AI crawlers and bots
Check your own AI-to-Google visit ratio against the 0.62% study
By Nora Blake · Oct 29, 2026 · 8 min read
Three studies published in the week before Oct 22, 2026 argue that AI assistants send a small fraction of Google's traffic, as reported by Search Engine Roundtable and MarTech. If you ship small sites, the useful question is not whether the studies are right on average. It is what your own ratio is, and how much of it you can trust.
This post gives you the formula, a worked example with made-up numbers, and the three places where the count stops being reliable.
What the studies reported
Marie Haynes looked at 12 months of GA4 data from 100 websites of varying sizes, with more than 145 million sessions. In her words, sessions from ChatGPT "amounted to just 0.62% of the volume driven by Google Organic." With all other LLMs included, the figure rose to 0.73%. MarTech points out that this is a share of Google organic traffic, not of total site traffic.
Two other studies landed the same week. SEO Works reported that organic search delivered 39 times more traffic than AI assistants. Kevin Indig looked at 12 sites and found that AI-driven direct traffic to long-tail pages was not yet detectable.
Source: Search Engine Journal, GEO Is A Distraction: Here's Where The Real Opportunity Is, Oct 2026. Retrieved Oct 26, 2026.
The limits are in the coverage itself. Haynes describes her dataset as small. MarTech notes that the study gives no exact reporting dates, no industry mix and no definition of how conversions were counted. Search Engine Journal's version of her column does not explain how sessions were classified as ChatGPT or as another LLM. None of that makes the 0.62% wrong. It means you are comparing your site to an average whose makeup you cannot see.
The ratio, and how to compute it
The formula is one line:
AI-to-Google ratio = visits from AI assistants / visits from Google search x 100
Haynes reports two versions, ChatGPT alone and ChatGPT plus other LLMs. Compute both. They answer different questions: the first is comparable to her 0.62%, the second to her 0.73%.
The numerator. Peeko counts visits from ChatGPT, Claude, Perplexity, Copilot and Gemini from ordinary page views, by their referrer, with no middleware involved. If you count by hand in any tool that lists referrers, match these hostnames and their subdomains exactly:
- chatgpt.com
- claude.ai
- perplexity.ai
- copilot.microsoft.com
- gemini.google.com
Do not match on the bare words google or microsoft. Our post on counting visits from AI assistants by referrer has the matching rule and a script.
The denominator. You have two honest choices, and you should pick one and stay with it.
- Visits whose referrer is Google search in your analytics. Leave gemini.google.com out of this number, because it belongs in the numerator.
- Clicks from Google Search Console. Google counts "any click that sends the user to a page outside of Google Search, Discover, or News," and a second click on the same link after returning counts once. Peeko connects Search Console read only and keeps its rows for sixteen months. Google publishes about three days late and revises its three most recent days, so end your window at least four days ago.
The two denominators will not match. Clicks are counted by Google, visits are counted by your page, and ad blockers and slow loads sit between them. Compare ratios only against ratios built the same way.
The window. Use 12 months if you have them, since that is what Haynes used. A newer site can use what it has, but a shorter window is a noisier one.
A worked example with made-up numbers
These numbers are an illustration, not a measurement. Say that over 12 months your site had 38 visits from the five assistants and 5,200 visits from Google search.
node -e "const ai = 38, google = 5200; console.log((ai / google * 100).toFixed(2) + '%')"
That prints 0.73%. The same numbers say Google sent about 137 times as many visits as the assistants. Your real figures will differ, and the point of running it on your own counts is to see that gap for your site instead of reading it for someone else's.
Where the count stops
A referrer count is a floor. It is a count of the AI visits that carried a referrer, not of all AI visits. Three things sit outside it.
Visits that arrive as Direct. Ahrefs published a study of 3,000 anonymized sites on Feb 6, 2025. It found that 0.17% of visitors came from AI chatbots and that 63% of sites had at least one AI visitor. Its own main caveat was that "this data relies on the referral source being trackable in analytics," and it concluded that AI traffic is likely being vastly underestimated. That study is more than a year old, but the mechanism has not changed. A study built on referrers has the same blind spot, and the coverage of the October studies does not say how they handled it.
