Will ChatGPT recommend your UK firm to a new client?

Ask a managing partner whether ChatGPT recommends the firm and the answer usually arrives as a budget question: who do we pay, and how much. Every page ranking for that query is published by someone selling the answer, which makes the premise hard to test.

The premise is wrong. OpenAI's publisher guidance says any public website can appear in ChatGPT search, and it publishes no way to buy a place in one. What it publishes instead is a permission, granted or refused in a single line of a text file that most UK practices have never opened. Cloudflare said more than a million of its customers had taken a one click option to block AI crawlers, and that every new domain is now asked at sign up whether to allow them. A firm whose website was built or moved in the last two years may have been opted out by a checkbox, without anyone at the practice being told it existed.

What actually decides whether ChatGPT names a firm

OpenAI documents four separate crawlers, and only one of them governs whether a firm appears in an answer. That crawler is OAI-SearchBot. OpenAI's crawler documentation says sites opted out of it will not be shown in ChatGPT search answers. The rest govern training and the checks behind advertising, and none of them changes what a prospective client sees when they ask for a recommendation.

The five minute check, on your own firm

Four lookups answer the question, and none needs a supplier or a paid tool. Run them in this order. The first two are the ones that settle it.

Why blocking the training crawler is not the same thing

OpenAI states plainly that each setting is independent of the others. A practice that refused to let its work train a model in 2024, which was a reasonable position and remains one, has not thereby removed itself from recommendations. The reverse is the expensive case: a blanket block written to keep the trainers out takes the search crawler with it.

There is one trap in the same documentation worth naming. OpenAI offers the noindex meta tag as the way to stay out of results, then adds that its crawler must be allowed to crawl a page in order to read that tag. A site that blocks the crawler and sets noindex has issued an instruction nobody can collect.

Where the block usually came from

Most blocks were set by a supplier, in a dashboard, as a default rather than a decision. Cloudflare now sorts AI activity by what the bot does on the site, separating collecting content to answer questions about it later from crawling it to train a model. That separation is recent. A switch thrown before it existed did not make the distinction, which is why a UK firm can hold a considered policy on training and still be missing from answers.

It also sits in an account. If the agency registered the domain and set up the delivery network, the setting is behind a login the practice may not have, which is the same problem as who owns the website an agency builds and worth establishing before the relationship is under strain.

The public record is not what is holding a firm back

Not one of the UK registers a professional firm is judged on blocks an OpenAI crawler. We read the robots.txt file of seven of them on 20 September 2026. ICAEW and ACCA publish long exclusion lists, none naming an OpenAI agent. The Law Society's Find a Solicitor directory excludes two paths. The Recruitment and Employment Confederation excludes application directories. The Financial Conduct Authority register allows everything and publishes four sitemaps. Companies House serves no file at all, which leaves it open. The government's own publishing platform names two AI crawlers and neither belongs to OpenAI.

That matters for where a firm looks next. The entries in those registers are readable, so what they say about a practice is available to be used. A firm missing from answers should look at its own domain before it looks anywhere else, and only then at whether its site answers the question a client would actually ask.

When not to hire anyone for this

Do not commission AI visibility work before running the four checks, because two of the four outcomes make the work unnecessary. If the analytics already show arrivals from ChatGPT, the firm is being cited and the money belongs somewhere else. If the robots file carries a block, removing one line is a ten minute job for whoever holds the hosting, not a retainer.

Hold off too where the firm cannot yet say which questions it wants to be the answer to. Every proposal in this market is priced against a volume of content, and content commissioned without that decision produces pages nobody asked for. The same caution applies to putting client information into ChatGPT, where the governing question is also a contractual one rather than a technical one. Agency Network Solutions introduces professional services firms to one vetted specialist, and a practice that has run its own checks briefs that specialist far better than one that has not. Where the agency holds the accounts, settle access before anything else moves.