Ask an AI assistant to recommend a plumber, an accountant or a florist in your town and it will name a handful of businesses. It will not name all of them. Being left out is the normal result, not a punishment.
You do not have to guess which side of that line you are on. The test below takes about ten minutes and needs nothing except a browser.
The question that matters is not whether AI knows your business name. It is whether AI names your business when a customer describes the problem you solve.
Before you start (two minutes)
- Use two or three tools, not one. ChatGPT, Google Gemini and Microsoft Copilot read different sources and will not agree. One tool is an anecdote; three is a pattern.
- Start a fresh chat and turn off memory. In ChatGPT, use a temporary chat. Otherwise the assistant may simply repeat what you told it last week, and you will test your own memory instead of its knowledge.
- Ask the way a customer asks. Not "tell me about my company". A customer describes a problem, a place and a budget.
- Write the answers down. In three months you will want something to compare against.
The five prompts
Replace the bracketed parts with your own details. Run each one in every tool you picked.
1. The name test
"What is [business name] in [town]? What do they do?"
This tells you whether your business exists at all in the assistant's picture of the world. It is the weakest of the five tests, and the one most people stop at. Passing it proves very little: customers who already know your name were never the problem.
2. The category test
"Who does [your service] in [town]? List five options and what makes each one different."
This is the test that matters. It is how a customer who has never heard of you would ask. If the answer names five of your competitors and not you, that is your finding, and the rest of this article is about why.
3. The fact test
"What services does [business name] offer, and what do they charge?"
Read this answer for errors, not for praise. Look for prices you no longer charge, services you stopped offering, an old address, a suburb you left years ago. An answer that is confidently wrong does more damage than no answer at all, because the customer has no reason to doubt it.
4. The comparison test
"Compare [your business] with [competitor A] and [competitor B] for a [type of customer]."
This shows you what the assistant believes you are good at, and who it thinks you serve. Pay attention to the exact words it uses about you. Those words came from somewhere — usually your own homepage, sometimes a directory listing you have not looked at in years.
5. The referral test
"I am a [type of customer] in [town] and I need [problem solved]. Who should I call first, and what should I ask them?"
This is the closest thing to a real customer conversation, and the hardest test to pass. It asks the assistant to commit to one recommendation rather than list options.
Score what you got back
Give yourself one point for each of these that is true:
- At least one tool named your business in the category test.
- More than one tool named it.
- Your services were described correctly, with no invented or outdated ones.
- The contact route was right: correct website, phone or service area.
- A tool named you in the referral test.
4–5: assistants can already recommend you. Protect that by keeping your pages current, and re-test after any big change to the site.
2–3: you exist but you are not convincing. Usually the facts are on the website but buried, vague, or spread across pages that do not say who they are for.
0–1: you are invisible to these systems. This is common and it is not a sign that something is broken. It usually means nothing on the web states your facts plainly enough to be repeated.
Six reasons an assistant cannot find you
- Your website never says where you work. "Serving customers nationwide" tells a machine nothing. Name the city, the suburbs, the region — in text, on the homepage and on every service page.
- All your services live in one paragraph. A single "What we do" sentence listing eight services gives an assistant nothing specific enough to match against a question. One page per service is the fix.
- Your facts are inside pictures. A photo of a price list, a PDF menu, opening hours baked into a banner image. People can read those. Crawlers mostly cannot. Type them out as text as well.
- Your details disagree across the web. An old phone number on Facebook, a previous address in a directory, a different business name on your invoices. When sources conflict, a cautious system leaves you out rather than guessing.
- Your site turns AI crawlers away. Open yoursite.com/robots.txt in a browser. If it blocks crawlers such as GPTBot, ChatGPT-User, Google-Extended or PerplexityBot, you have opted out of exactly the thing you are testing for. Some site builders set this by default.
- Nothing points at you. A new site with no listings, no reviews and no mentions has no corroboration. Local directories and industry listings matter here more than they do for ordinary search.
What to fix first
- Check robots.txt. It takes one minute and nothing else works until it is right.
- Put your service area, hours and how to start in plain text on the homepage.
- Give your three most valuable services a page each, with prices or a price range if you can.
- Make your name, address and phone number identical everywhere they appear online.
- Add structured data so the facts are machine-readable rather than inferred. This is the part most small business sites are missing, and it is the core of xEO optimisation.
If the site is three pages of photos with the details in a PDF, fixing it can cost more than replacing it. A small site built with xEO included is often the cheaper route.
Test again in 90 days
Run the same five prompts, in the same tools, and compare against what you wrote down. Answers vary between runs even when nothing has changed, so ignore small differences in wording. Look for one thing only: named, or not named.
Being named is the beginning. The enquiry still has to be answered while the customer is paying attention, which is a different problem with a different fix — a custom AI agent can reply the moment a message arrives. There is no point building that first, though. An agent has nothing to do if nobody is asking.
Questions people ask about this test
Does ChatGPT read my website directly?
Sometimes. Some answers come from what the model learned during training, and some come from a live search it runs while answering. You cannot tell which from the outside, and you do not need to: both work better when your site states its facts in plain text.
Why do I get a different answer each time?
These systems are probabilistic, so the same prompt produces different wording each run. Run each prompt twice and look for the pattern rather than the exact sentences. If you are named in one run out of four, treat that as not named.
Can I just ask ChatGPT to add my business?
No. There is no submission form, and nothing you type into a chat changes what other people see. The only reliable route is making the facts clear on pages that these systems can read.
How long after fixing my site will the answers change?
Weeks to months, and it varies by tool. Search-backed answers can update within days of a page being recrawled. Answers that rely on training data can take much longer, and no one outside those companies can see the schedule.
Is this the same as SEO?
It overlaps but it is not the same. Ranking on Google and being named by an assistant are different outcomes with different signals behind them. Our guide to xEO explains how SEO, GEO, AIO and AEO fit together.
If you run the test and do not like the result, send us the answers you got. We will tell you which of the six reasons applies to your site and what to do about it first.