Who tracks how often AI assistants cite my website?

Most answers to this question list four tools and stop. The more useful thing to know first is which engines can be measured at all, because that determines what any vendor can honestly sell you.

What is measurable, engine by engine

The fourth column is ours and it is the uncomfortable one. Perplexity is the only engine we hold an API key for. We publish that here rather than behind a sales call, because a vendor’s coverage claim is the first thing worth checking and “all the AI platforms” almost always conceals a manual process.

1. First-party reports, from the engine itself

The best data, and there is almost none of it. Microsoft publishes an AI Performance report inside Bing Webmaster Tools, which gives site owners Copilot citation data from the source rather than from a sample of questions. If you care about Copilot, that is the report to open first, and it is free.

Google Search Console has no separate AI Overviews breakout. Impressions from AI Overviews are counted inside ordinary Web search totals, which means you cannot isolate them there. OpenAI, Anthropic and Perplexity offer no publisher-side citation dashboard at all. Check each provider’s current documentation before you rely on any of this — it is the part of the field that changes fastest.

2. Third-party trackers

These ask the engines a fixed set of questions on a schedule and record who is named and which sources are cited. That is a sample, not a census: it tells you what the engine says when asked your questions, which is the number that matters commercially, but it is not a count of every time you were cited to anybody.

When we asked Perplexity this exact question on 19 September 2026, as part of our own audit, it named Scrunch, LLM Pulse, Beamtrace and UltraScout, and pointed to Bing Webmaster Tools for Copilot. Asked a related question about tracking tools generally, the same run named Peec AI, Profound, Otterly, AthenaHQ, Ahrefs Brand Radar and the HubSpot AEO Grader. It did not name Bungad in either answer, or in any of the eight questions we asked.

We compare the main ones, including ourselves and including where we are the wrong choice, on AEO tools compared.

3. Doing it yourself

Entirely possible, and for a small site it is the right first step. Query the engines that expose an API with the questions your buyers ask, parse the answer for your brand name and read the citation list. Keep every run.

Two things make this harder than it sounds. Answers are non-deterministic, so one run is an anecdote — and engine errors must never be recorded as “not mentioned”, or your score silently becomes a measure of your own timeouts. Our method, including the failure rules and the exact weighting, is written out on how we measure AI visibility, and the whole audit can be reproduced with one command.

Why analytics cannot answer this question

Referral traffic and citation frequency are different measurements, and confusing them is the most common mistake in this area.

A large share of AI answers end the journey. Your analytics will not record that as a lost visit, because it never was one — which is exactly why the loss is invisible until you go and ask. Use both numbers: citations tell you whether you are in the answer, referrals tell you what that was worth.

How often to check

Monthly against a fixed question set, at minimum. A single check proves nothing, because engines are non-deterministic: ask the same question twice and the wording, the brands and the sources move.

We measured that on ourselves. We ran our own audit twice, about a minute apart, and kept both records. The headline scores reproduced exactly — 0 of 8 in both, with the same five competitors named the same number of times. The per-domain citation counts did not: ahrefs.com went from 3 to 4, blog.hubspot.com from 5 to 4, peec.ai from 4 to 2.

So treat per-domain counts as estimates and the trend across months as the real signal. And keep the question set fixed — changing the questions resets the series, which is also the easiest way for a vendor to show you a rising number that means nothing.

The three numbers worth tracking

• Mention rate — the share of answers that name you at all.
• Citation share — the share of cited sources that point at your own domain.
• Position — where you appear among the brands named, because the first name carries most of the intent.

Track them monthly against a fixed question set. Changing the questions resets the series, which is why our own audit stores the prompt list verbatim in every record.

Where we are the wrong choice

• You need Copilot citation counts. Open Bing Webmaster Tools. It is free, first-party, and more complete than any sample-based tracker including ours.
• You need ChatGPT numbers this month. We do not have them yet. Buy a tracker that does — Peec AI and Otterly both cover it.
• You only want the data. A self-serve tracker is cheaper than our $99 plan and gives you more dashboards. We are priced for doing the fixing, not for the reading.

Common questions