How to Audit Your AI Search Visibility in 2026

Most expert-led businesses have no idea whether AI engines recommend them, ignore them, or recommend their competitors instead. They guess. The fix is a 30-minute audit you run every quarter, the same way you'd check your finances.

Key takeaways

An AI visibility audit tests whether ChatGPT, Claude, Perplexity, and Microsoft Copilot recommend you for the questions buyers actually ask, not for your brand name.
Use 8 discovery-intent questions per engine (32 prompts total) so you catch patterns instead of one-off results.
The 4 engines weight different signals: Perplexity surfaces sources, Claude weighs author trust, ChatGPT favors clear answers, Copilot pulls from a wider open web.
On-page, three signals matter most: one clear answer per page, a real Person bio with credentials, and structured data that names you and your offer.
BakingSubs earned 162,500 Microsoft Copilot citations to date, with 112,500 of those in just the last three months, because the audit became a quarterly habit, not a one-time check.
The free AI Visibility Check automates the question-testing step in 60 seconds.

Why a one-time check is not enough

AI engines re-rank what they recommend every few weeks as their training data, retrieval indexes, and ranking signals shift. A site that ranked first in ChatGPT in March can be invisible in June, not because the site changed, but because the engine did.

That's why the audit needs to be a quarterly habit. Once a quarter you run the same questions, against the same engines, and you compare. You're looking for drift. Are you getting cited more, less, or the same? Is a new competitor showing up? Did Claude stop linking to you?

Think of it the way a coach checks their pipeline numbers each month. You don't wait until clients dry up to look at the data. Same here.

A quick example. Imagine Tomás, a leadership consultant in Austin who works with first-time tech founders. He ran an audit in January and Claude cited his blog for "how to coach a first-time founder through their first layoff." Three months later, the same question pointed to a competitor in Denver. The Denver site hadn't done anything special. They'd just published two more posts on adjacent questions in that window, which made Claude treat them as the deeper source on the topic. Tomás only caught it because he ran the audit again.

Step 1: Write the right test questions

This is where most audits fail before they start. People type their own name into ChatGPT, see a nice summary of themselves, and assume they're visible. They aren't. They're just findable when someone already knows who they are.

What you want to test is the question a buyer types when they are looking for someone like you but doesn't know you exist yet. We call these discovery-intent questions.

Bad test prompts (these only check brand recall):

"Tell me about [your name]"
"Is [your business] good?"
"What does [your company] do?"

Good test prompts (these check whether you get recommended):

"Who are the best leadership coaches for first-time tech founders?"
"I'm a new mom struggling with postpartum anxiety. Who should I work with?"
"Recommend a business coach for solo consultants making under $200k."
"Best personal trainers in Brooklyn for runners over 40."

Write 8 of these per engine. Mix three angles: the buyer's situation ("I'm a new mom and..."), the buyer's outcome ("I want to lose 20 pounds and..."), and the buyer's category ("best executive coaches for..."). For more on how buyers actually phrase these queries, the 8 buyer questions worth testing covers the patterns.

Step 2: Run the questions against all 4 engines

Don't just check ChatGPT. The 4 engines behave differently enough that being cited by one means almost nothing about the others.

Here's the short version of what each engine weights most heavily:

Engine	What it weights	What this means for you
ChatGPT	A clear, quotable answer near the top of the page	Lead each post with a 2-sentence direct answer
Claude	Author signals and source trust	A real Person schema and a credible bio matter more here
Perplexity	Recency and visible sources	Publishing cadence and fresh dates help
Microsoft Copilot	Open-web breadth, less brand-bias	Niche depth wins, even without a big name

Run all 32 prompts (8 questions, 4 engines). For each, record three things: did you get cited, who did get cited, and what page on their site got pulled.

That third one is the gold. When a competitor gets cited, click through to the exact URL the engine quoted. You'll often find a single post that answers the question better than anything you've published. That's your benchmark. The why ChatGPT recommends your competitor walkthrough goes deeper on what to do once you spot the pattern.

For the engine-specific behavior, the breakdowns of how coaches show up in ChatGPT search and how to be recommended by Perplexity are the two I'd start with.

Step 3: Score the result

For each question, score your site one of four ways:

Winning. You're cited, by name, with a link. Best outcome.
Mixed. You're mentioned but not linked, or you're one of several options without being the top pick.
Invisible. You don't appear at all. A competitor does.
Empty niche. The engine answers generically without recommending anyone specific. This is actually opportunity, not failure. It means whoever publishes the right post first will own the question.

