Perplexity visibility: the engine that cites the most, and sends you nobody
Perplexity visibility is how often the engine cites your site in its answers. Perplexity searches on every question it is asked, where ChatGPT searches on only three in ten. It therefore issues more than half of all citations on the market. Fast Growth Advisors measures that gap engine by engine, with a dated reading and the link to reproduce it.
Note, 9 October 2026. This page describes what engine crawlers and corpus pipelines read: their reading decides citations. Browsing agents (ChatGPT agent, Claude in Chrome, Comet, the Perplexity browser) read something else. They open the page in a real browser, on behalf of a user, and see what JavaScript displays. CSS-hidden text and collapsed FAQs escape them. The text says so where it matters.
How do you improve your Perplexity visibility and get cited?
Publish your message as text served by your server, then make it extractable: those are two different requirements.
PerplexityBot is the crawler that continuously browses the web to feed Perplexity’s index. It does not execute JavaScript. Like the OpenAI crawlers, it downloads the files without running them. Anything appearing after page load escapes it: client logo carousels, third-party review widgets, animated figures, case studies in modals. As a result, it can say what your company does, without being able to say what sets it apart.
Good news, though: the effort goes further here than elsewhere, because the engine searches on every question and each of your readable pages has a chance of entering an answer.
Good extraction does not rescue vague text.
Once extracted, the page still has to say what you do and for whom, or the engine has nothing to cite. That is the part the messaging audit of Fast Growth Advisors measures, with the fifteen Message-Market Fit criteria our Observatory applied to 369 startups.
Why does Perplexity cite more than the other engines?
Because searching is part of its design, as the Fast Growth Advisors readings show.
It is not a model that decides case by case whether to look at the web: it is an answer engine, and it retrieves sources on every question. ChatGPT is the opposite, a model that may choose to search, and does so only one time in three.
| Engine | Searches | Citations issued | Of which to our sites |
|---|---|---|---|
| Perplexity | 100 % | 394 | 11 |
| Claude | 85 % | 269 | 4 |
| Gemini | 35 % | 71 | 10 |
| ChatGPT | 30 % | 34 | 2 |
Perplexity issues 394 of the 768 citations recorded, more than half on its own. ChatGPT issues 34. The same editorial work therefore does not carry the same value depending on which engine your buyers use.
One nuance the table does not show.
Gemini cites little, 71 times, but ten of those citations point to our sites. Perplexity issues five times more for eleven. An engine’s citation volume says nothing about your chances of appearing in it.
Does Perplexity use its own models?
Only in part, and this point is often misunderstood. Perplexity does not train a foundation model from scratch. Its in-house family, Sonar, consists of fine-tuned versions of Meta open models, and its own legal page files them under the Llama licence. Depending on the plan and the question, it also calls OpenAI’s GPT, Anthropic’s Claude or Google’s Gemini.
It does, however, run its own index, fed by two crawlers with distinct roles.
| Crawler | Declared role | Obeys robots.txt |
|---|---|---|
| PerplexityBot | Crawls the web continuously to feed the index | Yes |
| Perplexity-User | Reads a page at the moment a user asks a question | No, as the request comes from a human |
Its documentation states that neither crawler is used to train foundation models, and publishes their IP ranges, which makes them verifiable.
That verification is not theoretical. A declared user agent can be forged in one line, and on the servers of Fast Growth Advisors most of the traffic presenting itself as an AI was not. 876 impostors, what your AI traffic really hides →
A practical consequence nobody writes down. Blocking GPTBot and blocking PerplexityBot do not have the same effect. The first is about model training, the second about your presence in answers. And since Perplexity-User ignores robots.txt, a disallow does not prevent a read triggered by a user.
Source: Perplexity documentation, docs.perplexity.ai, consulted on 2 September 2026. This paragraph reports what the vendor publishes. Our measurements cover observed behaviour, not internal architecture.
What this measurement says, and what it does not. Twenty questions across four engines, two domains, a single day. It is an order of magnitude, not a definitive ranking, and the gap between 100 % and 30 % is too wide to come from chance. We have not measured whether these proportions hold on another question set.
Does being cited by Perplexity bring traffic?
Not mechanically, and this is the most uncomfortable result of the Fast Growth Advisors measurements.
That same week, the logs of our two sites show 1,376 visits from crawlers whose identity we could verify, plus 22 fetches launched live by a user’s question. Behind each of those 22 fetches, a real person was waiting for an answer the engine came to our pages to build.
