ChatGPT and your repair shop: what gets fetched, and what a page must say

What OpenAI's documentation says each crawler does

OpenAI publishes a page on its crawlers, and for a shop it settles one question: which fetches feed an answer and which feed training. Four user agents appear there, and three of them have jobs that reach a shop's site (the fourth, OAI-AdsBot, visits only pages submitted as ads); a robots.txt file addresses them one at a time, because under RFC 9309 a group of rules written for one product token does not bind another.

That is what the documentation settles, and all of it is about access. It says nothing about how ChatGPT chooses between two shops whose pages it can read, and we make no claim about that either. What a shop can check is whether its pages can be fetched and whether they say what a driver asks.

Two routes to what ChatGPT says about a shop

A reply can come from what the model learned in training, or from pages found when ChatGPT searches the web. The first stops at a date, so it can describe a shop as it was: the old address, a mechanic who retired, the hours before you closed on Saturdays. The second reads pages as they stood when last fetched, which is the route a correction on your site can reach.

That decides what to fix first. A new service, a move, new hours, the start of hybrid work: each has to be stated on a page OAI-SearchBot is allowed to fetch, since a site that shuts it out is removed from ChatGPT's search answers. Whatever you decide about GPTBot does not change that.

What a crawler receives from a repair shop's site

An AI crawler receives the page your server sends, usually with no scripts run. Our audit of 9,483 small business websites (40 trades, 90 metros, late August to early September 2026) measured the general version of what goes missing: 7.2% served our crawler a challenge page instead of the site, 3.0% rendered their content only in JavaScript, and 2.1% kept real content, such as a price list or a service menu, in a PDF. Those figures are sample-wide; the audit publishes no cut for repair shops.

On a shop's site, these are the places a useful fact can sit out of that fetch's reach.

• The online scheduler or estimate form. Loaded by a script from your shop software's provider, it usually reaches a fetcher as an empty frame. Write the booking terms beside it: which jobs take walk-ins, how far ahead you are booked, the drop-off window.
• Coupons and specials posted as images. A picture of a brake special has no text to read. Put the offer in a sentence, with its terms and its end date.
• The service menu or price list as a PDF. A PDF is a second file that a fetcher may or may not open. Put the menu on a web page and keep the PDF as the printable copy.
• Hours in a banner image, or a map embed with no address beside it. Write both as text.
• The whole site behind bot protection. A challenge page stops OAI-SearchBot at the door whatever robots.txt says, and the setting lives in your host's or security service's bot controls.

Scripts are the quiet loss. Google describes rendering JavaScript as a separate step that comes later, in its own documentation, and nothing on OpenAI's crawler page promises a rendering step at all. The words a driver needs belong in the HTML.

How to check what OpenAI's crawlers can reach

1. Read your own /robots.txt and note whether any group of rules names OAI-SearchBot, ChatGPT-User or GPTBot. Where none does, OAI-SearchBot and GPTBot follow the * group. ChatGPT-User is different: OpenAI's page says that because its fetches are started by a user, robots.txt rules may not apply to it. robots.txt for AI crawlers explains how groups are read and has a file to copy.
2. Run the crawler access check. It reads your file the way each of nineteen tokens would, OpenAI's three among them, and reports which are allowed.
3. Fetch a service page as the search crawler would, and read the status code: curl -A "OAI-SearchBot" -I https://www.example.com/brakes. A 200 means the page came back. A 403, a 503 or a page asking the visitor to prove they are human means the fetch stopped at the door.
4. View the source of your brake page, your diagnostics page and your contact page with scripts turned off. The words still there are the words a fetcher receives.
5. Paste the address of one service page into ChatGPT and ask the question it should answer, such as "what does this shop charge to diagnose a check engine light". That fetch is ChatGPT-User's; if the reply cannot see the page, or gets the answer wrong, the page is where to look.

These checks settle access. Whether ChatGPT then names the shop for a driver's question only repeated asking can show, and checking whether your shop is named sets out the panel for it.

What a service page has to state before a question can be answered from it

A question can be answered from a page only when a sentence on the page answers it. For a repair shop that means each service page carries, near the top and in words, the facts drivers ask about that job, written so a passage lifted from the page still says whose facts they are. The repair shop overview lists the facts; the table shows them on one page.

The closest thing to a controlled study of what helps a source points the same way. Aggarwal et al., in the paper that named generative engine optimization, found that adding statistics, quotations and citations to a source raised its visibility in generated answers by up to about 40% on their benchmark. That is a finding about sources in general, not about ChatGPT or repair shops, and we cite it for the mechanism only.

Put together, a brake page built that way reads like the template below. Swap the job and it serves for diagnostics, tires or air conditioning.

Brake repair in [town]

We replace brake pads, rotors, calipers and brake lines on [makes],
including hybrids and EVs from [makes].

Price: front pads from [price], including [parts grade] pads, labor
and a brake inspection. Rotors add [price] per axle when they measure
below the maker's minimum. We quote in writing and do no extra work
without your approval.

Time: about [time] for pads. Same day when you arrive before [time].

Warranty: [months] or [distance], whichever comes first, on parts
and labor. See the warranty page.

Getting home: a shuttle within [area], or wait in [waiting area].

Book online or call [phone]. Walk-in brake checks until [time].
[Shop name], [street], [town]. Open [days and hours].