Is your page cited for the question it answers?

What this is, and is not

One answer, from one engine, on one day. If your page is cited, that's real. If not, it may be the phrasing, the engine or the day. The Snapshot asks 15 approved questions across all four engines and reads every answer. Method: methodology.

Why one citation matters

A citation is a source link. A mention is your name in the prose. They're different things, and mixing them up is the most common way people misread their own results.

Here's the difference. An AI assistant can say "Acme is a well-known supplier" and never link to you. That's a mention. Or it can answer a question using your page as a source and list it in the sources at the bottom, without ever naming your company in the text. That's a citation. The check on this page looks for the second one: is your page in the assistant's source list?

Why care? Because a citation means the engine read your page and used it. It's the closest thing to a page-level signal you can test in one go. A mention tells you the model knows you exist. A citation tells you the page did its job.

One caveat, and it matters. A citation is a source link, not a click. The check can't tell you about traffic. It also can't tell you why a model picked your page, or whether it'll pick it again next week.

How to read the result

The tool takes your page's heading and turns it into the question. It asks Perplexity with web search on, from London. Then it reports one of two things: your page is in the sources, or it isn't. If it isn't, it also shows who was cited instead.

Start with the verdict. Cited? Good, your page answered that question well enough to be used. Not cited? Read the list of who was.

That second list is the useful bit. Look at what the cited pages have in common. Do they open with a direct answer? Are they from rivals, or from neutral sources like directories and publications? If it's mostly rivals, you've got a competitive gap. If it's mostly neutral sources, the question may simply be one that engines answer from reference material.

Also keep the limits in mind. This is one question, asked once, in a fresh conversation. The same question tomorrow can get a different answer. So treat a single result as a sample, not a ruling. Run it again before you panic, or celebrate.

When your page is not cited

Most fixes are dull, and that's good news. They're quick to check.

Start with the first lines of the page. If the question is "what does this cost to set up" or "how do I choose a supplier", the answer should sit in the opening paragraph, in plain words. Engines favour pages that answer before they warm up.

Next, make the text visible without JavaScript. Most AI crawlers don't run it. If your page builds its content in the browser, the crawler may see an empty shell. The readiness check looks for 600 or more characters of text in the raw HTML, which is a fair baseline.

Then check crawler access. Your robots.txt might block the crawlers that fetch pages at answer time: OAI-SearchBot, Claude-SearchBot, PerplexityBot, ChatGPT-User, Claude-User and Perplexity-User. Blocking only the training crawlers (GPTBot, ClaudeBot, Google-Extended) is a legitimate choice, so don't undo that by reflex. The two groups do different jobs.

Finally, say who you are in structured data. JSON-LD that names the organisation, whether as Organization, LocalBusiness or SoftwareApplication, gives a crawler a clear statement instead of making it guess.

Four things to look at, in this order:

• Does the page answer the question in its first lines?
• Is the text in the HTML before any script runs?
• Can answer-time crawlers get in?
• Does structured data say who you are?

The free readiness check and the robots.txt check cover the last three without an account.

From one question to a full picture

One question is a spot check. It's useful, but it's thin.

A buyer doesn't ask one thing. They ask a dozen, across different parts of your market, and most of those questions never include your name. So the Snapshot widens the test. We read your site and propose 15 prompts across 3 to 5 verticals. You edit and approve every one before anything runs, and you're charged only after you approve.

Most of the prompts are unbranded, the questions a buyer asks before they know who to call. Two are branded, to check the engines know the company exists. One is framed as a comparison. Each prompt goes to ChatGPT, Claude, Gemini and Perplexity, so 15 questions across 4 engines gives 60 answers.

For each answer we record whether you were mentioned, whether you were cited, and where your first citation ranked. Mention and citation are never blended into one score. Every matched mention is printed verbatim and checked before it counts.

The result is a verdict per vertical: Recognised but rarely cited (named in answers, which points to a page problem); Entrenched (one rival appears in most answers); or Open (no vendor named in more than about a quarter of answers). The thresholds are printed next to the counts.

So the single check tells you about one page. The Snapshot tells you where the pattern sits.