The state of AI search
Every day, AI engines crawl, cite, and recommend millions of pages across 4,897 properties and 8 models. This is the live picture of what they are doing — and who is winning.
What the data says right now
/ The three biggest shifts91% of all AI referral traffic now comes from ChatGPT alone — more than every other model combined.
Ask eight models the same question and their answers overlap just 43% of the time.
88.5% of pages are fetched exactly once, then never crawled again — first impressions are everything.
Rankings
Who's winning right now?
Every model is quietly ranking your brand against everyone else. This is the live AI 500 — 9,000+ brands scored on how often AI engines surface them, across 500 industries. The day's biggest movers are up top.
Unable to load rankings data
Traffic
Who's sending the traffic?
AI engines don't just answer questions — they send people. ChatGPT alone drives 91% of every AI referral we measure, more than all other models combined. Here's where the visits actually come from.
ChatGPT
Gemini
Claude
Perplexity
Grok
DeepSeekAI referral traffic by source
Rescaled so the highest visible point = 100%
Aggregated from 4,897 anonymized GA4 properties. Market share = % of total AI sessions (30d). Updated daily.
Citations
Who gets recommended?
When AI cites a source, it cites the same few over and over. Citations follow a steep power law \u2014 and most brands live in the long tail.
You've seen how AI picks its winners. Where does your brand land?
Track your citations, rankings, and movement across ChatGPT, Claude, Gemini, Perplexity, and Google AIOs — updated daily.
Find your brandModel agreement
Do AI models agree?
Ask eight models the same question and you'll get eight different answers. On average they agree just 43% of the time — Claude and Deepseek line up most, while Meta and Perplexity diverge hardest. Winning in one model tells you almost nothing about the rest.
| Open |  Clau | Gemi |  AIO | Grok | Deep |  Meta | Perp | |
|---|---|---|---|---|---|---|---|---|
| OpenAI | - | 27 | 21 | 20 | 24 | 25 | 17 | 17 |
| Claude | 27 | - | 26 | 19 | 35 | 35 | 23 | 15 |
| Gemini | 21 | 26 | - | 17 | 24 | 26 | 16 | 12 |
| AIO | 20 | 19 | 17 | - | 17 | 18 | 12 | 17 |
| Grok | 24 | 35 | 24 | 17 | - | 31 | 21 | 14 |
| Deepseek | 25 | 35 | 26 | 18 | 31 | - | 22 | 15 |
| Meta | 17 | 23 | 16 | 12 | 21 | 22 | - | 10 |
| Perplexity | 17 | 15 | 12 | 17 | 14 | 15 | 10 | - |
Crawlers
How do they find you?
Before AI can recommend you, its crawlers have to read you. ChatGPT's bots now account for 72% of all crawler visits — more than every other crawler combined — and 88.5% of pages are fetched exactly once, then never again.
ChatGPT TrainingChatGPT SearchMeta
AmazonClaude TrainingContent
What content wins?
Crawlers visit everything; models only cite a little of it. Review / Directory pages punch far above their weight — 4.8× cited relative to how often they're crawled — while Homepage pages are heavily crawled but rarely recommended. That gap is where AI visibility is won.
Markdown
Do crawlers prefer Markdown?
So we ran an experiment. We serve half of 9,035 eligible pages as Markdown and half as HTML, then watch which version AI crawlers fetch. Markdown is 79% smaller — but so far no crawler shows a decisive preference. The experiment is still running.
llms.txt
Does llms.txt actually help?
Everyone's rushing to add an llms.txt file. We checked 37,894 AI-cited domains — only 13.3% have one, and there's no statistical sign it lifts citations (p = 0.85). The honest verdict, for now, is inconclusive.
of AI-cited domains have adopted llms.txt. While adoption is growing, the file's effect on citation rates remains statistically indistinguishable from noise.
Translation
What happens to your search?
The prompt a person types is almost never the query AI runs. Only 0.17% reach search unchanged — the other 33% get rewritten completely, with years, superlatives, and brand names you never typed quietly added in.
Share of rewritten queries that gained each word. "brands", "best", and "official" reveal AI's commercial slant.
How we measure this
Every number here is live. We aggregate public AI-search telemetry — 18.6M citations across 4,897 GA4 properties, crawler logs spanning 37.9K domains, and cross-model responses from 8 AI models — then refresh it daily. No surveys, no panels, no estimates.
The research behind this page
Updated 1d agoYou've seen the whole map. Now find your brand on it.
Track how you appear across ChatGPT, Claude, Gemini, Perplexity, and Google AIOs — your citations, rankings, and movement, updated daily.
Start free14-day free trial · Cancel anytime