
This is one site.
Four customers told the owner they found him through ChatGPT in the site's first month. That is their account, not something I measured.
What to do
- Link your important pages to each other. On your phone, check that a few taps from the home page reach the page you want read.
- Check that you don't block OAI-SearchBot. It builds ChatGPT's search results. OpenAI says sites that block it won't be shown. How to check.
- Read your visit log. Here is how to find it. Look for bot names and the deep pages they fetched.
- Count ChatGPT's visitors by the tag (explained below). In Google Analytics, look for chatgpt.com as a source. In your visit log, search for `utm_source=chatgpt.com`.
- Make the page that gets fetched repeatedly more complete. State the service, the area, the price and the questions owners ask. That is my next step, not a result.
What I looked at
The site is Ki-Sol Performance, a mobile BMW, MINI and Rolls-Royce electrical and programming specialist in the Bay Area.
I built it, and it went live on 2026-10-03.
I read its visit log (the technical name is access log). That is the list your website keeps of every visit, bots included.
Each line records the page asked for, the time and the bot's name. A bot is a program that reads pages automatically.
The log covers 2026-10-09 to 2026-11-15, 51,427 requests.
I also used Google Analytics, the free tool that counts visits, for 2026-10-03 to 2026-11-14.
A "fetch" is a bot asking for a page. A "visit" is a person.
What I found
1. ChatGPT's search bot reached every page through links, not the sitemap
OAI-SearchBot (the one that builds ChatGPT's search results) made 6,034 requests. 320 came on 2026-10-09, the first day in the log.
It read robots.txt 194 times. It never asked for the sitemap, the list of pages a site hands to search engines.
Yet it fetched all 85 pages in the sitemap at least once, and no page outside it.
For all 85 pages, the first request I can see named another page of the site that links to it. For the two pages I published during the log, this is clear.
The 11 pages linked from the header and footer took 62% of the requests.
In this log, links decided what the search bot read. The sitemap did not.
2. The bot that fetches while ChatGPT answers mostly fetched the home page, on a timer
ChatGPT-User (the one that fetches a page while someone asks ChatGPT a question) made 929 requests.
885 were the home page: about 24 a day, at every hour, at a steady pace, day and night.
That looks like a timer, not people. I don't know what drives it, and I left those 885 out of everything below.
The other 42 fetches covered 30 different pages:
- fault codes (the codes a car's computer stores): 17
- services: 13
- work-log pages: 4
- parts: 3
- models: 2
- one each: the contact page, a battery article, a symptom page
One example: `/codes/2e7c`.
I can't see what anyone typed, so the log can't confirm that any one fetch answered a question.
Ignore the steady home-page drip. The deep-page fetches are the part worth reading.
3. ChatGPT's visitors came on phones, landed on deep pages, and carried a tag in the address
The log shows 28 visits from ChatGPT, the first on 2026-10-17. Of the 28, 24 were on phones and 23 from the US.
The first page they opened: services (10), fault codes (8), the home page (5), parts (3) and the contact page (2).
None showed chatgpt.com as the page they came from. For 27 of the 28, the first request had no "came from" note at all.
So searching your log for chatgpt.com would have found nothing.
What identifies them is a UTM tag: a label added to a web address saying where the click came from.
You see it at the end of the address, like `?utm_source=chatgpt.com`. It was on 27 visits, and `utm_source=openai` on one.
Another site owner reported the same pattern.
Google Analytics also shows 28 visits from chatgpt.com, with 4 taps on the call button. A tap is not a completed call.
You can find ChatGPT's visitors in your own log, but only by the tag in the address.
I'm not promising that your pages will be fetched or recommended. This is what one site's log showed.
If you'd rather have me look at yours, request my free 30-minute consultation.
For the curious
One thin service page was fetched on eight days
`/services/dme-unlock` (unlocking a BMW's engine computer) was fetched 9 times, from 9 different addresses, on 8 separate days.
The days run from 2026-10-22 to 2026-11-13. No other page was fetched on more than two days.
Nine addresses means it wasn't one person retrying.
It was also the most common first page for ChatGPT visitors after the home page: 4 of 28 visits, against 5 for the home page.
The page is thin: 274 words, a price, one customer review. Apart from the site menu it links only to the contact and services pages.
When one page keeps getting fetched, improve it, then re-measure.
A page being read is not a page being used
Eleven symptom pages describe something a driver notices, such as a battery that drains overnight. The search bot read them 1 to 57 times each.
Five of them were read 39 to 57 times. Together they got 1 of the 42 answer-time fetches.
The DME page was read 4 times and fetched 9.
The "BMW won't start" hub, published 2026-10-24, was read twice and never fetched. It is young, so that is a hint, not a verdict.
A page the search bot reads often is not necessarily one ChatGPT picks.

Other assistants read a little and sent nobody I could see
- PerplexityBot made 16 requests and ClaudeBot 147 (on two days in October).
- Claude-User made 2, Meta's meta-externalagent 207 and meta-webindexer 158.
- Perplexity-User and meta-externalfetcher made none.
OpenAI's three bots made 7,080 of the 7,630 requests from named AI bots, 93%.
I found no visit from Perplexity, Claude, Meta AI, Gemini or Copilot in the log or in Google Analytics.
Names and jobs: which AI bots visit my website.
Only OpenAI both read the pages and sent visitors.
How this compares with what others found
Space & Story (15 small sites, 2026-11-08) found 70.8% of assistant fetches were the home page and 2.6% service pages.
Here 885 of 929 were the home page. Karbon Agency saw the same: 17 service-page fetches among 2,292.
They differ in two places. The first is the timer.
A 34-day log of one small business site read a round-the-clock fetch every 26 minutes or so as real people asking.
Space & Story read one site's steady homepage fetches as automated, and I read mine that way. Nobody can prove either.
The second is the first page visitors opened. LovedByAI reported the home page as 42% of a service business's ChatGPT visitors on the average site.
That is across 23,740 visits to small-business sites. Here 5 of 28 (18%) landed on the home page.
Theirs is an average over many sites. Mine is one site with many reference pages.
What I add: log evidence of how OpenAI's search bot reached each page of one small site, and the per-page gap between read and fetched.
I didn't find the link-following question tested in the pages I read, but my search was limited.
Limits
- In my own check on 2026-08-19, no Bay Area competitor had coding or retrofit pages. Thin competition could produce a mention whatever the pages looked like.
- The log has no entries before 2026-10-09, so it opens six days after launch. Other search engines' bots were crawling the site then, so I can't separate OpenAI's bot from them.
- Dates are UTC. I read totals only, with no visitor addresses.
- I can't see prompts. What people asked is inferred from the page fetched.
- The numbers are small: 28 visits. Google Analytics counts only visitors who accepted cookies.
- Only OpenAI's bots were matched against OpenAI's published address ranges. A few requests used another bot's name falsely, so treat the other counts as approximate.
- The site studied
- OpenAI: Overview of OpenAI crawlers
- Space & Story: How Much AI Traffic Is People? 15 Small Business Sites (2026-11-08)
- dev.to: I Logged Every AI Crawler for 34 Days (2026-08-16)
- Karbon Agency: AI crawler and referral report 2026
- LovedByAI: ChatGPT referral traffic (2026-10-20)
- Indie Hackers: Traffic from ChatGPT now brings a quarter of our sign-ups (2026-09-26)
