Free tools / AI Share of Voice
Who gets named instead
Who does AI name instead of you?
free · no signupAsk ChatGPT or Gemini to recommend a business like yours and it names a few. This counts them. You get every other business that turned up across the whole check, and how many of your questions each one appeared on — beside your own count, on the same axis.
One box, your website address, plain English back. No account and no email.
Count who gets named
up to 10 questions · ChatGPT and Gemini · asked once each
1
We read your website to work out what you sell and where you are.
2
We write up to ten questions a customer of yours would type — and show them to you before we go.
3
We ask ChatGPT and Gemini, then count every other business named in the answers.
This check runs on our servers, and it has to. A web browser cannot ask ChatGPT a question on your behalf. The four generators linked further down genuinely do run in your browser, and only those say so.
It is the same check as the visibility checker, and it costs the same. If you have already run the visibility checker on this address recently, our server answers this page from that measurement instead of asking the assistants again — and the result says so, with the date it was actually measured.
Northlight
example.com EXAMPLE DATA6
Questions counted
at least one AI answered
2
You were named on
cited or named from memory
3
Other businesses
named across the six
4
Not measured
excluded from every count
Example figures on a fictional business. One axis, one denominator: the six questions at least one assistant answered. The four that nothing answered are in neither half of any fraction — they are not a zero for anybody, including the competitors. No percentage is printed, because several businesses can be named on one question and these counts do not divide a whole.
What the count is
What gets counted, and what deliberately does not
A share-of-voice number is easy to fake and hard to read. Here is exactly what this one is, before you run it — so nothing on the result needs a footnote to be honest.
Counted per question, never per assistant
Each question's list of businesses is a union across every assistant that answered it. So a count here means "named on this question by at least one assistant" — and it can never be turned into "Gemini named them". That attribution is not in the data, so it is not offered.
measuredA count with its denominator, never a percentage
Every line reads named on 4 of 6 measured questions. There is no percentage anywhere on the result, because several businesses can be named on one question — these counts do not divide a whole and cannot add up to a hundred.
measuredThe denominator is questions that were answered
A question no assistant answered is left out of both halves of the fraction and listed separately with its reason. It is not a question your competitors were absent from, and it is not one you were absent from either. Nothing was answered, so there is nothing to divide.
not measuredNamed is not the same as read
This list is built from the sites the assistants cited in the answers where your own site was not cited. A business named from an assistant's memory, with no source behind it, is a different fact from a page that was fetched — and the two are kept apart everywhere in the product, including here.
namedwhats the best crm for a small law firm
For a small law firm, the options that come up most often are:
1. competitor-one.com
matter management and trust accounting built in
2. competitor-two.com
cheapest of the three, with a client portal
3. competitor-three.com
strongest document automation
Free and paid, said plainly
This page names who wins. It does not show which questions they win.
That is the line, and it is drawn on purpose rather than by accident. You will see the gap on your own result, labelled, in the place the detail would have been.
What this free page gives you
Every other business that turned up anywhere in your check, how many of your measured questions each one appeared on, and your own count on the same axis. The questions nothing answered, listed with the reason each one failed. Which assistants ran, and which did not.
All of it is the whole-check total. It is enough to know whether you are being out-named, and by whom.
What is held back, and why you can see the gap
Which of your questions each business won. On the paid plans a competitor's row opens into the exact questions they were named on; here that row says so and stops. Same measurement, same scan — the detail is what the paid plan buys.
It is shown as a labelled gap rather than quietly left out. A gap you can see is an offer. A gap you cannot see is a bug, and it is the kind that makes the rest of the numbers unreadable.
4 of 6
competitor-one.com — named on 4 of the 6 questions at least one assistant answered
The number that would be easiest to sell is "67% share of voice". It is also the one that is not true: several businesses are named on the same question, so nothing here divides a whole, and a rate over six questions would claim a precision this check does not have.
The failure that matters
A check that did not run is never a zero
This is the way a page like this ships a lie: a business shows up with a small number, or your own row shows a big one, because a check never happened. Four rules stop it, and all four are visible on the result rather than promised here.
An unrun check is excluded, not scored
If an assistant was rate-limited, timed out, refused the question, or has no key installed, that question leaves the denominator and is listed on its own with the reason in plain words. It never becomes a zero for you, and it never becomes a zero for anybody else on the chart.
