Free ways to check your AI visibility
Checking your AI visibility for free takes five methods used together: ask the four main assistants the buying questions your customers ask, log every answer in a spreadsheet so repeated runs become a trend, check your robots.txt for blocked AI crawlers, find the roundups and directories that rank for your category and see whether you are on them, and look for referrals from assistant domains in your analytics.
What a free check can and cannot tell you
Everything useful in this field comes from asking assistants questions and recording the answers, and nothing stops you doing that yourself. A free check can tell you whether you are named, how often, who is named instead, what sources the answers lean on, and whether anything is technically blocking the crawlers that feed those answers.
What it cannot tell you is what any real person was told. Those conversations are private between the user and the provider, and no method, free or paid, has access to them. Every approach below produces a directional reading from questions you ask, which is a real measurement as long as you are honest about what it is.
Method one: ask the four main assistants yourself
Open a temporary or signed-out chat in ChatGPT, Claude, Gemini and Perplexity. Copilot is worth adding if your customers are business users on Microsoft 365. Ask each the same questions, worded identically, in a fresh conversation every time.
Write the questions as a customer would, with your business name nowhere in them. "Who are the best joiners in Mansfield, England?" is a test. "Is my firm a good joiner?" is not. Always include the country — Mansfield, Boston, Perth and Hamilton all exist in more than one, and an unqualified town name regularly produces an answer about the wrong continent.
Run each question five times. Models are non-deterministic and the shortlist moves between runs, so what you are collecting is a rate, not a verdict. The full method, including the pitfalls that quietly invalidate a test, is set out in how to check if ChatGPT recommends your business.
Method two: a spreadsheet that turns answers into a trend
One row per run, in any spreadsheet you already have. Columns: date, assistant, question, named yes or no, your position in the list, other businesses named in order, sources cited, and notes.
Add two summary cells and the sheet starts earning its keep: your named rate for the month (runs that named you, divided by total runs) and a tally of the competitor names. Repeat monthly on the same questions and you have a trend line, which is the only thing that lets you tell a real change from normal variation.
Resist the urge to improve the wording of your questions each month. A better question resets your history. Keep the originals running and add new ones alongside them if you want to test a different angle.
One habit is worth building in from the start: paste the first two sentences of each answer into the notes column rather than only ticking a box. It takes seconds and it captures how you were described, which is frequently where the useful finding is. A gift retailer who discovers they are consistently described as a card shop has learned more from the wording than from the named rate.
Method three: check your robots.txt for crawler blocks
Visit yourdomain.com/robots.txt in a browser. It costs nothing and takes a minute, and we regularly find sites quietly blocking the crawlers that assistants depend on.
Look for Disallow rules against user agents such as GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended and Applebot-Extended. Some of these fetch pages to answer a live question, others gather training data, and the distinction matters: blocking the first kind can keep you out of answers today. A blanket block added by a plugin or a previous developer is a common and entirely fixable cause of absence, and the full breakdown is in AI crawlers and your robots.txt.
While you are there, check that your main service pages return normally to a plain fetch rather than requiring JavaScript to show their text, and that no firewall rule is rejecting these agents by IP. Robots.txt is the common culprit but not the only one.
Method four: find the pages that get cited in your category
Search your core buying question in an ordinary search engine and note the roundups, directories, review platforms and local publications that rank. Then compare that list against the sources your assistant answers actually cited in method one. The overlap is usually substantial.
Go through each one and check whether you are listed, whether the listing is accurate, and how you are described. Being absent from the single page most often cited for your category is a specific, solvable problem, and it is free to find. Why these third-party pages carry so much weight is covered in reviews, directories and third-party signals.
Note the ones you can join at no cost and the ones that want payment or an editorial pitch. The free ones are the obvious first pass.
Do the same for the pages on your own site that these answers ought to be able to quote. Open your main service page and read it as a stranger would: does any single sentence state plainly what you do, where you do it, and who for? A page built around a hero image and a contact form often contains nothing extractable at all, which is a free thing to discover and a cheap thing to fix.
Method five: look for assistant referrals in your analytics
In whatever analytics you run, filter acquisition by referral source and look for chatgpt.com, perplexity.ai, copilot.microsoft.com, claude.ai, gemini.google.com and similar. Anything you find is a person who read an answer and clicked through to you.
Treat the numbers as a floor rather than a measurement. Volumes are small for most small businesses, attribution is imperfect, and clicks are the exception rather than the rule in assistant answers — the core of the zero-click problem. What the figure is good for is direction: a referral count that grows month over month, alongside a rising named rate in your spreadsheet, is two independent readings pointing the same way.
Two practical notes. Set the date range to at least ninety days, because a month of single-figure referrals looks like nothing and a quarter of them looks like a pattern. And check your server logs if you have access to them, since they show assistant crawlers fetching your pages even when no human ever clicks — evidence that you are being read, which analytics alone will never give you.
Putting it together: an hour a month
A workable free routine: once a month, open a temporary chat in each assistant, run your five questions five times each, fill in the spreadsheet, and update the named rate. Once a quarter, re-check robots.txt and the directory listings. That is roughly an hour a month, and it produces a defensible record of whether your visibility is improving, flat, or sliding while competitors gain.
Where the free approach runs out
Three things break it, and they are worth naming plainly.
- Consistency. Doing it by hand means slightly different wording, a different time of day, a different browser session. Small inconsistencies accumulate into a trend line you cannot fully trust.
- Time. Five questions, five runs, four assistants is a hundred answers to read and log. Two locations doubles it. Ten locations makes it somebody's job.
- Discipline. This is the real one. Most people run the check once, find it interesting, and never run it again — and a single snapshot answers almost nothing, because the whole value is in the comparison with last month.
If you can hold the routine, the free method is genuinely sufficient for a single-location business and you should not pay for anything. If month three is when it quietly stops happening, that is the honest argument for automating it — whether with your own script against the assistant APIs, or with a product. If you would rather see a baseline before deciding, the free check we run at Nusan does the first pass without a signup, and the spreadsheet above will still be waiting if you prefer to keep it by hand.
If the manual routine stops being realistic — more than a couple of locations, or a month where you simply do not get to it — the question becomes how to judge the products that do it for you, which is covered in what to look for in an AI visibility tracking tool.
Common questions
How long does a free AI visibility check take?
Budget about ninety minutes for the first pass, most of which is setting up the spreadsheet and writing your questions. After that, a monthly repeat takes thirty to forty-five minutes for a single location and five or six questions across two or three assistants.
Do I need a paid ChatGPT account to run the checks?
No, and the free tier is arguably better for this. You want an answer that owes nothing to your account memory or custom instructions, so a signed-out or temporary chat is the cleaner test. The same applies to Claude, Gemini, Perplexity and Copilot.
Can I see AI traffic in Google Analytics?
Partially. Referrals from chatgpt.com, perplexity.ai, copilot.microsoft.com and similar domains appear when somebody clicks a link in an answer. What you cannot see is the far more common case where the assistant answered and nobody clicked, which is exactly the traffic you are trying to understand.
Is a free check as good as a paid tool?
The evidence is the same kind, and for one location with a handful of questions it is good enough. What you give up is consistency and scale: identical questions run the same way every month, several markets at once, and a record that survives your own forgetfulness in month three.
See whether the assistants name you
Three real customer questions for your trade and town, no signup, about a minute.
Run my free check