The Oct 9 to Oct 14, 2026 window. From Oct 9, many ChatGPT links reached sites with no utm_source tag and often no referrer. Tools that count ChatGPT by referrer, ours included, filed those visits under Direct. Adobe's sample of about 500 pages showed ChatGPT visits down by about 31,000 a day and Direct up by about 26,000 a day, over only four sampled days. Adobe's Cedric Huesler later said ChatGPT was tagging links again and called the change a bug. If your window covers those dates, compute the ratio with and without them and write the affected days down as an undercount. Our post on the October 2026 shift of ChatGPT visits to Direct shows the three daily series to compare.
Answer fetches. An assistant that reads your page while replying to somebody leaves a server hit, not a visit. That hit says your page was read. The question itself never reaches your server, so nobody can report it. Crawler hits take a separate server side path, because crawlers do not run JavaScript. Our @usepeeko/bot-track middleware handles that path, and those hits never belong in the numerator. Keep them next to the ratio as a second number. Our post on AI visibility scores versus server side counts explains why the two measurements disagree.
Small counts are noisy
If your numerator is in the dozens, one month tells you very little. In the example above, a swing between 32 and 44 visits, which is roughly what random variation alone produces at that size, moves the ratio from 0.62% to 0.85%. That range straddles the study's figures by itself.
So do not read a trend from one month. Compute the ratio over the longest window you have, then recheck it monthly and watch whether the numerator moves by more than a handful of visits.
What the ratio cannot tell you
A visit ratio measures volume. It says nothing about value, and the study's value numbers are the least defined part of it. Haynes reports that ChatGPT visitors converted at 13.8% against 3.02% for Google organic, and that only 55 of the 100 sites tracked conversions in GA4. Google organic still drove 27 times as many conversions as all LLMs combined. The coverage does not say what counted as a conversion.
Source: Search Engine Journal, GEO Is A Distraction: Here's Where The Real Opportunity Is, Oct 2026. Retrieved Oct 26, 2026.
This is where your own data does something a study cannot. Peeko can show your site's own ratio and your own AI-referred signups. It cannot tell you whether the averages in any study apply to your site, and neither can the study.
A goal in Peeko is a definition rather than a counter, so a signup goal you write today counts over the months before you wrote it. That lets you compare AI-referred signups with Google-referred signups on history you already have, without waiting for new data. The no-cookie way to follow a visitor from an AI referral to a signup is in following an AI referral to signup without cookies.
How to read your number
Treat the result as a position relative to one study, not as a verdict.
- Near or below 0.6% of Google visits: your site looks like the study's average. AI assistants are not where your visits come from today, and your time is better spent where the visits are.
- Several times higher: something about your site draws AI answers, or your Google traffic is small. Check the denominator before celebrating.
- Different by a wide margin once you remove Oct 9 to 14: the referrer gap is large enough that you should look at your Direct visits and your answer fetches before you conclude anything.
For scale only, the Ahrefs study found the top sites picking up roughly 6% of their total traffic from AI, and one site at 18%. That is a share of total traffic, not of Google organic, so do not set it beside 0.62%. It does show that averages hide a wide spread.
Run it this week
- Pull 12 months of visits from the five assistants and from Google search, ending at least four days ago.
- Compute both ratios, ChatGPT alone and all five.
- Repeat with Oct 9 to 14, 2026 removed.
- Write down the number and the date, and recompute next month.
If you work with an AI coding agent, it can pull the same counts for you. npx @usepeeko/cli speaks the same MCP endpoint an agent does, so a command and a tool call return the same numbers. Our post on connecting Claude Code or Cursor to your site analytics shows the setup.
The October studies make an argument about averages. Your ratio is a measurement of one site, with a known floor. Together they tell you whether to spend an afternoon on AI visibility or on the channel that already sends the visits.