If nothing at all came back, there is no chart. You get a panel saying nothing was measured and why — not an empty list, which reads as "nobody was named".
A refusal is a refusal, not a result
There is a hard ceiling on how many questions we will pay to ask the assistants in a day, shared by everyone using these tools. When it is reached, or when a check cannot start for any other reason, this page says which reason and shows nothing about your website. No assistant was asked, so nothing was measured.
The counter above the box is read from our server, and it is the same pot every other check on this site draws from.
Working today
The rest of the free shelf
Being out-named is one finding. These are the checks and files that decide whether an assistant can read you at all — which is the half you control. The four generators run in your browser:
FAQ Schema Generator
Turn your questions and answers into a small block of text an AI can read cleanly. Paste it into your site, or hand it to whoever built it.
free · runs in your browserOrganization Schema Generator
Tell AI who you are: your real name, where you are, your official profiles. Stops it confusing you with a business that has a similar name.
free · runs in your browserrobots.txt Generator
One small file that decides which AI companies are allowed to read your site. Plenty of businesses are blocking them all by accident.
free · runs in your browserllms.txt Generator
A short summary of your site, written for AI to read. Worth an hour — though no AI company has confirmed they use it, and we will not pretend otherwise.
free · runs in your browserFour more free checks fetch your pages from our server and hand you back what an assistant gets: the AI Visibility Checker (the same scan as this page, read question by question instead of counted), the AI Crawler Checker (whether you are accidentally blocking AI from reading your site), the Render Check (whether your pages look empty to AI), and the AI Readiness Score (those in one pass).
Fair questions.
These six answers are also the version we hand to search engines. Nothing hidden, nothing extra.
- measured
- At least one assistant answered this question, so it is in the denominator.
- named
- This business turned up in the answer. Whether a page of theirs was read is a separate fact.
- missing
- The answer ran and never named you. Someone else got the recommendation.
- not measured
- The check did not return. Reported with its reason and left out of every count.
What is AI share of voice?
It is how often other businesses turn up in the AI answers to your customers' questions, next to how often you do. On this page it is a plain count over one check: how many of the questions each business appeared on, out of the questions at least one assistant actually answered.
It is not a market-share figure and it is not a percentage. Several businesses get named in the same answer, so these counts do not divide a whole.
Where does the list of competitors come from?
From the answers themselves. It is the sites the assistants cited on the questions where your own site was not cited, with your own domain removed, and with search engines, social networks, general reference sites, software review and directory and comparison sites, and government domains excluded by name.
So it is derived, not curated. A business on the list was cited alongside questions about you and has not been verified as your competitor, and the exclusion list is not exhaustive — a directory it does not yet name can still appear. The result says this on itself, in the same words.
Why can't I see which questions each competitor won?
Because that is the paid feature. This free page names who wins across the whole check; the per-question breakdown — which of your questions each business was named on — is on the Starter plan and above. Your result marks the gap where the detail would be rather than leaving it out silently.
Which AI assistants do you ask, and how many questions?
The free check asks up to ten questions across two assistants, ChatGPT and Gemini, once each — as many questions as your homepage supports, and the result says how many it asked. Repeating each question several times and reporting the spread is on the paid plans. Perplexity is on no plan today. Google's AI Overviews is on no plan: the only route to it is a paid search-data vendor, and we do not buy that data.
What happens if an assistant does not answer?
That question is taken out of the denominator and listed on its own with the reason — rate-limited, timed out, refused, or no key installed. It is never counted as a question you lost, and it is never counted as a question a competitor won.
If nothing came back at all, there is no chart and no list. You get a panel saying nothing was measured and why, because an empty list reads as "nobody was named" and that would be a finding nobody made.
Who sees my results?
You do. Your check — your business, the businesses named instead of you, and where you are losing — is not published, not used in our public data, and not used to train anything, unless you separately say yes. It will never be buried in a terms-of-use box.
One thing to be straight about: this check runs on our servers, not in your browser. It has to — a web browser cannot ask ChatGPT a question on your behalf. The four generators further up genuinely do run in your browser, and only those say so.
Knowing who wins is the first half. Knowing where is the other.
The free count names every business that turned up. The paid plans open each row into the exact questions they were named on, ask each question three times instead of once, and run on a schedule so you can watch it move.