How to get cited by Perplexity

Every AI engine decides which sources deserve a place in its answer. Perplexity simply offers more places. Cross-engine citation analyses put it at roughly 8.2 sources per answer on average, about 3.4 times what ChatGPT shows. Where a ChatGPT answer might cite two or three pages, a Perplexity answer routinely cites eight, and every one of those slots is an opening for a page you control or a page that names your brand. If you are choosing where to spend your first month of AI visibility effort, the engine with the most doors is a rational place to start.

The audience is smaller than ChatGPT's but unusually valuable. Perplexity handles an estimated 50 million weekly queries, against an estimated 250 to 500 million for ChatGPT Search. What the raw numbers hide is intent: G2's 2026 buyer research found 44 percent of B2B software buyers use Perplexity during shortlisting, the stage where vendors get cut and deals get shaped. A citation that appears while a buyer trims a shortlist is worth more than most impressions anywhere else in the funnel.

There is one more structural advantage: transparency. Perplexity shows its sources prominently on every answer, which means you can see precisely which pages beat you, question by question. No other major engine gives you a cleaner map of what to fix. This playbook is essentially a method for reading that map and acting on it.

Set expectations before you begin: this is a weeks-long loop, and each pass through it compounds. The first cycle establishes a baseline and usually stings, because most brands discover they hold few or none of the citation slots on the questions that matter to them. G2's broader finding, that 51 percent of B2B tech brands have zero citations across ChatGPT, Perplexity and Gemini, suggests the sting is the normal starting point rather than a verdict. The brands that close the gap are simply the ones that run the loop more than once.

Perplexity is retrieval-first by design. Rather than leaning on a model's memorized knowledge, it searches its own index, reported at more than 50 billion pages and refreshed by its crawler, PerplexityBot, then reads the top results and composes an answer with citations attached. That architecture has a practical consequence for brands: nearly everything Perplexity says about your category flows through pages that exist on the web right now, which means nearly everything is influenceable on a timeline of weeks.

Contrast that with ChatGPT, where a large share of brand knowledge sits in training data and moves on retraining timelines nobody outside OpenAI controls. With Perplexity, the loop is tighter: publish or earn a better source, get it crawled, and the answer can change at the next retrieval. The fuller anatomy of the engine, including how it ranks within its index, is covered in how Perplexity picks its sources.

One warning before you reuse work from other engines: the overlap is thin. Cross-platform citation analyses find only about 11 percent of domains are cited by both ChatGPT and Perplexity. A page that wins ChatGPT answers earns no automatic standing here, and vice versa. Treat Perplexity as its own campaign with its own source diet, which is exactly what the next steps map out.

Write down 20 to 25 unbranded questions your buyers ask when they shop your category: the comparisons, the "best X for Y" queries, the problem-shaped questions that precede a purchase. Run each one in Perplexity and record two things: which brands the answer names, and which sources it cites. The second list is the strategic gold, because it tells you which specific pages currently own your buyers' answers.

Run each question more than once. AI answers are probabilistic: SparkToro measured under a 1 percent chance that two identical runs of an AI assistant return the same brand list, and Perplexity's retrieval adds its own variability as the index refreshes. Three to five runs per question over a few days gives you a distribution instead of an anecdote. Keep branded questions out of your score entirely: an answer to "is your brand good" mentions you by construction and teaches you nothing about discovery.

This baseline is tedious by hand, which is why it usually gets done once and never repeated. Reachroller runs the same question set on a schedule through official APIs, stores every raw answer for audit, and separates branded from unbranded questions automatically. Perplexity tracking is built and rolling out, with ChatGPT live today, and the methodology documents exactly how a mention is counted. However you run it, do not skip the baseline: without one, you will never know whether the work that follows moved anything.

Perplexity does not read the web the way you might guess. Its single largest source is Reddit, with estimates of Reddit's citation share ranging from roughly 17 to 24 percent, and one analysis measuring Reddit at 46.7 percent of Perplexity's top-10 citation share. Relative to other engines, Perplexity also skews toward LinkedIn, NIH and G2. Community threads, professional posts and review platforms carry weight here that traditional publishers would envy, and major newspapers underperform their prestige across every engine studied.

The pattern holds across the research: Otterly's 2026 AI Citations Report, built on more than a million data points, and Semrush's three-month most-cited-domains study both confirm community-content dominance with a long, fragmented tail of niche sites filling the remaining slots. Reddit's citation share in commercial categories grew roughly 73 percent across 2025 and 2026, so the skew is strengthening rather than fading. The cross-engine picture is mapped in how AI assistants choose sources.

Your move in this step is specific, and it comes straight from your step 1 logs: list the exact domains and threads Perplexity cited for each question you lost. That list becomes your outreach map in step 4 and your competitive benchmark in step 3. If a four-year-old Reddit thread owns the answer to your most valuable buying question, you now know precisely what you are competing with, and it is beatable.

Perplexity fills eight citation slots per answer, and your own site can earn one of them on merit. The evidence for what merits look like comes from the Princeton and Georgia Tech GEO study published at KDD 2024: adding quotations, statistics and cited sources boosted visibility in generative engine responses by up to roughly 40 percent, with the best methods improving about 22 percent on position-adjusted word count and 37 percent on subjective impression versus baseline. Keyword stuffing, the reflex tactic from classic SEO, performed near the bottom.

Shape each page around one buyer question. Use the question as the h1, answer it completely in the first 90 to 130 words so the paragraph can be quoted standalone, then support it with attributed numbers, named sources and an honest FAQ. Perplexity synthesizes across its eight sources, so a page that contributes a concrete fact the others lack, a price, a benchmark, a direct quotation, earns its slot even against bigger domains. The full page anatomy is in how to write content AI engines actually cite.

