How to Tell If AI Assistants Cite Your Business

Rankings you can check. Traffic you can check. But when a prospect asks ChatGPT "who's a good commercial HVAC contractor in Phoenix" and gets a three-name answer, you have no line in your analytics that tells you whether you were one of the three. AI assistants don't send referral headers the way search engines do, and most don't link out consistently. So a lot of business owners are flying blind on a channel that's quietly steering decisions.

The good news: you can measure this. Not perfectly, but well enough to know where you stand and whether your content work is moving the needle. Below is a method you can run yourself in an afternoon, plus how to read the results.

Why your normal analytics won't show this

Traditional SEO measurement leans on referral data. Google Search Console shows impressions and clicks. Your analytics shows which page a visitor landed on and where they came from. AI assistants break that chain in two ways.

First, many answers don't produce a click at all. The assistant summarizes, the user reads it, the decision gets made, and nobody visits your site. That's a mention with zero measurable traffic. Second, when assistants do link, the referral often shows up as direct traffic or gets grouped under a generic domain, so it's easy to miss or misattribute.

This means you have to measure the source directly — by asking the assistants the questions your prospects ask and recording what they say. It's manual, but it's the only method that reflects reality right now.

How do I check if ChatGPT or Perplexity mentions my business?

Run the actual buyer questions through each assistant yourself and log whether your business is named, linked, or absent. There is no reliable automated dashboard that covers every assistant accurately, so a structured manual test is the honest baseline. Here's the sequence:

1. Build a question list. Write 15 to 25 questions a real prospect would type. Mix them: category questions ("best family dentist in Naperville"), problem questions ("my furnace is short cycling, who do I call"), comparison questions ("X company vs Y company"), and your own brand name ("is [your company] any good").
2. Pick your assistants. At minimum test ChatGPT, Perplexity, and Google's AI answers. Add Gemini and Microsoft Copilot if your audience skews toward those. Perplexity is the most useful early signal because it links sources visibly.
3. Use a clean session. Log out or use a private window so your own history doesn't bias the answer. Turn on web browsing / search mode where the assistant offers it — that's what mirrors a real user getting current information.
4. Ask each question in each assistant. Paste the exact question. Don't lead the model or add your company name unless the question is a brand-check question.
5. Record three things per answer: Were you mentioned by name? Were you linked? Were competitors named instead? Copy the relevant sentence so you can see how you were described.

That's the whole method. It's tedious, but it produces something you can compare month to month.

A simple scoring sheet

Log results in a spreadsheet so "are we showing up in AI" becomes a number instead of a vibe. Here's a format that works:

Score each cell: Linked = 2, Mentioned = 1, Absent = 0. Add up the columns and the whole grid. Now you have a baseline total — say 18 out of a possible 90 — and per-assistant totals that tell you where you're weakest. Re-run the same list in 60 to 90 days and watch the number, not any single answer. Individual answers shift constantly; the trend across 20 questions is what matters.

What to watch beyond the yes/no

The copied sentences are as valuable as the score. Read how you're described. If the assistant says "a local contractor" for a competitor but "a well-reviewed company known for same-day service" for you, that specificity came from somewhere — usually your own content and your reviews. If you're described inaccurately, that's a content gap you can fix. If a competitor keeps getting named for a service you also offer, look at what they've published that you haven't.

Why am I not being cited when I rank well on Google?

Ranking and being cited draw on overlapping but different signals, so a page can rank page-one and still never get pulled into an AI answer. Assistants tend to favor content that answers a specific question cleanly, states facts in plain sentences, and can be quoted without surrounding fluff. A page that ranks on brand strength or backlinks but buries the answer in marketing language often gets skipped by the model even when a human searcher finds it.

Three common reasons a well-ranking business goes uncited:

• The answer isn't extractable. The information exists on your page but it's wrapped in persuasion. Assistants prefer "A furnace short cycles most often because of a clogged filter, an oversized unit, or a faulty flame sensor" over "When it comes to furnace problems, our experienced team is here to help."
• You have no third-party corroboration. Assistants cross-check. If your claims appear only on your own site and nowhere else — no directory listings, no reviews mentioning specifics, no mentions on other pages — the model has less confidence naming you.
• Your content doesn't match the question shape. You've written about "HVAC maintenance services" but nobody asks that. They ask "how often should I service my rooftop unit." The question-shaped page gets cited; the service-page-shaped one doesn't.

Turning the measurement into action

Once you've run the baseline, sort your spreadsheet by the zero-score rows. Those are questions where you're invisible and a competitor or a directory is winning. For each cluster of zeros, ask one question: does a page on my site answer this directly, in plain language, in the first two sentences?

A worked example. Suppose a dental practice in Boise runs the test and finds it's absent for "how much does Invisalign cost in Boise," "is Invisalign worth it," and "Invisalign vs braces for adults." All three are zeros across every assistant. The practice has an Invisalign service page, but it opens with "Our caring team offers modern orthodontic solutions." The fix isn't a new marketing campaign. It's writing three question-shaped pages that lead with the answer: a genuine price range with the factors that move it, an honest "worth it" assessment with tradeoffs, and a straight comparison table. Re-test in 90 days. If those pages get indexed and referenced elsewhere, some of those zeros start becoming ones and twos.

You won't win every question, and you shouldn't try. Focus on the 8 to 10 questions closest to a buying decision in your market. Being cited for "emergency AC repair open now" matters more than being cited for "history of air conditioning."

A note on tools that claim to track this

Several products now offer AI visibility tracking. Some are genuinely useful for scaling the manual method across hundreds of prompts. Just understand what they're doing: running queries and parsing answers, the same thing you did by hand. They can't see inside the models, and results vary by the day and the phrasing. Treat any single-tool number as one reading, not gospel. If you'd rather not run and re-run these tests yourself, this is the kind of measurement a content program like ClearPath Content builds into its reporting, alongside the pages that actually move the score.

The practical takeaway

You don't need a new platform to know whether AI assistants cite you. Write down the 20 questions your best prospects ask, run them through three assistants in a clean session, score mentioned/linked/absent, and total it. Do it again in a quarter. That single number — and the description sentences underneath it — tells you more about your AI visibility than any dashboard, and it points directly at the pages you need to write next.