Tally your scores. If more than half your test prompts come back Invisible or Empty, the issue is content depth, not technical SEO. You don't have enough on the topic for the engines to confidently recommend you.

This is the most important judgment call in the audit. Most people see "Invisible" and assume their site is broken. Usually it isn't. It just hasn't published enough specific content on the right buyer questions. The fix is publishing, not tweaking.

Step 4: Inspect the on-page signals

Now look at your own pages. Specifically the pages that should be ranking for the questions that came back Mixed or Invisible.

Three things matter, in this order:

1. One clear answer per page.

The first 100 words of the page should directly answer the question the title promises. Not a setup, not a personal story, not a definition of terms. The answer. AI engines pull the first quotable block, so make that block worth pulling.

Audit check: read the first paragraph of each page out loud. If it doesn't contain the answer, it's not extractable.

2. A real Person bio with credentials.

Claude in particular weights author signals heavily. A blog post written by "Admin" or "The Team" is treated as lower-trust than a post written by a named person with a bio, a photo, credentials, and a link to their About page.

Audit check: every blog post should name an author. That author should have an About page with their full bio, real photo, and a Person schema block (the hidden tag that tells AI engines this page is about a specific human, not a brand).

3. Structured data that says who you are.

You need at minimum a Person schema on your About page and an Article schema on each blog post. These are the hidden tags that tell engines this page is about a person named X with credentials Y, or this article was written by Z on date D. Without them, engines have to guess.

For a deeper look at the structural side, what ChatGPT actually looks for when recommending experts breaks down the technical signals one by one.

How BakingSubs uses this audit

I run this exact audit on BakingSubs every quarter. The site has earned 162,500 Microsoft Copilot citations to date, and 112,500 of those landed in just the last three months. That acceleration didn't happen by accident.

Each quarter I find 5 to 10 questions where BakingSubs is Mixed instead of Winning, or Invisible where a smaller competitor got cited. Then I publish or improve content specifically for those questions. Three months later, most of them flip to Winning. It's a slow, boring loop. It also compounds.

The system is called the Citation Cluster Method and the audit is the diagnostic step that tells you which cluster to build next. The full BakingSubs case study shows the exact ladder.

A faster way to run step 1

The question-testing step is the most useful part of the audit and also the most tedious. Typing 32 prompts and recording each result takes about 45 minutes if you're careful.

The AI Visibility Check automates that step. It runs 8 discovery-intent questions per engine across ChatGPT, Claude, Perplexity, and Copilot, and gives you the same 4-branch scoring (Invisible / Mixed / Winning / Empty niche). It takes about 60 seconds. You still do steps 2 through 4 yourself, but you get the data faster.

Frequently asked questions

How often should I run an AI visibility audit?

Once a quarter is the right cadence for most expert-led businesses. AI engines re-rank what they recommend every few weeks, so a monthly audit gives you too much noise and an annual audit catches problems too late. Quarterly lets you spot drift before it costs you clients.

Can I just search my brand name in ChatGPT to see if I'm visible?

No. Searching your own name only checks if the engine remembers you when prompted. What matters is whether the engine recommends you when a buyer asks for someone like you but doesn't know you exist. Those are completely different tests. The discovery-intent questions in the 8 buyer questions guide are the right starting set.

What if every test comes back Invisible?

That usually means a content depth problem, not a technical problem. The engines don't have enough on your site to confidently recommend you for those questions. The fix is to publish 5 to 10 posts that each directly answer one of your test questions, using your real name as the author. Most sites see citations start showing up within 8 to 12 weeks.

Do I need to fix every Invisible result?

No. Pick the 5 questions where the buyer is closest to hiring (someone asking "best business coach for solo consultants" is closer to buying than someone asking "what is business coaching"). Fix those first. Ignore the rest until you've won the high-intent ones.

Should I audit my competitors too?

Yes. When a competitor gets cited for one of your test questions, click through to the exact page the engine quoted. That page is your benchmark. You're not copying it. You're learning what depth and structure the engine treated as the best answer, so you can publish something better. The how to find which competitors AI recommends guide covers the workflow.

What to do next

Block 30 minutes on your calendar this week. Write your 8 discovery-intent questions. Run them through ChatGPT, Claude, Perplexity, and Copilot, and score each result Winning, Mixed, Invisible, or Empty. Save the scores in a doc with today's date so you can compare next quarter.

If you'd rather skip the manual testing, the free AI Visibility Check runs the question-testing step for you and hands back the same 4-branch scoring. Either way, the goal is the same: stop guessing whether AI engines recommend you, and start knowing.