You cannot manage what you never see. When a prospect asks ChatGPT or Perplexity for a recommendation, the answer appears in a private conversation — no rank tracker records it, no analytics pixel fires, and no report lands in your inbox. If an AI engine is recommending your competitor by name three times a week, you will not find out from your dashboard. You find out by asking the engines yourself, on a schedule, and writing down what they say. This post walks through how to do that properly.
Why track AI citations at all?
Because AI answers are already influencing buying decisions, and they change without warning. A citation is when an answer engine names or links your business as a source or recommendation — and unlike a Google ranking, it is invisible unless you go looking. Models get updated, retrieval sources get reshuffled, and an answer that mentioned you in March may recommend someone else in April. Gartner has predicted that traditional search volume will drop roughly 25% by 2026 as users shift to AI assistants, which means a growing slice of your pipeline forms inside answers you never see. Tracking citations is how you catch wins, losses, and outright errors — including hallucinations about your brand — before they cost you customers.
How do you build a question panel?
Start with 10–20 questions a real buyer would ask before choosing a business like yours. This panel is your measurement instrument, so build it once and keep it stable. Good panels mix three question types:
- Category questions — "What is the best [service] in [city]?" or "Best [product type] for [use case]?" These reveal whether you appear in consideration sets at all.
- Comparison questions — "[You] vs [competitor]" or "Alternatives to [market leader]." These show how engines frame you against rivals.
- Branded questions — "Is [your business] legitimate?" "What does [your business] cost?" These catch factual errors and stale information.
Pull the wording from how customers actually talk — sales calls, support tickets, and the "People Also Ask" boxes in search results are all good sources. If you want a fuller method, we cover it in question research for AEO. Resist the urge to rewrite the panel every month; if the questions keep changing, you cannot tell whether the answers changed or your yardstick did.
How do you run a monthly ask-and-record process?
Pick one day a month, ask every panel question on each major engine, and record the results the same way every time. The core four to cover:
- ChatGPT (with web browsing active, since that is where citations appear)
- Perplexity (citations are numbered and explicit, making it the easiest to log)
- Google AI Overviews (search your questions on Google and note when an AI answer appears and who it links)
- Microsoft Copilot (its answers draw heavily on Bing's index, so your Bing visibility matters here)
Two rules keep the data honest. First, use fresh sessions — log out or open a clean chat so earlier conversation history does not steer the answer. Second, accept that answers are non-deterministic: the same question can produce different responses on different runs. That is exactly why consistency matters more than any single result. One snapshot is an anecdote; six months of monthly snapshots is a trend line.
What should you log for each answer?
Log five things per question, per engine: whether you were mentioned, whether you were cited with a link, which competitors were mentioned or cited, what the engine actually said about you, and a screenshot. A simple spreadsheet works — one row per question-engine pair, one column set per month.
Screenshots deserve emphasis. AI answers are ephemeral: there is no archive to consult later, no cached page to retrieve. If an engine says something wrong or wonderful about you today, the screenshot is your only evidence next quarter. Screenshots also settle internal arguments — "Perplexity used to recommend us" is a claim; a dated screenshot is a fact. Store them in a folder named by date and question so anyone can reconstruct the history.
How do you measure share of voice against competitors?
Share of voice is the percentage of your panel questions where an engine mentions you versus your competitors. If you appear in 4 of 15 category questions and your main rival appears in 9, you have a 27% presence to their 60% — a much more useful number than a binary "are we cited?" Track it per engine, because visibility often diverges: it is common to look strong on Perplexity while being absent from AI Overviews, since each engine retrieves from different sources. Watching which sources engines repeatedly cite in your category also tells you where to earn coverage — if a particular industry directory or review site keeps showing up as a source, appearing there matters. That is the logic behind digital PR for AEO.
Can branded search be a proxy for AI visibility?
Yes — imperfect, but useful. Many people who see your name in an AI answer do not click a citation; they open a new tab and Google you. So a rise in branded search impressions (visible in Google Search Console) often shadows growing AI visibility, especially when it is not explained by ads or press. Pair it with referral data: some AI clicks do arrive with identifiable referrers, and our guide to tracking AI referral traffic in GA4 shows how to segment them. Neither signal is complete on its own, which is why the manual panel remains the backbone.
What tools exist for tracking AI citations?
The tooling landscape is young. A wave of "AI visibility" trackers has appeared, and established SEO platforms are adding AI Overview reporting, but there is no equivalent yet of the mature rank-tracking ecosystem — coverage is partial, methodologies vary, and products change month to month. Treat any tool as a supplement to, not a replacement for, your own panel: the spreadsheet method costs an hour a month and you control the methodology completely. If you would rather not spend that hour, AEO Fail runs monitoring for $19/month — same discipline, done for you, with the screenshots archived.
Frequently asked questions
How often should I check AI citations?
Monthly is the sweet spot for most businesses. AI answers do shift, but rarely so fast that weekly checks reveal much beyond noise. Check more often only after a major event — a rebrand, a PR push, or discovering a hallucination you have worked to correct.
Why do I get different answers to the same question?
AI engines are probabilistic: the same prompt can retrieve different sources and phrase answers differently on each run. This is normal. It is also why trends across months matter more than any single response, and why some teams run each question two or three times and record the consensus.
Do personalized results skew my tracking?
They can. Logged-in sessions with long chat histories may color answers, and Google results vary by location. Use fresh sessions or logged-out windows where possible, and note your location in the log so results are comparable month to month.
What should I do when a competitor keeps getting cited and I don't?
Study the sources the engine is citing for those answers — that is your roadmap. Usually the gap comes down to quotable content, structured data, and third-party coverage. Start with our guide on how AI answer engines work to understand what they reward.
Start with a baseline
Tracking only pays off if you begin before you need the data. Build the panel this week, run your first ask-and-record session, and file the screenshots — that baseline makes every future change measurable. And if you want to know where you stand right now, request a free AEO audit and we will show you how visible your site is to the engines your customers are already asking.