Where Perplexity gets its information

Perplexity gets most of what it tells you from web pages it searches for and fetches at the time of the question, and it lists those pages as numbered citations beside the answer. That visible list is the most useful thing on the screen if you are trying to understand your own position, because it shows the actual sources rather than leaving you guessing.

What tends to appear in those citations

From our own reading of these answers across many questions, the following five kinds of source come up repeatedly. The proportions change with the question, and the list itself is consistent.

• Official and government pages, when the question touches rules, registration or public records. These are usually the right answer and will not be displaced.
• Vendors' own pages, more often than people expect, especially when the vendor publishes clear pricing, coverage or documentation.
• Comparison and listing pages, because one page can answer a question about a whole category.
• Forums and community threads, particularly for questions about what something is actually like to use.
• News and trade publications, for anything recent.

If you want a shortcut to understanding your own category, count which of those five dominates your top ten questions. A category answered mainly from official pages is one where you should spend very little. A category answered mainly from listings is one where outreach beats writing. A category answered mainly from vendor pages is one you can win with a keyboard.

How to use the citation list

Ask your buyers' questions, then collect every domain cited across all of them. Sort by how often each appears. That ranked list is where the assistant looks for your category. Being present on the top few is usually a faster win than writing another page on your own site, because you are joining a source it already trusts instead of asking it to trust a new one.

Keep the full URLs as well as the domains. When you go to ask a review site to include you, knowing that it is one specific article rather than the whole site that keeps being cited makes the request concrete and much more likely to succeed.

A worked example

We ran this tally on our own site rather than describing how one would look. Across 24 batches and 72 conversations on aiknowsus.com in September 2026, we counted which domains Perplexity cited most often. The domains at the top were the assistants' own documentation and the vendors' own websites, with a single well known review site appearing far down the list. So the sources deciding these answers in our category were mostly official pages and companies' own pages rather than independent reviews.

The same capture carries the other half of the picture. Across those 24 batches, the phrase recording that we were not cited appears 161 times in the engines' own audits of their own answers. A tally tells you where the answers come from. It does not tell you that you are in them.

Read your own tally as a budget. A source you cannot join, such as an official documentation site, is not a target and appearing beside it is not the goal. A directory or listing that keeps appearing is your highest value target, because it is cited more often than anything else you could realistically be on, and getting listed is usually a form and a phone call. News and trade sites are a medium term relationship rather than a campaign. Competitors' own pages tell you what a vendor page has to contain to be used at all in your category, so open them and note exactly which facts they state that you do not.

The one off blogs are the trap. They are tempting because they look easy to influence, and being cited once each is exactly what tells you they are not deciding your answers.

Why your own page can still win

A vendor page is not disqualified. It gets used when it carries something the other sources do not: a real count, a dated price, a documented method, a named list. What loses is not being a vendor. It is being vague.

Documentation is the underused version of this. Help pages and technical docs are specific by nature, they rarely exaggerate, and they answer exactly the questions a careful buyer asks. If yours are behind a login, you have hidden your most citable asset.

The one source type most businesses forget

Community threads. Questions about what something is actually like to live with, what breaks, and what people regret get answered from forums far more often than from any vendor page. You cannot write those threads yourself, and you should not try to. What you can do is answer the same questions honestly on your own site, because a page that admits the difficult parts is the only vendor page that competes with a candid thread.

The limitations to keep in mind

The citation list shows what was used for that answer, not everything considered. A page can be fetched, read and left out of the list. So absence from the citations is a weaker signal than presence in them.

Perplexity also does not publish how it chooses or weights the pages it fetches, and its own product changes: different modes retrieve to different depths, and what a research mode reads is not what a quick answer reads. So record which mode you used, and do not compare a deep research reading with a quick one.

Finally, the citation list is evidence about one moment. The set moves between runs, so a domain tally is only meaningful across many questions and several readings.

How often to redo this

Monthly is enough for the domain tally. The list is fairly stable, and what changes is your own position within it. Redoing it weekly mostly measures noise, although weekly is fine for the two or three questions you care about most.

Common questions

If my page is cited, why am I not named in the answer?

Because your page was useful for a fact and not persuasive as a recommendation. That is the most common outcome for a vendor page, and it is a content problem rather than a technical one. Look at what the page says where it should state something checkable: a number, a scope, a limit, a date.

How many sources does it read for one answer?

Not published, and it varies by mode and question. You can see how many it cited, which is a floor rather than the real number. Be wary of anyone who states a figure.

Why did the citations change completely between two runs?

A fresh search runs each time, so the retrieved set can differ. Ordinary search results also move. If the change is dramatic and repeatable, check whether you asked the question in slightly different words, because a small rewording is a different query.

It cited a competitor for a fact about us. Why not our own page?

Either your page does not state that fact plainly, or it does not exist as a page about that question, or it could not be fetched. Check in that order. The fix is to state the fact where a buyer asking that question would land, in text, dated, and to make sure the page returns its content without scripts.

Should I be chasing reviews on review sites?

Where those sites are actually cited in your category, yes, and the tally above tells you whether they are. In several categories they barely appear, and in those the effort is better spent on your own documentation and on the directories that do get cited.

Is a citation worth anything if nobody clicks it?

It is worth something even without the click, because it is how your facts get into answers. But you should not pretend a citation is a visit. Count citations, mentions and visits as three separate things, and decide which of them your business actually needs before you optimise for it.

What to do first

Open a sheet and run five of your buyer questions through Perplexity, pasting every citation URL into a column. Then sort by domain and count. The single domain at the top of that list, which you are not on, is your first piece of work, and it is almost never the one you expected.