How do I track whether ChatGPT and other AI engines are actually mentioning my business?

Google Search Console gives you a stable, queryable record of impressions and positions because Google's results for a given query are largely consistent across users at a given moment. AI engines break this assumption in two ways: there is no publisher-facing analytics product that reports what ChatGPT or Perplexity said about your business, and the answers themselves are non-deterministic, meaning the same prompt run twice in a row can surface different businesses, cite different sources, or phrase the recommendation differently, because the model is generating a fresh response each time rather than retrieving a fixed ranked list.

Personalization compounds this. A logged-in ChatGPT session with memory enabled, or a Perplexity account with search history, can skew an answer based on that user's prior activity, which means one person's experience testing a prompt is not necessarily representative of what a new customer would see. Any tracking method has to account for both of these realities rather than treating a single test as definitive.

Write 10 to 15 prompts that mirror how an actual customer would phrase a request, not how a marketer would phrase a keyword. For a roofing company in Tampa, that looks like "who is the best roofing company in Tampa for storm damage repair" rather than "roofing company Tampa," since AI engines respond to natural language intent, not keyword strings, and the phrasing genuinely changes which sources get pulled.

Test every prompt in a fresh, logged-out or private browsing session for each engine, since a logged-in session carries memory and history that will bias the answer toward or away from businesses you have previously interacted with. Run the full list on a fixed cadence, weekly is reasonable for most local businesses, and always at a similar time of day, since some engines pull from time-sensitive web content that shifts throughout the week.

Keep a single running spreadsheet rather than scattered notes, since the value of this exercise comes from spotting trends over months, not from any single week's result.

Do not expect to be named in every answer to every prompt; that is not how any business, even a category leader, performs in a non-deterministic system. A reasonable early target for a local business actively working on AI visibility is a roughly 15 percent citation rate across your top five highest-intent prompts, meaning you show up by name in about one out of every six to seven test runs on those specific prompts. Treat that as a floor to build from rather than a ceiling, and track the trend line over eight to twelve weeks rather than reacting to any single week's number.

A sudden drop across multiple engines in the same week usually points to a technical issue (a crawl block, a site outage, a schema error) rather than genuine competitive loss, so cross-reference a dip against the crawlability and indexing checks you run separately before assuming you lost ground on content or reputation alone.

While there is no citation dashboard, you can identify actual click-through traffic from AI engines in your existing analytics. In Google Analytics 4, go to Reports > Acquisition > Traffic acquisition and look at the Session source / medium dimension for referrer domains including chatgpt.com, perplexity.ai, bing.com/chat (Copilot), and gemini.google.com. Set up a custom channel group or a simple filtered exploration if these are getting buried under generic "referral" or "organic search" buckets, since some platforms are still misclassified by default.

Keep expectations calibrated here too: most AI visibility currently produces a zero-click brand impression rather than a session, meaning the model names your business and the user takes the information at face value without clicking through to your site at all. A low or flat referral number does not mean you are not being mentioned; it means the mention is not converting to a visit, which is a different problem with a different fix (usually a stronger, more specific call to action embedded in the content the AI engine is pulling from).

Manual tracking is sufficient and appropriately cheap for a single-location business monitoring a handful of prompts. Once you are managing multiple locations, multiple competitors worth benchmarking, or a prompt list that has grown past what one person can test weekly without burning hours, tools built specifically for this, such as Peec.ai, Otterly, or Profound, become worth the subscription cost. These platforms automate prompt testing across engines on a schedule, track citation share against named competitors over time, and surface trend data without manual spreadsheet upkeep.

The tradeoff is that these tools are priced and built for agencies and multi-location brands rather than a single local storefront, so evaluate whether the time saved actually exceeds the monthly cost before subscribing, particularly if you are only tracking one location.