How to run an AI visibility audit, in one sentence: ask ChatGPT, Perplexity, and Google four questions about your business, grade what comes back against a 17-point checklist, and fix the heaviest failures first. Ten minutes. Free tools, no signup anywhere. The exact prompts and every check are below, so you can run the whole thing from this page.
Why audit AI visibility at all?
You can't fix what you haven't measured, and almost nobody has measured this. Somewhere today, a potential customer asked an assistant for a recommendation in your category, and the answer either named you or handed the lead to someone else. Most owners have never read that answer. Not once.
An audit does two things. It shows you the real answers AI gives about your business, in the assistants your customers already use, with no vendor filtering them for you. And it converts "we should do something about AI search" into a short list of specific gaps in priority order. Guesses about your AI visibility don't survive contact with an actual ChatGPT answer about your business. One reading replaces a quarter of speculation.
To be clear about what this is: a guided self-check, not a scanner. No tool can fake the part that matters, which is you watching an assistant recommend someone else and reading exactly which sources it used to do it.
Run the four tests first
Fill in the brackets, run each prompt, and write down what happens. Evidence first, interpretation later.
Test 1: the recommendation test
Open a fresh ChatGPT chat with web search on and paste:
I'm looking for a [your category] in [your city]. Which specific
ones would you recommend, and why? Cite your sources.
Run the same prompt in Perplexity and check its cited sources, not just the answer. You're looking for two things: whether you're named, and which pages the assistant built its shortlist from. Those sources are the reading list for your category. Not a local business? Drop the city and run it anyway.
Test 2: the incognito search test
Open an incognito window and search:
best [your category] in [your city]
Check three things. Are you on page one, or in the map pack if you're local? If an AI Overview appears, expand its sources and see whether you're one of them. And note which roundups and directories fill the page, because those same pages feed the assistant answers from test 1.
Test 3: the brand test
Ask any assistant:
What do you know about [your business name] in [your city]?
Is it reputable? What do reviews say?
You're checking accuracy, not fishing for compliments. An accurate answer passes. An answer that's outdated, blank, or confused with another business is an entity problem, and that has specific causes and fixes.
Test 4: the comparison test
[your business name] vs other [your category]s in [your city]:
how do they compare?
Somebody is framing this comparison. If you never published one, the assistant assembles it from whoever did: competitors, affiliates, review sites. Note which claims it repeats and where each one came from.
Score yourself on three areas
Now grade what you found against 17 checks, or 13 if you're not local, since four checks only apply when there's a city involved. Answer each one yes, no, or unsure. Play it straight; unsure counts as half credit, and gaming a checklist you're grading yourself on is a strange way to spend ten minutes. Our free AI Visibility Audit generates the four prompts filled in with your details, walks the same checks, and hands you the prioritized fix list at the end. The three areas map to the phases of the Reforge Method.
Assay: what do the machines say?
Five checks, straight from your test results. ChatGPT names you. Perplexity names you. You rank top ten, or in the map pack, for your main search. AI Overviews cite you when one appears (what wins those citations is its own playbook). And a search for your brand name returns your site, profiles, and reviews rather than competitors or junk.
Reforge: are your pages built to be the answer?
Six checks on your own site. Your homepage makes it obvious what you do, who it's for, and where, without scrolling. You have one real page per service you sell, not a blurb listing all of them. Your top page carries the target phrase in title tag, meta description, URL, H1, and first sentence. Your pages answer the questions buyers actually ask, cost and timeline and is-it-worth-it, instead of just describing you. Structured data shows up in a validator. And you have fair comparison content, because AI answers lift comparison tables constantly.
Temper: do third parties back you up?
Six checks off your site, four of them local-only. A complete Google Business Profile. Ten or more reviews in the last 90 days. Reviews that name specific services, since "great company!" gives a model nothing to quote. Name, address, and phone matching everywhere. Third-party mentions beyond your own site: directories, lists, press, podcasts. And profiles that all link back to your domain. This is the corroboration layer, the evidence LLMs weigh before naming any brand.
How do you read your score?
Read the three areas separately, because they fail for different reasons.
Assay is the outcome, not the cause. A low Assay score tells you you're missing from answers; it can't tell you why. The why lives in the other two areas, so read them as diagnosis. A low Reforge score is the fastest problem you'll ever fix, because every failing check sits on pages you fully control. Rewrites and new service pages move in weeks.
A low Temper score is the slow one. Reviews, mentions, and consistent listings accrue over months, and no clever page structure substitutes for them. The mix matters too: strong Reforge with weak Temper reads to a machine like a well-built site nobody vouches for, while the reverse is a loved business whose site answers nothing. Machines hedge on both, for opposite reasons.
One more reading note: count your unsures. A pile of them usually means you've never looked at your own evidence from the outside, and half of those turn into quick yeses once you spend twenty minutes checking. The other half are real gaps hiding behind "probably fine."
What do you do with the fix list?
Don't work all 17 items. Every check carries a weight from 1 to 3, and the fix list orders your failures with the weight-3 misses on top: things like being absent from ChatGPT, a homepage that explains nothing, or a review drought. Take the top two or three and ignore the rest until next quarter. In the tool, each failed check links to the article that teaches its fix, and if you want human eyes on it, you can email your results to us for a free review. That's optional, and transparently how we meet clients.
Then make it a loop. Put a re-test on the calendar for 90 days out, and set up AI referral tracking in GA4 so you can watch movement between audits. If you'd rather work from a full quarter plan than a fix list, the 21-item AI SEO checklist is the expanded version of everything above.
Run test 1 before you close this tab. Paste the recommendation prompt into ChatGPT with your category and city, read who it names, and open the sources. Whatever you feel reading that answer is the entire case for doing the other nine minutes.