Plain answers about AI search, each one traced to its source.
We read the studies, court filings and dashboards behind the numbers people quote about ChatGPT, Google AI Overviews and AI Mode. This is what they show so far.
Is AI search taking clicks from my site?
Yes, where AI answers appear.
On keywords that show a Google AI Overview, the top organic result’s click-through rate fell from 7.3% to 1.6% between December 2023 and December 2025. In a controlled experiment, switching people to AI Mode cut their clicks to publishers by 18.8 percentage points.
Source: Ahrefs, 300,000 keywords. Preregistered field experiment with 1,100 participants.
The evidence
Do I need to rank on Google to get cited by AI?
Less than you would think.
61.7% of the pages AI chatbots cite sit outside Google’s organic top 100 for the same query. Rank tracking cannot see most of the places AI citations go.
Source: Profound, 340 million prompts.
The evidence
Does AI prefer new content?
It prefers recently edited pages.
Of AI-cited pages with both dates readable, 72% looked fresh by their last update, but only 42% were first published within the past year. The freshness comes from editing old pages, not from publishing new ones.
Source: Seer Interactive, 7,683 pages cited by ChatGPT, Gemini and Perplexity.
The evidence
Is AI referral traffic worth having?
Small, but it converts.
ChatGPT referrals converted at 15.9%, against 1.76% for organic search. Most AI visits are invisible in analytics: 70.6% arrive with no referrer and show up as Direct.
Source: Orbit Media, 446,000 AI visits.
The evidence
Does llms.txt help AI find and cite my pages?
No evidence that it does.
One cited URL in 94,614 came from an llms.txt page, and Google says on the record that its AI systems do not use the file. Of 584,107 llms.txt files in one crawl, 68.27% were templated.
Source: Allmo citation analysis, June 2026. Common Crawl, July 2026 crawl.
The evidence
Can I pay to appear in ChatGPT’s answers?
Not with ads, on current evidence.
Ads appeared on 25.94% of 50,006 commercial ChatGPT prompts. The advertiser was missing from the answer’s own cited sources 96.37% of the time.
Source: SE Ranking, 50,006 prompts. The only public measurement so far.
The evidence
Can I control what AI companies take from my site?
Partly, and the trade is hard to measure.
Google’s new Search Console opt-out removes a site from AI Overviews and AI Mode, but its reporting carries no clicks. In one index, 39.5% of sites that disallow OpenAI’s GPTBot still served it a page.
Source: Google Search Console documentation. AI Crawler Block Index raw counts.
The evidence
Can I trust the AI search statistics I see quoted?
Check the source first.
Most numbers we traced were real, but changed on the way: a new denominator, a dropped range, a different firm credited. A LinkedIn post quoted 93% where its source printed 92%. A Reuters Institute explainer attached Pew’s 1% AI Overview click rate to over 68,000 searches, where Pew prints the base for it as 12,593.
Source: Our traces of quoted statistics back to their primary source, since July 2026.
The evidence
Latest entries
All 32 entriesPew's 1% counts 12,593 searches. The sentence quoting it says 68,000.
A Reuters Institute explainer of 13 October attaches Pew's 1% AI Overview click rate to over 68,000 searches. Pew's base is the 12,593 that produced a summary.
LinkedIn's VP quotes 93%. The report he cites prints 92%.
A LinkedIn marketing post cites 93% of the 24 most-cited articles. The Meltwater page it links prints 92%, which is 22 of 24. The studies under it disagree too.
39.5% and 47.1% describe the same test. One has the wrong denominator.
A crawler-block index reports that 39.5% of GPTBot bans go unenforced. The chart in its own README says 47.1%. The raw data file shows what changed underneath.
The release says 60%. The study it links says 68.01%.
A paid release headlines 60% of searches ending without a click. The SparkToro study its own report page links reports 68.01%, counted a different way.
One test says 93 of 100 brands. The study citing it says 63 of 100.
Ahrefs' branded AI Overviews study cites a 93-of-100 brand test and reports 63 of 100 from its own panel. Five published numbers, five denominators.
Three click-through ranges in one filing. The widest ran in neither story.
A 92-page federal filing gives three Microsoft click-through ranges. TechCrunch printed one number, TheWrap printed two ranges, and 51% to 94% ran nowhere.
The index, monthly. Who AI engines cite, who they dropped, every number sourced.