Can I submit my business to ChatGPT?

No. There is no submission process, no listing form, and no paid placement. Assistants draw on their training data and, when browsing, on live web results. The only lever you have is what those sources say about you.

How long does it take to change?

Changes to your own site can be picked up within days by assistants that browse. Changes that depend on training data — where the model recalls you without searching — move far more slowly, over months. Expect the browsing-based improvement first.

Does paying for ads help?

Not directly. Ads do not enter training data or grounded search results as recommendations. Indirectly, advertising that produces reviews, press coverage or directory listings can help, because those are readable sources.

Is this the same as SEO?

The inputs overlap but the target differs. SEO competes for a position in a list of links. This competes to be named inside a sentence. A page can rank well and still never be quoted.

How to get your business recommended by ChatGPT (UK guide, 2026)

To get named by ChatGPT, your business needs to be described consistently across sources the model can read — your own site, directories, review platforms and local press — in plain language that states what you do and where you do it. There is no submission form and no way to pay for placement. What you can do is make yourself easy to describe accurately.

Why you cannot simply submit your business

The first thing to understand is that there is no door to knock on. Assistants do not have a business directory you can join, an application to fill in, or an advertising product that buys you a mention in an answer. When someone asks ChatGPT for a joiner in Mansfield, the model is doing one of two things: recalling what it absorbed during training, or searching the live web and summarising what it finds.

Both routes have the same root. They depend on what already exists, in writing, about your business. So the work is not persuasion — it is making yourself unambiguous and easy to describe.

What actually appears to influence being named

In rough order of leverage, based on what shows up in the citations when we scan:

1. Third-party corroboration, described consistently

This is the big one, and it is the one most businesses underinvest in. If your business appears on Checkatrade, a trade association register, a local news piece and two directory listings, and all of them describe you the same way — same trading name, same town, same service — then a model has several independent sources agreeing. That is what confidence looks like to a system trained on text.

Inconsistency is what kills it. "T.C. Joinery" on one site, "TC Joinery Ltd" on another and "Tom Carter Joinery" on a third look like three weak signals rather than one strong one. Pick the name you trade under and use it identically everywhere.

2. Plain, extractable statements on your own site

Assistants quote sentences that make a claim clearly. Compare:

"Your trusted partner for bespoke fenestration solutions across the region."

"We fit and repair sash windows across Nottinghamshire, including Mansfield, Worksop and Newark. Most jobs are completed in a day."

The second is quotable. It names the service, the area and a concrete detail. The first says nothing a model could repeat to a customer. This costs nothing to fix and is the single change most worth making today.

3. Crawler access

If your host or a bot-blocking plugin refuses the AI crawlers, you cannot be cited from your own site at all — no amount of good content will help. Check that your robots.txt does not disallow GPTBot, ClaudeBot, PerplexityBot or Google-Extended, and that your firewall or CDN is not blocking them. This is a five-minute check that occasionally explains everything.

4. Structured data

LocalBusiness schema with an accurate name, address, phone number, service area and opening hours gives machines an unambiguous record. It is not magic and it will not on its own get you named, but it removes doubt about the facts.

5. The pages people actually ask for

Assistants lean heavily on comparison and buying-guide content when answering discovery questions. "How to choose a joiner", "what should a fitted wardrobe cost in Nottinghamshire", "sash window repair versus replacement" — these get cited because they answer the question the user actually asked. If a competitor has written that page and you have not, their page is the one being read.

What appears to make no difference

Keyword density. Repeating "best joiner Mansfield" does not help and reads badly to a human, which is the audience that matters when a model is summarising.
Paid ads. Not directly. They do not enter the sources an answer is built from.
Meta keywords. Still dead, still occasionally sold to people.
Volume of thin pages. Fifty near-identical town pages give a model fifty weak signals and give Google a reason to distrust the site.

An honest note on measurement

Nobody can show you a transcript of what an assistant told your customer. Those conversations are private, personalised by that user's history, and different every time. What can be measured is how a model behaves across a consistent set of questions, asked repeatedly. That is a directional index — genuinely useful for spotting whether you are absent, whether you are improving, and who is named instead — but it is not a transcript, and any tool claiming otherwise is overselling.

Where to start this week

Find out whether you are named at all today. You cannot improve a number you have never measured.
Fix your trading name so it is identical everywhere it appears.
Rewrite your homepage opening so it states plainly what you do and where.
Check the AI crawlers are not blocked.
Write the one buying-guide page your customers keep asking you about in person.

Then measure again in a month. The answers vary run to run, so a single check tells you very little — the trend across a dozen questions is the signal.