Volume matters, because you likely lost 10 or 15 questions in step 1 and each needs its own page. Reachroller generates a publish-ready fix page for every lost question, complete with slug, title tag, meta description and schema markup, at ten credits per fix; you review and publish on your own domain. Whether you write by hand or generate and edit, hold the same bar: every claim attributable, every number sourced, nothing you would be embarrassed to see quoted verbatim inside an AI answer. Because with Perplexity, verbatim quoting is exactly what you are trying to earn.

Given Reddit's outsized share of Perplexity citations, a Perplexity strategy without a Reddit strategy is half a strategy. The wrong version is astroturfing: fake accounts praising your product get detected, banned and publicly documented, and the callout threads then get crawled and cited by the same engines you were trying to game. The durable version is disclosed, useful participation: a named brand account answering real questions in your niche subreddits, contributing benchmarks and honest comparisons, taking criticism without deleting it. The complete approach is in a Reddit strategy for brands that AI engines respect.

G2 matters twice for software brands: it skews high in Perplexity's citation diet, and it is where a large share of B2B buyers already research. Keep the profile complete and current, and cultivate genuine reviews at a steady pace rather than in suspicious bursts. LinkedIn, another Perplexity-favored surface, rewards substantive posts published under named people. An engineer's detailed comparison post can end up cited in answers where your product pages never appear.

Then work your step 2 outreach map. The roundups and comparison posts Perplexity already cites for your lost questions are the highest-leverage pitches available to you: one added mention on a page the engine already trusts can appear in the very next retrieval. Citation analyses from Ahrefs and Semrush consistently show engines leaning on independent third-party pages over brand-owned domains, so treat earned mentions as equal partners to your own publishing, not as a nice-to-have.

None of the content work matters if PerplexityBot cannot reach your pages. Check robots.txt first: blanket AI-blocking rules added during the 2023 panic often still deny every AI user agent, which quietly removes you from Perplexity's 50 billion page index. Allow PerplexityBot explicitly if you want citations. Distinguishing search crawlers, which produce citations and referrals, from training crawlers is a policy decision worth making deliberately, and the bot-by-bot breakdown is in the AI crawler boom, measured.

Keep the basics sound: clean HTML that renders without JavaScript acrobatics, fast responses, a current sitemap, and canonical URLs that do not fragment your content across duplicates. On structured data, stay evidence-based. SE Ranking found roughly 71 percent of pages cited by ChatGPT carry structured data, a correlation, while Ahrefs' May 2026 test on 1,885 already-cited pages found no measurable citation lift from adding JSON-LD. Ship schema because it is cheap and may help parsing; do not expect it to substitute for substance.

Finally, give new pages time to enter the index before judging anything. One to two weeks is the honest interval between publishing and a fair recheck. Perplexity's refresh cadence works in your favor compared to slower engines, but it is not instant, and rechecking on day two only manufactures false negatives.

A note on scale for teams running this across many questions: the technical pass is a one-time audit, but the content and outreach steps repeat per question, and the arithmetic adds up quickly. Fifteen lost questions means fifteen citable pages plus fifteen small outreach campaigns, which is where most manual efforts quietly stall after the third page. Decide up front what your sustainable weekly cadence is, one page and one pitch per week beats five in week one and none after, and let the baseline data pick the order: highest-intent questions first, questions with churning citation lists before questions locked up by giants.

After the one to two week interval, re-run your full question set exactly the way you ran the baseline: same questions, multiple runs each, every answer and citation logged. Look for movement in the distribution rather than a single dramatic flip. Your page appearing in the citation list for three of five runs where it appeared in none is a real win; one appearance in one run is weather. Perplexity's visible citations make this audit easier than on any other engine, which is another reason it rewards systematic effort.

Two outcomes are worth distinguishing when you read the results. A citation means your page was retrieved and used as a source, with a visible link. A mention means your brand was named in the answer text, whether or not your site was the source. Both matter, they move through different mechanics, and a good measurement habit tracks them separately. Reachroller counts a mention only when the brand name literally appears in the stored answer text, links every score to the raw answer behind it, and schedules the recheck for you, so the before-and-after comparison survives scrutiny.

Then keep the loop turning. The brands that hold Perplexity citations are the ones that keep publishing citable pages, keep earning community mentions and keep measuring, because the index refreshes constantly and every competitor reading a playbook like this one is a retrieval away from taking your slot. If ChatGPT is your other priority surface, the companion playbook is how to get your brand mentioned by ChatGPT.

Keyword stuffing and its cousins. The GEO study measured classic keyword repetition near the bottom of all tested methods for generative engines, worse than leaving the page untouched. A retrieval engine that reads eight sources and synthesizes across them is choosing pages for extractable substance, and a page that repeats its target phrase offers none. The same logic sinks thin programmatic pages: a hundred templated variants of the same shallow answer give Perplexity a hundred reasons to cite the one community thread that says something concrete.

Buying your way in.There is no paid placement in Perplexity's organic citations, and services that promise guaranteed citations are promising a mechanism they do not control. The related temptation, paying for bulk fake Reddit activity, is worse than useless: detection leads to bans and public callout threads, and those threads live on the exact surface Perplexity cites most. The engine's Reddit habit makes manufactured community praise a liability with distribution.

Unproven technical rituals. llms.txt remains an unratified proposal that no engine has confirmed using, so give it ten minutes if you want the optionality and no more. And skip one-off visibility checks as a management tool: a single favorable Perplexity answer proves as little as a single unfavorable one. The volatility research is clear that distributions, tracked over repeated runs, are the only readings worth reporting, which is why every step in this playbook ends in a logged, repeatable measurement rather than a screenshot.