To audit your own AI citations, write 15 to 30 real buyer questions in your category, ask each one in ChatGPT, Perplexity, Gemini and Google AI Mode, and log three things per prompt: were you named, was your domain linked, and which competitor showed up instead. Repeat the exact same prompts monthly. The pattern of misses is your gap report, and it points straight at the page or third-party source you need to fix or earn. No paid tool required to start.
Why audit your AI citations at all?
Because invisibility is silent. When you drop out of Google rankings, your traffic graph dips and you notice. When ChatGPT recommends three cleaning brands and yours is not one of them, nothing breaks, no alert fires, and the sale just quietly goes somewhere else. The buyer never sees your name, so you never see the loss. An AI citation audit is how you make that silent gap visible.
The second reason is that optimization you cannot measure is just hope. Every article tells you to "be extractable" and "earn citations," but almost none tell you how to verify it worked. The audit closes that loop. It turns "I think we should show up" into "we win 9 of 20 buyer prompts, we lose these 11, and here is who beats us." That is a to-do list, not a vibe.
You do not need our software to do this. We run a single visibility engine across more than 10 brands, and the core of it is exactly the manual loop below, just automated and scheduled. Start by hand. Once you trust the method, decide whether to automate it.
What exactly counts as a citation?
Be precise about what you are counting, or your log turns to mush. A citation is not one thing; it is a small ladder, and you should record where each result lands:
- Named — the answer mentions your brand by name in the body text, even without a link. This still builds awareness and trust.
- Linked — the answer attaches your domain as a source chip, sidebar card, or footnote. This is the strongest signal and the one that can send a click.
- Adjacent — a third-party page about you (a review site, a directory, a Reddit thread) is cited instead of your own site. You influenced the answer, but you do not own the citation.
- Absent — your brand does not appear at all, and a competitor takes the spot. This is the row you care about most.
Recording the rung, not just a yes/no, is what makes the audit actionable. "Named but never linked" is a different problem from "absent," and they have different fixes.
Step one: build a prompt list that mirrors real buyers
The whole audit lives or dies on the prompt list. Generic prompts give you generic, useless answers. Write the questions a real buyer types right before they choose a brand. For a non-toxic cleaning brand that might be "best non-toxic all-purpose cleaner for homes with pets," "is concentrate cleaner cheaper than ready-to-use," or "what cleaning brands are actually plant-based and not greenwashing."
Aim for 15 to 30 prompts spread across three buckets: category questions (the buyer does not know brands yet), comparison questions (you versus a named rival), and branded questions (they type your name to check facts). Keep the list in a sheet so it never changes between runs, because consistency is what lets you compare month to month. This is the same intent-mapping idea behind measuring AI share of voice — your prompt list defines the questions you are competing to own.
Step two: run every prompt across every engine
Open four tabs: ChatGPT, Perplexity, Gemini (or Google AI Mode), and Claude. Paste each prompt into each engine, exactly as written. Use a clean session or a temporary chat so your own history does not bias the answer toward brands you have already searched. This matters more than people expect; a logged-in account that has visited your site repeatedly can surface you when a cold buyer would never see you.
Run all four because each engine grounds its answers differently and you will get different citations. Perplexity leans on community sources and freshness, ChatGPT leans on encyclopedic and editorial references, Gemini and AI Mode favor video and multimodal coverage, and Claude weights clear, authoritative pages. A brand that wins on one can be invisible on another, and you only learn that by checking all four. If you want the per-engine playbook, our guides on how to get cited by Claude and how to get cited by ChatGPT break down the deltas.
Step three: log it in a way you can repeat
Your log is one row per prompt, per engine, with a handful of columns: the prompt, the engine, the date, the rung (named / linked / adjacent / absent), the competitors that appeared, and the specific source the engine cited. That last column is gold. If Perplexity keeps citing a Reddit thread instead of your page, you now know exactly where the answer is being formed and where you need presence.
Keep the structure dead simple. A plain spreadsheet beats a clever tool you will not maintain. The goal is a log you can run identically next month so the deltas mean something. For a fuller framework you can lift, our weekly ChatGPT mention tracking guide lays out the exact columns and cadence, and the DIY AI visibility audit walks the whole loop end to end.
Step four: read the gaps and act on them
Now sort your log by the "absent" rows and group them. Patterns appear fast. Maybe you win branded questions but lose every category question, which means buyers who do not know your name never find you. Maybe you lose a whole engine, which means your content is not extractable in the way that engine prefers. Maybe one competitor keeps winning a single sub-question, which means they own one page you do not.
Each pattern has a fix. Lost category prompts usually mean you need a clean, answer-first page targeting that exact question. Lost on one engine usually means a sourcing problem, not a content problem, so you go earn the kind of citation that engine trusts. A rival owning a sub-question means you study their page and write a better, more extractable one. The gap report is not a scoreboard to feel bad about; it is the most honest content brief you will ever get.
How often, and when does the work pay off?
Run a full audit monthly and a lighter weekly check on your five to ten highest-value prompts. Monthly catches model and crawl shifts; weekly catches fast-moving comparison and freshness-driven prompts before a competitor cements a lead. The cadence is the point — a one-time audit is a snapshot, and snapshots cannot show you a trend.
Set honest expectations on the payoff. After you fix a page or earn a citation, engines have to re-crawl, re-index, and in some cases retrain before your change shows up in answers. Meaningful movement typically takes weeks to a few months, not days. Anyone who promises an overnight jump into AI answers is selling something. What the audit gives you is the proof that your changes are working, slowly and then steadily, instead of you guessing in the dark.
Questions people ask
Build a list of 15 to 30 buyer questions in your category, ask each one in ChatGPT, Perplexity, Gemini and Google AI Mode, and log whether your brand is named, whether your domain is linked, and which competitors appear instead. Repeat the same prompts on a schedule so you can see movement over time rather than a single snapshot.
Run a full audit monthly and a lighter weekly check on your highest-value prompts. AI answers change as engines re-crawl the web and update their models, so a single audit is a snapshot, not a trend. A consistent cadence is what turns the audit into a measurement system instead of a one-time curiosity.
Usually because the competitor has a cleaner, more extractable answer to the exact sub-question, more consistent facts across the web, or stronger third-party citations on sources the engine trusts. The gap column in your audit log tells you which prompts you lose, and that points you straight at the page or source you need to fix or earn.
Want this done for you?
We run this audit monthly across 10+ brands. Want yours? Start with a free AI visibility audit.
Get a free AI Visibility Audit →