.69, or 32.31 percentage points, and the first is about 3.06 times the second. The two are from different months and different panels, so the gap is not a like-for-like difference.

Why do the published figures disagree?

They disagree because the measurers count different queries on different dates, and the pages that report them rarely describe the query set. Omnibound's own explanation is that Semrush's panel leans toward short research-style keywords of mixed intent, while BrightEdge tracks commercial verticals with longer, more informational queries. That is the page's interpretation, and it does not show the two panels' contents.

Omnibound's advice, which is opinion, is to calibrate to your own keyword portfolio and not to one industry-wide figure. That advice is the case for sampling your own queries. The table lists the places where the sources conflict, so you know which figures to hold loosely.

Read the table as a warning about secondary statistics pages. When a single article gives two statements about branded queries that pull in opposite directions, the safe move is to treat neither as settled and to measure your own set.

Which kinds of queries trigger an AI Overview?

Informational queries trigger the large majority, and the sources agree on direction while disagreeing on size. Elementor says nearly nine in ten triggering queries are informational, and Safari Digital says 99%.

Rates by category and query type also vary widely. The table gives each figure with the page that reports it; the BrightEdge figures are as reported by Omnibound, and the rest come from Safari Digital with no study named.

Two readings follow, and both are analysis. A marketing-queries sample is mostly informational, so it should land nearer the high end of the published range than an e-commerce one. The B2B technology figure of 82% is the closest published benchmark, but it describes technology companies' queries, not queries about marketing.

Elementor adds that over 68% of terms triggering AI Overviews get 100 or fewer monthly searches, and almost 80% of triggering keywords fall in the 0% to 40% difficulty range. If your query list leans on high-volume head terms, expect a lower rate than the headline figures.

What does Google itself say about when they appear?

Google says AI Overviews appear only when its systems determine they add something to classic Search, and that they often do not trigger. That sentence in Google's AI features guide is the official explanation of why no rate can be assumed for any query.

The same guide says a page must be indexed and eligible to show with a snippet to be linked in AI Overviews or AI Mode, with no additional technical requirements. It also says sites appearing in these features are counted in Search Console's Performance report under the Web search type. For the separate report on AI feature impressions, see Kanso's guide to Search Console's Generative AI performance report.

Google's Help page adds three points that bear on sampling. The feature is offered in a listed set of countries and languages, including India and Kannada, and it cannot be turned off, though the Web filter shows only text links without it. The page also carries a notice that Google is updating Search settings, so its text may not match what you see.

Google's guidance is one category of evidence and the statistics pages are another. For a graded view of what is official guidance, research or opinion across AI search tactics, see GEO without the hype.

Why check by hand rather than by script?

Kanso's working premise is that automated querying of Google Search is not permitted by Google's terms of service, so the method here is manual. The sources supplied for this page do not reproduce those terms, so this page cannot quote them. Read Google's current terms yourself before choosing any tool that sends queries to Google on your behalf.

A manual check has real limits, and they should be stated plainly. It is small, it takes time, and one person's screen is one person's screen. The sources do not say how far personalisation or location changes whether an overview appears, so the protocol holds those conditions fixed and records them.

What a manual check gives you is a figure for the queries you care about, collected the same way each month, with every result saved. The vendor panels in the table above (Semrush's 10M+ keywords, Conductor's 21.9 million queries) are far larger, but the extracted text does not describe how those panels collect their queries, and you cannot inspect them.

How do you run a manual monthly sample?

You fix a query list and a set of conditions once, check every query by hand in one short window each month, and log the same fields every time. The steps below are the author's design, built from the limits in the sources above, and not a method any source prescribes.

1. Fix the list. Take your queries from your own Search Console data and your own keyword plan, write down how you chose them, and do not swap queries between months. A Search Console workflow for striking-distance queries is one way to pull a list that reflects what you already rank for.
2. Tag every query. Record the intent bucket, whether it is branded, the number of words and whether it is a question. The published rates split on exactly these lines.
3. Fix the conditions. Choose one country, one language, one device and one browser state, and keep them for every query and every month. Google lists availability by country and language, so conditions are part of the result.
4. Check in one window. Type each query into Google yourself, in one sitting where you can, in the same week each month. Do not use scripts, browser automation or a tool that drives Google Search for you.
5. Record and save. Log whether an overview appeared, whether your domain was linked and what kinds of source were linked. Save a screenshot for each query.
6. Use the Web filter as a control. Google says the Web filter shows text-based links without AI Overviews, so it gives you the plain results for the same query when you need to see what sits below an overview.
7. Compute and compare. Work out the presence rate, then compare it with last month only if the change is larger than the noise (see the sample-size section).

presence rate = queries with an AI Overview ÷ queries checked × 100

citation rate = queries where your domain was linked ÷ queries with an AI Overview × 100

What goes in the log?

