How to check if AI recommends your business

Ask the assistants the questions your customers actually ask, from where your customers are, and count how often you are named. Do it across all four major assistants and more than one question, because they disagree with each other and with themselves. Most checks fail on the first step: they ask a generic question, and get a meaningless answer.

Doing it by hand

You can do this yourself in about twenty minutes. It is worth doing once even if you later automate it, because you will learn what your customers are actually being told.

1. Write down three questions a real customer would type. Not "best [your category]" — the question they would actually put to an assistant. One broad ("what's the best barn wedding venue in the Cotswolds?"), one with a real requirement ("which venues near Cirencester take 150 guests with rooms on site?"), and one comparison that still asks who.
2. Ask each of them, separately, to ChatGPT, Perplexity, Gemini and Claude. Use a fresh conversation every time — a chat that already knows you are researching your own business will tell you what you want to hear.
3. Note, for every answer, whether you are named, and who is named instead. Twelve answers in total.
4. Count. Named in 9 or more of 12 is strong. Under half means a gap worth closing. The names that keep appearing instead of yours are your real competition in AI answers — often not who you would expect.

The three mistakes that make most checks meaningless

1. Asking a generic question

This is the big one. Across 39 UK businesses, the median was named in 75% of answers to a question shaped around what they actually do. Ask the generic category question instead and a business can drop to nothing — we watched a Guildford live-music venue score 0 of 3 on "best wedding venue in London" and 12 of 12 on "best live music venues in Guildford". If your check asks a question your customers do not ask, its answer tells you nothing.

2. Asking one assistant

The four assistants read the web differently and frequently give four different shortlists. A business can be named every time by Perplexity and never by Claude. One assistant is a sample of one.

3. Asking once

The same assistant asked the same question disagrees with itself about 4.7% of the time in our measurements. Across several questions that is enough to make a single run misleading. A pattern across questions and assistants is real; a single answer is a coin that landed one way.

Where you ask from matters too

Assistants answer for the person asking. Ask from the wrong place and a UK shoulder surgeon can lose to American ones, or a Cotswolds venue to a Napa winery. If you check from a VPN, or the assistant does not know where you are, frame the question with the location: "I'm in Cheltenham — which…".

Doing it automatically

That is what Ask All Four does: it reads your website, drafts the three questions a real customer would ask, puts them to all four assistants from your location, and shows you who was named in each answer and who was named instead. The check is free and takes about twenty seconds.

Questions people also ask

Can I just ask ChatGPT whether it recommends my business?

You can, but the answer is unreliable — asked about you by name, an assistant will usually find something nice to say. The test is whether it names you when it does not know you are asking.

How many questions should I check?

At least three, of different kinds: a broad one, one with a specific requirement, and a comparison. One question samples one way of asking.

What if every assistant names me?

Good — and worth watching, because answers move as models update. The first sign of losing it is usually one assistant dropping you, long before your enquiries notice.

Is a free AI visibility check accurate?

It is accurate for the questions it asks. Check what those questions were. A free check that asks a generic category question can make a well-recommended business look invisible.