AI Search Statistics
How Often Do AI Engines Name or Cite a Website?
KinetixSEO is an SEO and AI-citation (GEO) checker, and the data below comes from its own production GEO tracking — not a survey, not a vendor claim. Refreshed monthly as more tracked checks land. Rows covering fewer than 30 distinct tracked keyword and prompt pairs are marked as not yet statistically reliable rather than hidden.
Two numbers, never added together: how often an AI model answering from memory named a tracked site, and how often an engine that searches the live web or reads a live results page cited one as a source. They answer different questions, so each has its own table. A third-party mentions dataset we query is shown separately, in neither number.
Data through October 2026 , last updated
Series began August 2026; next snapshot 2 Dec.
Original research (November 2026): 10,000 sites drawn at random from the Tranco top 100,000, 6,767 homepages scanned. SEO Statistics 2026: What 6,767 Real Homepages Show →
Key findings
What the latest reliable data shows
- In October 2026, AI engines that search the live web or read a live results page cited a tracked site as a source in 8% of 627 checks.
- The live-search and results-page engines with a reliable sample in October 2026 (Google AI Mode and Grok) each cited a tracked site as a source in 0% of their checks.
- This benchmark series has produced statistically reliable data (at least 30 distinct keyword and prompt pairs) for 2 months, starting in September 2026.
Monthly rate by engine
The series so far
Cited as a source by live search
Engines that search the live web, or a live results page such as Google’s AI Overview. The rate is the share of checks in which the tracked site was among the sources the answer cited.
| Month | Engine | Cited rate | Sample size | Sentiment (pos / neu / neg) |
|---|---|---|---|---|
| October 2026 | All live-search engines combined | 8% | n=627 checks, 52 keyword-prompt pairs | 0 / 0 / 0 |
| October 2026 | Google AI Mode | 0% | n=75 checks, 31 keyword-prompt pairs | 0 / 0 / 0 |
| October 2026 | Microsoft Copilot | 0%* | n=1 checks, 1 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| October 2026 | Gemini (web search) | 0%* | n=51 checks, 26 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| October 2026 | Google AI Overviews | 18.2%* | n=270 checks, 21 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| October 2026 | Grok | 0% | n=75 checks, 31 keyword-prompt pairs | 0 / 0 / 0 |
| October 2026 | Meta AI | 0%* | n=51 checks, 26 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| October 2026 | ChatGPT (web search) | 0%* | n=51 checks, 26 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| October 2026 | Perplexity | 1.9%* | n=53 checks, 26 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| September 2026 | All live-search engines combined | 21.4% | n=196 checks, 34 keyword-prompt pairs | 0 / 0 / 0 |
| September 2026 | Google AI Mode | 4.9%* | n=41 checks, 25 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| September 2026 | Microsoft Copilot | 0%* | n=3 checks, 3 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| September 2026 | Google AI Overviews | 35.1%* | n=111 checks, 9 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| September 2026 | Grok | 2.4%* | n=41 checks, 25 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| August 2026 | All live-search engines combined | 0%* | n=20 checks, 5 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| August 2026 | Microsoft Copilot | 0%* | n=5 checks, 5 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| August 2026 | Google AI Overviews | 0%* | n=15 checks, 5 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
Mentions in a third-party dataset
A third-party, aggregated dataset of AI answers that mention a domain. We query the dataset; we do not read the answers ourselves, so it is shown on its own and is in neither combined rate above.
| Month | Engine | Mention rate | Sample size | Sentiment (pos / neu / neg) |
|---|---|---|---|---|
| September 2026 | ChatGPT | 0%* | n=9 checks, 8 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
| August 2026 | ChatGPT | 0%* | n=5 checks, 5 keyword-prompt pairs (sample size too small — not yet statistically reliable) | 0 / 0 / 0 |
* Provisional — fewer than 30 distinct keyword and prompt pairs, not yet statistically reliable. Sentiment is only classified when a model answering from memory names the site, not on live-search or results-page readings, and only since August 2026. So the three sentiment counts never sum to the sample size.
Methodology
Where this data comes from
Every row is aggregated from real GEO citation checks run against tracked keywords across KinetixSEO customer accounts — the same checks that power each customer's own AI Visibility dashboard. A citation rate built from fewer than 30 distinct tracked keyword and prompt pairs in a given month is marked unreliable rather than dropped, so a thin early month never silently disappears from the series once volume catches up. See the full scoring methodology.
Series break, September 2026: AI citation checks moved from every night to once a week per tracked keyword and prompt, part-way through that month. From then on each month holds roughly a seventh as many checks per keyword, so check counts before and after the switch are not comparable. Citation rates remain a share of checks run, but treat any move across September 2026 as a possible effect of the change rather than a trend. From the same change, a row is only called statistically reliable when it covers at least 30 distinct tracked keyword and prompt pairs, rather than 30 checks.
From 10 October 2026, a results page that showed no AI answer (no AI Overview or other AI feature on it) is no longer counted as "not cited": there was no answer to cite anyone in. Every earlier month was recomputed the same way, so cited rates for results-page engines are higher than the figures first published, and check counts lower.
Perplexity changed measure on 6 October 2026. Until then its checks were the model's own account of whether its answer named the site, so they sit with the named rates. From that day we read the sources its answer actually cited, so they sit with the cited rates. October 2026 therefore has two Perplexity rows, one per measure.
This dataset is published under a CC BY 4.0 license and is free to reuse with attribution. Reliable rows only — download the CSV.
FAQ
Frequently asked questions about this data
What is the difference between "named" and "cited" in this data? +
They are two separate measures and this page never adds them together. Named: an AI model answers the tracked keyword from its training memory, with no web search, and the check records whether its answer named the tracked domain. That is what the model remembers, not a link a reader can follow. Cited: an engine that searches the live web, or a live results page such as Google's AI Overview, lists the tracked domain among the sources of its answer. Both come from the same automated checks that power each KinetixSEO customer's own AI Visibility dashboard, not a survey or a vendor-reported estimate.
Why do some rows show a sample size warning instead of a rate? +
A (provider, month) row needs checks from at least 30 distinct tracked keyword and prompt pairs before its citation rate is treated as statistically reliable on this page — repeated checks of one keyword are not independent evidence. Thinner rows are still shown, marked, rather than hidden — hiding them would make an early month look like a gap once volume catches up.
How often is this data updated? +
A new monthly snapshot is computed on the 2nd of each month, aggregating the previous full calendar month of tracked citation checks. The "last updated" date near the top of this page reflects when that snapshot was actually computed, not when this page's template last changed.
Can I reuse this data? +
Yes — this dataset is published under a CC BY 4.0 license and is free to use with attribution. A CSV export of the reliable rows is linked from the methodology section below.
Does a higher citation rate mean an AI engine is easier to rank in? +
Not directly. Each rate measures how often an engine named or cited any tracked site for its tracked keywords, not how any single site ranks. Engines also differ in how often they name or cite anyone at all for a given query type, which moves the combined rows independent of any one site's visibility. And a memory engine's named rate and a live-search engine's cited rate answer different questions, so compare engines only within the same measure.
Track your own AI citation rate
Start with a free GEO Readiness check, then track whether ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Claude, Perplexity, Microsoft Copilot, Grok, Meta AI and DeepSeek actually name or cite your site over time.