Platform
Perplexity visibility
An answer engine that cites its sources is the closest thing this field has to a window into the retrieval step. If you only measure one platform closely, measure this one.
AI Says is not affiliated with or endorsed by OpenAI, Google, Anthropic, Perplexity or other AI/search providers. Platform names are used descriptively to identify the services being assessed.
Citations turn guesswork into evidence
The hardest thing about measuring AI visibility is that the interesting part is hidden. A system retrieves some documents, weighs them against what it already holds, and produces a paragraph. You see the paragraph. You do not see the documents — so when your business is absent, you cannot tell whether you were never retrieved, retrieved and passed over, or retrieved and paraphrased without attribution.
Answer engine that cites its sources — the clearest read on citations. That one behavioural difference changes what can be learned. When the sources are listed, absence becomes diagnosable. If a regional directory is cited for a question about your own specialism, the problem is that the directory page was a better source than yours. If three competitors are cited and you are not, the problem is competitive, not technical. If your own site is cited but a competitor is recommended, the problem is positioning. Three very different briefs, distinguishable only because the citations are visible.
This is why we treat observed data from a citing platform as the strongest evidence available for the Citations & Authority component, which carries 15% of the overall score. Everywhere else that component has to lean on estimated signals from your own site; here it can be counted.
Integration required
Observed answers need a connected key
Observed data for this platform requires a Perplexity API key (PERPLEXITY_API_KEY) configured for the account with the provider enabled. Without it the platform panel reads Integration required, and no citation rate is shown — an uncounted citation rate is not an estimable one.
Three separate rates
Being mentioned, being cited and being recommended are measured independently. They move for different reasons and they are fixed by different work. Accuracy is measured as a fourth rate, claim by claim against what is verifiably true.
Diagnosis
What each citation pattern tells you
Four patterns, four different problems. The citation list is what lets you tell them apart.
01Your domain is cited and you are recommended
The mechanism is working end to end. Your page was retrieved, judged the best available source, and the business behind it was put forward. The work here is breadth — which other questions produce this result, and which do not.
02Your domain is cited but a competitor is recommended
You are a credible source about the category and a weaker candidate within it. This is almost always a positioning and proof problem: the page explains the subject well but says little about why this business in particular should be chosen.
03A directory or listicle is cited, and you appear inside it
You are visible only at second hand. The aggregator page answered the question in a more comparable form than your own site did. Your page needs the same compactness — category, location, services and differentiators stated plainly and early.
04Neither you nor anything mentioning you is cited
A retrieval failure rather than a selection failure. Check crawl access, indexability, whether a page for this specific question exists at all, and whether your entity is resolvable enough to be matched to the query in the first place.
The full visibility report records these patterns across the whole question library rather than one answer at a time, which is what turns an anecdote into a brief.
What to change
What makes a page citable
Citability is a property of a page, not of a brand. These are the properties that distinguish a page worth citing from a page that merely exists.
It answers one question, not eight
A citation points at a page as evidence for a claim. A page about one thing is evidence for one thing; a page about everything is evidence for nothing in particular.
The answer is in the first two sentences
Extraction does not read to the bottom. State the answer, then qualify it — the inverse of how most marketing copy is written.
Facts are stated, not implied
Prices in ranges, coverage as named regions, timeframes as numbers, credentials as names. Implication survives a human reader and does not survive extraction.
The page is stable and canonical
A citable URL needs to keep existing at the same address. Pages that move, get rebuilt or live behind parameters accumulate no standing as a source.
Structured data agrees with the prose
Schema that contradicts the visible text lowers confidence rather than raising it. The markup should restate what the page says, not a tidier version of it.
Something about it is verifiably current
A visible date, a changelog, a recent addition. Recency gives a system a reason to prefer your page over an older third-party summary of you.
Next
Read this alongside
Citation behaviour varies enormously between platforms, which is why a single platform is never the whole assessment.
ChatGPT visibility
Often answers from internal representation rather than retrieval, which makes absence harder to diagnose and accuracy harder to correct.
Read the pageGemini visibility
Rewards entity hygiene, and sits alongside AI Overviews — a separate surface requiring a licensed data source.
Read the pageGEO, described honestly
What can and cannot be claimed about influencing retrieval and citation in generative systems.
Read the page
Questions
Questions about Perplexity visibility
Because it shows its working. An answer engine that attaches citations to its claims turns an otherwise invisible process into something observable: you can see which pages were drawn on, whether any of them were yours, and which competitor or directory was used instead. On platforms that cite sparingly, the same information has to be inferred.
It means the directory page was judged a better source for that question than anything on your domain. Usually that is because the directory states the answer in a compact, comparable form — name, category, location, services, a contact detail — while your equivalent page states it across three sections of marketing copy. It is a retrievability and answerability problem, not an authority problem.
No, and keeping them separate is important. Your site can be cited as evidence for a claim about the category while a competitor is the business actually recommended. We measure citation and recommendation as distinct rates for exactly this reason — improving one does not automatically improve the other.
Within limits, yes, and more here than on most platforms. Citations favour pages that are retrievable, specific and self-contained. A page that answers one question clearly, carries the relevant structured data, and does not hide the answer behind preamble is a far more citable unit than a broad services page. That is a concrete, testable change.
Recorded answers with their citation lists. Visibility, recommendation, citation and accuracy become counts over real observations rather than estimates, and the citation data feeds the Citations & Authority component as observed rather than estimated evidence. Without a configured Perplexity key, the panel reads "Integration required".
AI Says is not affiliated with or endorsed by OpenAI, Google, Anthropic, Perplexity or other AI/search providers. Platform names are used descriptively to identify the services being assessed.
Find out what AI says about your business
Enter your website and get your AI Says Score. No credit card required.