AI Search Statistics That Matter in 2026

That is the widest supply-and-demand mismatch in search right now: demand multiplied in a year, while the leading assistant has almost nothing to recommend. The mechanism explains it — ChatGPT draws the local businesses it surfaces first from the Foursquare Places API, a listing source almost no local business has claimed, while nearly every local agency optimizes Google Business Profile alone.

That range is two vendors reporting their own sites, not an industry spread: Ahrefs measured 23× against its own signups, Semrush 4.4× against its own conversions. Treat it as the shape of the effect rather than a number to forecast with.

Both figures also understate the channel: a meaningful share of AI referral sessions arrive with no referrer header and land in Direct regardless of analytics setup, so measured AI traffic is systematically under-counted.

The honest read for a local business: AI search is not where most traffic comes from in 2026. It is where the highest-converting traffic comes from, and the visibility numbers above say the cost of entry is still near its floor.

Position inside the page matters more than length. Analysis of 1.2 million citations found the top 30% of a page generating 55% of them, which is why answers belong high and key-takeaways blocks belong above the fold.

Length is not the lever it is sold as: across 174,048 cited pages, Ahrefs measured the correlation between word count and citation position at 0.04 — essentially zero. Read that as "padding does not buy position", not "shorter is better"; SE Ranking's separate count-based analysis did find longer pages collecting more citations. Freshness is a real but modest effect — AI-cited content runs about 25.7% fresher than the organic top ten, not the inflated 4.3× figure that circulates.

Speed shows up too. In SE Ranking's analysis of 216,524 pages, those with first contentful paint under 0.4 seconds averaged 6.7 ChatGPT citations against 2.1 for pages slower than 1.1 seconds — roughly a threefold gap. Tight, fast, answer-first pages win.

Three supporting numbers complete the picture, each from a different publisher. Yext's analysis of 17.2 million citations found first-party websites earning 4.31 citations per cited URL against 2.46 for directory listings — a 1.75× advantage for owning the page you get cited from, not the 4.31× multiplier that gets quoted from that table. In niche business-to-business SaaS categories, Overthink Group's 1,260-prompt run found 70.8% of citations going to "best" and "top" lineups, which is why inclusion in other people's roundups outperforms another service page. And wire-syndicated press releases on channels like Yahoo and MSN account for 0.04% of citations in BuzzStream's four-million-citation dataset: effectively zero return on the oldest PR product still being sold.

The schema numbers deserve precision, because the tactic is still widely invoiced. Ahrefs' matched-control test reported −4.6% in AI Overviews, +2.4% in AI Mode and +2.2% in ChatGPT, the last two statistically indistinguishable from zero. Large language models read visible HTML, not hidden JSON-LD. Schema still earns rich results and Knowledge Graph entity reconciliation — legitimate reasons to do it — but it is not a citation lever.

Related: Google deprecated the FAQPage rich result on May 7, 2026. The visible question-and-answer format is more valuable than ever, because it is what gets extracted; the markup now buys nothing. Any 2026 proposal charging for llms.txt, FAQ schema, or "training data inclusion" is charging for the parts of this discipline the evidence has already retired.