Perplexity prints its sources beside the answer, so a citation there can be checked by looking. That is rarer than it sounds. Being cited is not the same thing as being mentioned...the gap between those two words is where most of the advice on this subject falls apart.

A mention is the brand name appearing in the answer text, and a citation is the domain appearing as a source link. Practitioners stare at the gap on a dashboard and assume something has broken...often it only means the tool changed its denominator halfway through. Check the denominator first.

Three Definitions Of Mention

Three vendors publish three different definitions of "mention" in their own help centers...the counting units do not match. That alone is enough to make two dashboards disagree, before any difference in prompt sets, run counts or collection method.

  • Ahrefs: One mention per response, no matter how often the brand is named.
  • Profound: Several mentions per response, stated as such in its own glossary.
  • Semrush: Mentions counted per prompt, and citations counted per response instead.
  • Citations: One per domain at Ahrefs, so five cited pages from a site count once.

Google, meanwhile, never uses the word citation in its own AI features documentation. Its vocabulary is links and supporting links...the whole mention-versus-citation framework is a vendor construction sitting on top of platforms that never adopted it.

The Missing Perplexity Data

Both of the large published studies of the mention-to-citation gap left Perplexity out entirely. Semrush's ghost citations study, run with Kevin Indig and Growth Memo, covered 3,981 domain appearances across 115 prompts and 14 countries on 9 June 2026. BuzzStream published the counterweight five weeks later, across twelve thousand AI responses and 2,967 prompts...once again Perplexity is missing.

  • Semrush: 17.6% of citations carried a brand mention, on its own published percentages.
  • BuzzStream: 69.9% on that same quantity, measured five weeks later.
  • Reverse Quantity: The two point opposite ways, 34.5% against 23.1%.

They disagree with each other by roughly fourfold...neither cites the other. Both studies are still up.

BuzzStream's internal breakdown suggests a mechanism without proving it. Those numbers sit in the same report as its headline. Single-brand queries produced a 39.0% mention-plus-citation rate and list or category queries produced 7.2%...a fivefold swing inside one dataset, driven by prompt type.

One Study With Perplexity

AirOps published a study in October 2025 covering 21,311 brand mentions from 500-plus commercial-intent queries, on GPT-5, Claude Sonnet 4.5 and Perplexity Sonar. The findings it published are stated across all three platforms together...little can honestly be drawn from them about Perplexity alone. AirOps sells visibility software.

  • External: 85% of brand mentions came from domains other than the brand's own.
  • Owned: 13.2% of mentions came directly from the brand's own domain.
  • Format: Nearly 90% of third-party mentions sat in listicles, comparisons or reviews.

Nearly 68% of the brands appeared on only one of the three platforms...cross-platform presence is the exception.

Perplexity's Published Crawler Rules

Perplexity's crawler documentation covers two agents and prints the complete user-agent string for both...a log-parsing regex built from it rests on the vendor's own words.

  • PerplexityBot: "designed to surface and link websites in search results on Perplexity."
  • Training: The same page says it "is not used to crawl content for AI foundation models."
  • Perplexity-User: Supports user actions, with IP ranges published at perplexity-user.json.
  • Propagation: A robots.txt edit takes up to 24 hours to reach Perplexity's systems.

Two statements about Perplexity-User sit in the same table cell and cannot both be operationally true. First, it "controls which sites these user requests can access". Then, "since a user requested the fetch, this fetcher generally ignores robots.txt rules"...a directive that is generally ignored is not a control.

What Perplexity does not publish is any citation measurement of its own, and neither does OpenAI, Anthropic or Google. Microsoft is the only assistant vendor shipping first-party citation data, through the AI Performance report in Bing Webmaster Tools, which covers Copilot, Bing's AI summaries and select partner integrations.

Where Perplexity Measurably Differs

Perplexity's citations track Google's rankings far more closely than any other assistant's do, a gap Ahrefs measured across 15,000 long-tail queries in early July 2025 on four assistants, with every cited URL then cross-referenced against the Google and Bing results for the same query.

  • Perplexity: 28.6% of cited URLs appeared in Google's top 10, and 16.6% in Bing's.
  • Gemini: 8.2% in Google's top 10, against a four-assistant average of 12%.
  • ChatGPT: 8.0% for in-text citations, roughly a third of Perplexity's figure.
  • Copilot: 8.6% in Google and 14.0% in Bing, the biggest Bing lean in the set.

Take the figure the right way round, though. It says that of the URLs Perplexity cites, 28.6% rank in Google's top 10...it says nothing about what share of top-10 pages get cited. Against that, a self-published preprint from a firm selling in this space measured the inverse and put position-one URLs at a 54% citation rate. Each figure answers a different question, and both can be true.

Two further differences turn up in preprints that have not been peer reviewed...both matter for a dashboard.

  • Density, per Sielinski: Perplexity runs 14.5 to 42.2 citations per response, Gemini 19.9 to 50.1.
  • SearchGPT, same paper: 4.2 to 6.3, a platform spread the author calls roughly sixfold.
  • Concentration, per St Gallen: A mean Gini of 0.671 for Perplexity, lowest of four engines.
  • AI Mode: A mean Gini of 0.782, the most concentrated engine in that comparison.

