← All articles
Product

ChatGPT Reads Small-Business Sites. Here Is What We Measured on Three of Our Own

By Iuliia Gorshkova · Co-founder · CEO & CTO
Share
Yes, ChatGPT reads small-business websites when it has a reason to. Between 19 August and 26 October 2026 we logged every AI bot on three small sites our own team runs (a guide to life and business in Spain, a robotics-and-AI learning hub and a course company), all publishing articles through Laspi Pro's autoblog. ChatGPT opened their pages 855 times while answering real people's questions, OpenAI's search index fetched 932 pages, Perplexity 332. The pages that got read were articles answering one concrete question. Our own thin directory pages on laspi.pro were opened twice in the same period.

Every few days someone asks us the same thing: does ChatGPT even look at a small site, or does it only quote the big directories? We had the logs to answer it, so we read them. Three small sites our team runs publish articles through the autoblog; the WordPress plugin on each of them counts every request from known AI bots. This is what ten weeks of those counts say.

What we measured and how

The plugin records, once a day per bot and per site, how many requests came from the user agents the AI companies document: ChatGPT-User (the browser ChatGPT uses while it is answering a person), OAI-SearchBot (the index behind ChatGPT search), GPTBot (OpenAI's training crawler), PerplexityBot, ClaudeBot, Meta-ExternalAgent and a few others. It stores a daily count per bot, not visits, sessions or IP addresses, so every number below is a count of requests, not of people. Two of the three sites stopped reporting on 4 October, so their series end there; the third runs to 26 October.

The three sites: a guide to life and business in Spain (filledblanks.eu), a robotics-and-AI learning hub (i.partenit.io) and a course company (gradebuilder.tech). All three belong to people on our team, which is why we can publish their logs. None of them is a client.

The numbers, 19 August to 26 October 2026

  • ChatGPT-User, live reads while answering someone: 855 in total (466 on the learning hub, 247 on the Spain guide, 142 on the course company).
  • OAI-SearchBot, the search index behind ChatGPT: 932.
  • PerplexityBot: 332. GPTBot, OpenAI's training crawler: 286.
  • Meta-ExternalAgent: 3,688. ClaudeBot: 39,065, almost all of it a crawl storm on one site in late August and September.

The live reads grew week by week on the site that published most consistently. The learning hub went from 40 pages opened by ChatGPT in the week of 17 August to 64 at the end of August, 79 in mid-September and 97 in the week of 28 September. The Spain guide held 36 to 50 weekly reads through mid-September, then fell to 9 the week it paused publishing, and stayed under 10 after that.

What got read, and what did not

The pages ChatGPT opened were articles that answer one concrete question a person might type into it: how a particular permit works, what a course covers, how to do one specific thing. Home pages were rarely opened. The contrast on our own side is sharp: in the same two months the business directory on laspi.pro, which describes each business in a paragraph, was opened by ChatGPT twice. Nineteen pages of description got two reads; a few dozen articles got hundreds.

Crawling and reading are different things. ClaudeBot made 39,065 requests to the course company's site and that changed nothing for the business: a crawler collecting pages is not an assistant quoting them. The 855 ChatGPT-User requests happened while a person was sitting in a chat asking something the page could answer.

What this means for a small business

  • AI assistants do not need your home page. They need a page that answers the question their user just asked, with the facts stated plainly and dated.
  • Regular publishing shows up in the logs within weeks. The consistent site rose for seven weeks running; the site that paused dropped within two weeks.
  • Being read is not yet being recommended. When we ask ChatGPT about these sites by name, it knows them and quotes them. When we ask it an open question in their field, it recommends large directories first. Reads come before recommendations, and recommendations follow reputation: reviews, mentions, links from places people already trust.
  • Bot counts are not customers. A read means the page was used while answering someone. Whether that person then visited or bought is not in a bot log, and we are not going to pretend it is.

How to see this on your own site

If the site runs WordPress, the Laspi Pro plugin keeps the same per-bot log and shows it in the autoblog report, whether the articles are written by you or through the autoblog from your own facts. On any other platform the hosting logs contain the same user agents: filter for ChatGPT-User and OAI-SearchBot and count by day. OpenAI and Perplexity publish the exact agent names and IP ranges, linked below. How the autoblog itself works is described on the content-for-sites page.

Frequently asked questions

Does ChatGPT actually visit small websites?
Yes. In our ten-week log ChatGPT's live agent opened 855 pages on three small sites while answering people, and its search index fetched 932 more. It opens pages that answer a specific question and rarely opens home pages.
How do I tell which bot is ChatGPT?
Three user agents matter. ChatGPT-User is the live fetch while a person is chatting. OAI-SearchBot builds the index behind ChatGPT search. GPTBot collects training data. OpenAI documents all three, including the IP ranges they use.
Did these reads bring customers?
A bot log cannot say. A read means the page was used while answering someone; whether that person then visited or bought is not in these numbers.
Why did one site fall from 44 to 9 weekly reads?
It paused publishing in September. The other site kept its weekly rhythm and its reads kept rising. Nothing else changed on either site in that period.
Are these your clients' sites?
No. All three are projects run by people on our team, which is why we can publish their logs. The plugin reports the same data to any site owner privately.

Sources

  1. OpenAI, 2026 — ChatGPT-User, OAI-SearchBot and GPTBot are OpenAI's documented crawlers; the page lists their purpose and IP ranges.
  2. Perplexity, 2026 — PerplexityBot and Perplexity-User are Perplexity's documented crawlers.

Read next

ChatGPT Reads Small-Business Sites: 10 Weeks of Our Own Logs · Laspi Pro