How Perplexity decides which companies to name

Perplexity turns your question into searches, fetches a small number of pages, and writes an answer it can support from those pages. So the companies it names are, in practice, the companies those particular pages named. If you are not in the set of pages it fetched, you were never in the running, whatever your reputation.

The three filters, in order

First, the search. Your question becomes one or more queries. If your pages are built for different words from the ones a buyer uses, you lose here, before anything else happens. This filter is invisible and it is where most businesses are eliminated.

Second, the fetch. A handful of results are retrieved. Pages that are slow, blocked, or empty without scripts can drop out at this stage even when they ranked. Perplexity names its crawlers, PerplexityBot for indexing and Perplexity-User for a fetch made on a user's behalf, so this filter is one you can check rather than guess at.

Third, the writing. The answer is assembled from what those pages actually say. This is where specifics win. If one page states a figure and names its scope and the others speak in general terms, the specific one shapes the answer, and often the recommendation.

Why a listing page can beat your own site

A comparison page or a directory naming ten companies in your category is a very efficient source for a question like "what are the best options". One fetch, ten candidates, apparent neutrality. Your own page is one company and is read as advocacy. This is why being present on the pages that list your category matters as much as your own writing, and it is the part most businesses skip.

It also explains a result that feels unfair, and we have a dated instance of it. On 17 September 2026, on a question about finding every case filed against a company, Claude ranked two enterprise vendors above clawlaw.in, our sister company, specifically because those vendors publish explicit court and tribunal coverage lists. The assistant said in the same answer that its ordering reflected price transparency and source authority rather than product quality. That is not a judgement of the products. It is arithmetic about which pages stated their scope plainly and which pages were in the fetch.

What the answer text is actually built from

Read a few answers closely and you will see the same six ingredients being assembled, each from a different kind of source.

• The candidate list, almost always from a roundup or directory page.
• The prices, from vendor pricing pages, because those are the only places prices are stated plainly.
• The feature claims, from vendor documentation more often than from marketing pages.
• The caveats, from forum threads and reviews, which is where the honest problems are written down.
• The rules or definitions, from official pages where the question touches anything regulated.
• The recommendation itself, which is the model's own sentence built from the five above.

That breakdown is useful because it tells you which kind of page to go and fix. If you are absent from the answer entirely, you have a candidate list problem, and no amount of rewriting your own pricing page solves it.

A worked example of the three filters

Start with a real reading of the third filter. On 17 September 2026, on Claude, pages from clawlaw.in, our sister company, ranked in the assistant's raw results and visibly shaped what it wrote. The company still ended up as a name inside a list rather than as a linked recommendation, because its own comparison pages read as vendor advocacy. The first two filters had been passed. The third one was where it lost, and more search work would not have changed that.

To find out which filter is costing you, ask the same question three ways and compare the citations each time.

First, in your own industry language: "field service management platform for SMEs." Second, in the words a buyer would use: "software to manage a team of technicians and their job visits." Third, with a constraint added: "software to manage technicians and job visits for a company with twenty staff in Pune."

If the sources change completely between the first two, your pages are probably written for the wrong words. That is a search problem, filter one, and the fix is vocabulary rather than depth. If the sources are similar but your page is never among them although it ranks in ordinary search, suspect filter two and go and check whether your page returns its content in raw HTML. If your page is cited in all three and you are still not named in the text, that is filter three, and the fix is to put something checkable in the page that the answer can repeat.

The third version is also where small companies win. Adding a constraint brings in different companies, because the broad question has many defenders and the constrained question has few. If your page states plainly who you serve, which sizes, which cities, which sectors, you become the obvious answer to the constrained question, and the constrained question is what a real buyer asks second.

What this means for you

Work on all three filters rather than only the third. Write for the words buyers use. Make sure the page can be fetched and read without scripts. Then give it something concrete to quote, and get your name onto the listing pages that already answer your category questions.

Do them in that order too. Fetchability is a day of work, vocabulary is a week, and getting onto listing pages is an ongoing task you can start now and let run in the background.

What we do not know, and what cannot be promised

Perplexity does not publish how many pages it fetches, how it ranks them against each other, or how it decides which named companies survive into the answer. Everything above is inference from the citations it shows, which is good evidence about inputs and no evidence about weights. Be suspicious of anybody describing this engine's internal scoring, including us.

Because the answer depends on a fresh search each time, results also move. Two people asking the same question minutes apart can get different companies. Treat any single answer as one reading, not as your position, and take several before you conclude anything. No one can guarantee you a place in an answer here or anywhere else.

Common questions

Why did I appear yesterday and not today?

Because a fresh search ran and retrieved a slightly different set of pages. This is the ordinary behaviour of a retrieval engine, not a penalty. Look at the citations in both runs. If a roundup page dropped out of the set, that explains it far better than anything you did.

How often should I check?

Weekly for your most valuable questions, monthly for the full set. This is the engine where frequent checking is justified, because it reacts to new pages quickly, so a weekly reading is genuine feedback rather than noise.

It cited a competitor's comparison page for a fact about us. What should I do?

Open the page and check the fact. If it is wrong or out of date, write to them with the correct value and a link to your own page stating it, and keep a dated copy of what it said. If the fact is correct, the problem is that their page is the only place it is clearly stated. Publish it on your own site, on a page built for that question, and it stops being their fact.

Is being cited the same as being recommended?

No. You can be a numbered citation in an answer that recommends two other companies, which means your page was useful and not persuasive. Keep the two in separate columns, because they need different fixes: citation is a mechanics and relevance problem, recommendation is a content problem.

Can I pay to be included?

Advertising and organic answers are different things, and you should keep them separate in your own thinking. Nothing on this page is about buying placement. If you do run ads anywhere, do not count those appearances in your visibility measurement, because mixing them produces a number that flatters you and tells you nothing.

Do longer pages get cited more often?

Length is not the mechanism. A page gets cited because it answers the question and carries something specific. A long page that hedges for two thousand words is less citable than a short one that states a fact with a date. Length helps only when it is more real content, which usually means more questions genuinely answered on the same page.

What to do first

Take your single most valuable buyer question and ask it in Perplexity three ways, as in the worked example. Write the citations down for each version. Whichever filter the comparison points at, that is your next piece of work, and you will have found it in twenty minutes instead of a quarter.