Just One Controlled Test

The only controlled experiment located that tested optimization tactics on Perplexity is the GEO paper, accepted to KDD 2024. It built a retrieve-then-generate pipeline on GPT-3.5, then re-ran a 200-sample subset on Perplexity.ai by uploading the source text as files, because the platform does not let a user specify source URLs. No live page, crawler or index was involved.

Half the category quotes 40% from that paper, and that 40% came from the simulated pipeline on a proxy metric...on the file-upload run the numbers came out smaller and depended on which metric got read.

  • Quotation Addition: Plus 20.7% on Position-Adjusted Word Count, plus 30.0% on Subjective Impression.
  • Statistics Addition: Plus 8.7% on the first metric and plus 37.2% on the second.
  • Cite Sources: Plus 11.2% on the first and minus 23.1% on the second, same runs.
  • Keyword Stuffing: Minus 9.1% on Position-Adjusted Word Count, plus 13.8% on the subjective one.

Its rank table points against the usual advice. On the paper's own GPT-3.5 pipeline, not on Perplexity, adding citations moved a rank-five source by 115.1% and a rank-one source by minus 30.3%, which the authors gloss as GEO being "especially helpful for lower ranked websites".

One Run Is Nothing

In Sielinski's data Perplexity is the steadiest of the three platforms tested, at a median Jaccard of 0.50 on cited domains between identical runs. He compared repeated runs of the same query, with 6,084 to 7,164 response pairs per platform-topic cell. The St Gallen team, measuring same-day re-runs across four engines, puts the ordering the other way round. On that data Gemini is steadiest at 0.505 and Perplexity sits second from bottom. Nobody has reconciled the two.

  • Sielinski: SearchGPT lands at 0.33 to 0.40, with Gemini lower again at 0.29 to 0.31.
  • St Gallen: ChatGPT 0.233, Perplexity 0.282, Google AI Mode 0.318, Gemini 0.505.
  • Zero Overlap: Occurs in 1 to 2% of Perplexity pairs, the lowest rate in Sielinski's study.
  • Aggregation: Brand presence over a campaign beats individual cited URLs, per St Gallen.
  • Sample Size: Roughly 100 queries to pin Perplexity's citation share within five points.

Note that the sample-size figure is citation share for a domain, not whether an assistant says a brand name. A manual test of 96 queries run in Minneapolis in July 2026 found Perplexity roughly 1.4x more volatile than ChatGPT. It ran once. No raw data was published...the agency that ran it sells the service.

Limits Of A Citation

Appearing in the source list confirms placement...it confirms nothing else. Tow Center researchers at Columbia Journalism Review ran 1,600 queries across eight AI search tools, each asked to name the source of a passage Google returns in its own first three results.

  • Error Rate: Perplexity answered 37% of the queries incorrectly, well under the group.
  • Collectively: The eight tools answered more than 60% of all queries incorrectly.
  • Confidence: Wrong answers arrived with almost no qualifying language, per the researchers.

Those queries measured news attribution, not business facts...the study does not establish how often an assistant gets a company's address wrong. What it establishes is that a source list is not a verification layer.

Citations Versus Mentions

Neither has been measured on Perplexity in the way the question implies...the honest answer is a division of labor rather than a winner. Citations are URLs, which means each one can be traced back to the page that earned it and then fixed. A mention has no address.

Every number people reach for on this question is Google-shaped. Ahrefs put branded web mentions at 0.664 against AI Overview visibility in May 2025, then restated the same figure as 0.656 in December, against 0.218 for backlinks.

Both runs drew only on domains above DR 40, and the 26% of brands with no AI Overview mentions were dropped before the arithmetic ran. Even the strongest result gets called moderate to very weak by Ahrefs itself. Seer measured 35% higher organic click-through when a brand is cited in an AI Overview, 0.70% against 0.52%...that firm also stated plainly it cannot prove the citation caused the lift.

Then Pew tracked 900 American adults across 68,879 Google searches in March 2025, 12,593 of which returned an AI summary. Pew is not a vendor. People clicked a link inside the AI summary in 1% of the visits where one appeared...no Perplexity equivalent has been published, which is worth sitting with before a budget gets attached to a citation count.

Where To Start

  1. Allow PerplexityBot in robots.txt, then wait a day, per Perplexity's 24 hour note.
  2. Write the questions buyers actually ask, in their own wording, and freeze the list.
  3. Run roughly a hundred of them rather than five...a handful of runs settles nothing.
  4. Log the cited URL and the brand mention as two separate columns, never as one.
  5. Look at the sentence attached to each citation, and fix the page when it reads wrong.

None of that needs a tool behind it. Where a monthly log is easier to keep than to remember, vizmeta HQ at hq.vizmeta.com runs it free every month across ChatGPT, Claude and Google AI. Perplexity is not among them, so that one stays a manual check.

Read the rest of this advice with the same eye. When a page states a Perplexity citation threshold, a reranker score or an optimal update interval, go looking for where the number came from...none of those figures appears in anything Perplexity has published. Any claim that will not survive one click is not a source.