The log has one row per query per month, with ten fields, and its job is to make next month's pass comparable with this one. Every field is there because a published figure splits on it or because a result cannot be audited without it.

Keep the log in a spreadsheet and never overwrite a month. If you change the query list or the conditions, start a new series and say so, because the old and new rates are no longer comparable.

How big does a manual sample need to be?

A sample of 100 queries gives a margin of error of about 10 percentage points at the 95% level, and a change between two months must exceed about 14 points to stand out from noise. These figures come from the standard normal-approximation formula for a proportion, not from any source, and they assume the worst case, a true rate of 50%.

margin = 1.96 × √( p × (1 − p) ÷ n )

For n = 100 and p = 0.5, the standard error is √(0.25 ÷ 100) = 0.05, and 1.96 × 0.05 = 0.098, or 9.8 points. For a difference between two independent monthly samples, multiply the standard error by √2: 1.96 × 1.414 × 0.05 = 0.139, or 13.9 points.

A worked comparison uses figures from the sources. Omnibound reports Semrush falling from 24.61% in July 2025 to 15.69% in November 2025, a drop of 8.92 points. Two monthly samples of 100 queries each could not reliably show a change that size, because 8.92 is below 13.9.

Two cautions apply. The formula assumes queries drawn at random from a defined population, and a hand-picked marketing list is not, so treat these margins as a floor on the real uncertainty. And a small sample suits a rough direction and a citation log better than a precise month-to-month trend.

Why record citations as well as presence?

You record citations because presence alone does not say whether your site gained or lost from an overview. Omnibound reports Seer Interactive data, from April 2026, in clicks per million impressions: 33,500 for queries without an AI Overview, 20,743 for cited brands and 9,445 for uncited brands.

The arithmetic is in the figures. Cited brands earn (20,743 − 9,445) ÷ 9,445 = 1.196, or about 120% more than uncited brands, which matches Omnibound's own 120%. Uncited brands get 9,445 ÷ 33,500 = 0.28 of the no-overview figure, less than a third.

The same page reports Seer Interactive's organic click-through rate on AI Overview queries falling from 1.76% in June 2024 to 0.61% in September 2025, then rising to 2.4% in February 2026. Pew Research Center, as reported by Omnibound and Elementor, found users clicked a traditional result 8% of the time with an AI Overview and 15% without, a relative drop of (15 − 8) ÷ 15 = 47%.

The sources also disagree on whether ranking predicts citation. Omnibound reports overlap between cited pages and the organic top 10 at 17% (BrightEdge, February 2026) or 38% (ALM Corp, March 2026), against about 76% in mid-2024. For a manual log, the practical point is to note whether your linked pages also rank, and what kind of source gets linked instead.

That last field has a direct use for sites with tutorial or how-to content. Omnibound reports YouTube at 23.3% of citations in a Surfer SEO analysis, while Safari Digital calls Reddit the most cited source with more than 3 million mentions. The two figures come from different methods, so log source types in your own sample and read your own pattern.

What should you disclose when you publish a sample?

Disclose enough that a reader could repeat your check and get a comparable answer. Every published figure on this page falls short of that in some way, because the pages reporting them rarely say how the queries were chosen.

• The number of queries checked and how they were chosen.
• The date window, the country, language and device.
• That the check was manual, with no automated querying.
• The presence rate, the margin from the table above and the raw counts behind them.
• The intent mix of the list, so readers can compare with the intent table.
• Any change to the list or conditions since the last pass.

For the design habit behind this, writing the method down before collecting data, see the pre-registered content refresh test, which also publishes a protocol with no results yet.

What is documented and what is analysis?

The facts below are attributed to the page that reports them, and everything in the second list is the author's interpretation or design. The split follows the standard for research pages on this site.

Analysis, not documented by any source read:

• That a sample of marketing queries should sit nearer the high end of the published range.
• The protocol, the log fields and the order of the steps.
• The sample-size table and every margin in it, which are computed with a standard formula.
• That the premise about Google's terms is correct. It is Kanso's working premise, and no source read here confirms it.

Not covered by any source supplied: a rate for marketing queries, a rate for India, how personalisation or location changes whether an overview appears, and how the vendor panels collect their queries. For those, the honest answer is that the sources are silent.

AI Overview statistics: frequently asked questions

Sources

• AI features and your website, Google Search Central, page last updated December 10, 2025
• Find information in faster & easier ways with AI Overviews in Google Search, Google Search Help (Android version), read October 28, 2026
• Google AI Overviews Statistics (2026): 56+ Data Points on Coverage, CTR Impact, Citations, and Industry Breakdown, Omnibound, last updated June 2026
• 26 AI SEO Statistics for 2026 & Insights They Reveal, Elementor
• 18 AI Overview (AIO) Statistics Worth Knowing in 2026, Safari Digital