How to Check Whether AI Engines Actually Cite Your Site

There is no rank tracker for AI search, so visibility has to be measured by asking the engines directly and recording what they say. The method is a fixed panel of buyer questions, run against each engine on a schedule, logging three things per answer: were you cited, at what position, and which domains were cited instead.

That third column is the one that changes what you do next. Most teams expect to find their competitors there and instead find review sites, listicles and forum threads they have never engaged with.

Here is how to run it, manually or with tooling.

Why readiness scores are not enough

Most AEO tools score whether your pages look citable: answer-first openings, schema, headings, bylines. Useful, and worth fixing. But it is an input, and it is a guess about a system nobody has the source code to.

The output is checkable. A page can score 95 for citability and never be cited once, because citation depends on authority, freshness and third-party corroboration that no on-page score can see. If you only measure readiness, you can work for six months, watch a number climb, and have no idea whether anything happened.

Measure the output. Use readiness to explain it.

Step 1: build a query panel

A query panel is a fixed set of questions you will ask repeatedly. Fixed is the important word — changing the questions between runs makes the results incomparable.

Aim for 15–20 queries across four types:

Type	Example shape	What it tells you
Brand probe	"What is {brand}?"	Whether you exist as an entity at all
Category	"Best tools for {job to be done}"	Commercial visibility, the hardest to win
Informational	"How do I {problem you solve}?"	Top-of-funnel citation opportunity
Comparison	"{Competitor} alternatives"	Displacement opportunities

Two rules that matter more than they look:

Exclude your brand name from everything except the brand probes. If you ask "is {brand} good for X", you have told the engine what to talk about. That measures nothing. Unprompted discovery is the whole point — you want to know if you come up when nobody mentions you.

Write them the way a person would speak. Assistant queries are conversational and longer than search keywords. "What's the best way to track whether AI is mentioning my company" is a realistic query; "AI visibility tracking tool" is a keyword.

Step 2: choose engines

Pick the ones your buyers actually use, and be consistent:

• ChatGPT with web search — largest assistant audience.
• Perplexity — the most citation-transparent; every answer lists sources. The best starting point if you only run one.
• Google AI Overviews — the highest-volume surface, and the most volatile. Note that it does not appear for every query.
• Gemini and Copilot — worth adding if they matter to your market.

Each engine has its own retrieval behaviour and its own preferred sources. Results differ substantially between them, so record per-engine rather than averaging into one number.

Step 3: record the right things

For every query-and-engine pair, log:

1. Cited (yes/no) — did your domain appear in the source list?
2. Position — where in the citation list. First is worth considerably more than fifth.
3. Brand mentioned but not cited — named in the text with no link. Partial credit; it means you are known but not treated as the source.
4. Every domain that was cited — the full list, not just yours.
5. The answer text — so you can see how you were characterised, and catch inaccuracies.

Point 5 catches things a score never will. It is fairly common to find an engine describing your product with a feature you removed a year ago, or attributing a competitor's limitation to you.

Step 4: read the citation leaderboard

Aggregate every cited domain across every query and rank by frequency. This is the single most useful artefact the exercise produces.

A typical result for a commercial category looks something like:

g2.com                 14 citations
reddit.com             11
somereviewsite.com      9
competitor-a.com        6
a-listicle-blog.com     5
yoursite.com            1

What this tells you is concrete and actionable:

• Your real competitive set is the list, not just the vendors in it.
• Aggregators and communities usually dominate. Being accurately represented on them is higher-leverage than another blog post on your own domain.
• Where a competitor's own site is cited, they have earned direct trust — worth studying what that page does.
• Your own count is the baseline you are trying to move.

Step 5: run it on a schedule

Once is a snapshot; the value is in the trend. Weekly is a reasonable cadence — frequent enough to catch movement, infrequent enough that noise does not swamp it.

Expect variance. The same query can produce different sources on consecutive days. Do not react to a single run. Look for direction over four to six weeks, and treat any single-run change as noise until it repeats.

Doing it manually

Perfectly viable at small scale, and worth doing once by hand even if you later automate, because reading the answers teaches you things a dashboard flattens.

1. Put your queries in a spreadsheet, one row per query-and-engine pair.
2. Open each engine in a fresh incognito window — logged-in sessions carry personalisation that contaminates results.
3. Ask each query, paste the answer, list the cited domains.
4. Mark your own citation and position.
5. Repeat weekly. Budget two to three hours for 20 queries across three engines.

The incognito detail matters more than it sounds. Assistants adapt to conversation history, and a session where you have been discussing your own company will cheerfully cite it.

Doing it with tooling

At 20 queries across 3 engines weekly, that is 60 manual checks a week, and the transcription is where errors creep in. This is the problem OPSYRA's AI Visibility tracker was built to solve: it constructs the panel from your business profile and validated topic model, runs it against live engines on a weekly schedule, and stores the full answer text and citation list for every run so the trend is comparable over time. The citation leaderboard and the off-page action plan are generated from the measured results rather than estimated.

Whatever you use — ours, someone else's, or a spreadsheet — the requirements are the same: a stable panel, unbranded queries, per-engine results, full citation lists, and history.

What to do with the results

The leaderboard points at the work:

• Cited nowhere, brand probes fail. You have an entity problem before you have an AEO problem. Focus on being described consistently across your own site, your social profiles and third-party listings.
• Cited for informational queries, absent from commercial ones. Normal, and the usual shape. Commercial queries are won through third-party mentions, not your own content.
• Brand mentioned but never cited. You are known but not trusted as a source. Usually a content-structure and authority problem on your own pages.
• Losing to one dominant aggregator. Get accurately listed there. It will outperform months of blogging.

FAQ

How do I check if ChatGPT mentions my website?

Ask ChatGPT, with web search enabled, a set of questions your buyers would ask without naming your brand, then check whether your domain appears in the cited sources. Use a fresh incognito session so conversation history does not influence the result, and repeat the same questions weekly to see a trend rather than a single snapshot.

What is share of voice in AI search?

Share of voice in AI search is the proportion of your query panel where your domain is cited, relative to competing domains cited for the same queries. If you are cited in 3 of 20 queries and a competitor in 9, they hold a substantially larger share of voice across that panel. It is only comparable run-to-run if the query panel stays fixed.

How often should I check AI visibility?

Weekly is a sensible cadence for most businesses. AI answers vary day to day, so a single check is unreliable, and daily checking mostly measures noise. Weekly runs on a fixed panel let you distinguish a real directional change over four to six weeks from ordinary variance.

Why does ChatGPT cite my competitor but not me?

Most often because a third-party source the engine trusts mentions them and not you. Engines answering commercial questions synthesise from review sites, listicles and forum threads rather than vendor websites, so the citation usually reflects presence on those pages rather than anything about your own site. Less commonly it is a content problem — no passage on your site directly answers the question — or a crawling problem where your content is not readable at all.

Can I track AI visibility in Google Analytics?

Only partially. Referral traffic from assistants shows up in analytics when a user clicks through, but a citation without a click leaves no trace, and that is the majority case. Analytics therefore understates AI visibility substantially. Measuring citation presence directly, by querying the engines, is the only way to see the part that generates no traffic.