Visitors arriving from an AI answer over the period: none.
Citation and visit are two different things. Readers get the answer inside the engine’s interface, and the link stays a footnote they rarely open. Counting citations to forecast traffic means confusing two measurements that do not follow each other.
That figure is also a starting line, measured before any optimisation. Whatever moves next will be attributable.
This goes beyond Perplexity. On the sites most exposed to Google AI Overviews, Ahrefs measured a 23.1 % drop in click-through rate over nine days, with impressions holding steady. The blind spot nobody measures →
What is left of your page when PerplexityBot reads it?
PerplexityBot, like any crawler, never sees your page the way you do: it gets whatever text an extractor kept. Three layers decide what gets through. The crawler must reach the page, the text must sit in the bytes returned, and the extractor must keep it. Almost nobody checks that last one.
Our measurements cover three B2B software vendor sites, two in HR software and one in workforce scheduling. Each page was fetched with the PerplexityBot user agent, exactly what the crawler receives, then run through the three reference extractors.
| Page | Text in served HTML | trafilatura | jusText | Resiliparse |
|---|---|---|---|---|
| HR software vendor A | 25,387 characters | 59 % | 26 % | 95 % |
| HR software vendor B | 13,425 characters | 23 % | 18 % | 40 % |
| Scheduling software vendor | 12,832 characters | 49 % | 35 % | 91 % |
On HR software vendor B, trafilatura keeps just 23 % of the served text, jusText 18 %. About four fifths never reach the engine.
Resiliparse keeps 95 % of vendor A’s text, where jusText keeps 26 %. Same page, same crawler: what reaches the engine depends on the extractor it uses.
Between the two sits the scheduling software vendor, with 35 to 91 % of its text getting through the three extractors.
Same fetch again with the GPTBot user agent, on the same pages at the same time. All three sites served byte-for-byte identical content to both crawlers. None of them varies its response according to the agent asking.
This measurement therefore holds for every crawler that does not execute JavaScript, and there are many. A page that extracts badly does so for Perplexity as much as for ChatGPT, and a single fix covers both.
Your site is not invisible to AI, it is unreadable →
The limits of this test. Only extraction is measured, on the bytes a crawler receives when it runs no JavaScript. Text that JavaScript later paints in a browser comes on top of that loss and does not make up for it. Three pages make no statistic; they show a mechanism you can rerun on your own site in ten minutes.
What extractors keep, measured on three sites →
White paper, the recoverable message →
What we do not know about Perplexity
Two blind spots, which Fast Growth Advisors prefers to name rather than fill with an assumption.
It does not expose the queries it issues. ChatGPT and Claude let you see what they are about to search for, which allows a comparison between the user’s intent and the keyword actually queried. Perplexity does not. So there is no way to say which language it searches in, nor how it rewrites.
We have not measured the stability of its answers.
On Google AI Overviews we replayed the same question eleven times: the sources change from one call to the next, with a median overlap of 62 %. There is a stable core and a rotating fringe. Assuming Perplexity behaves the same way would be a guess, since the experiment has not been repeated there.
Three states, never merged. Either the engine cites you, or it answers without citing you, or it does not answer that question. That third state is not invisibility: it is an absence of measurement. A tool that merges the last two reports a problem nothing established.
The Google AI Overviews barometer, method and editions →
Make your expertise citable by ChatGPT and Claude →
FAQ
How do you get cited by Perplexity?
Publish text served by your server, then make that text extractable.
PerplexityBot does not execute JavaScript: anything rendered after page load does not exist for it. Since the engine searches on every question it is asked, each of your readable pages has a chance of entering an answer, provided it can read them.
Does Perplexity cite more than ChatGPT?
Far more.
Across twenty questions put to four engines, the Fast Growth Advisors measurements of 6 August 2026 count 768 citations in total, 394 of them issued by Perplexity and 34 by ChatGPT. That gap comes from a decision taken upstream: Perplexity searches on 100 % of questions, ChatGPT on 30 %.
Does a Perplexity citation send visitors to your site?
Not mechanically.
Over the same week our two sites received 1,376 verified crawler visits and 22 fetches triggered by a live user question, for zero visitors arriving from an AI answer. Citation and visit are two different things, and the second does not follow the first.
The rest of this series:
Measure AI visibility, the method and the instruments →