Perplexity AI SEO: How Perplexity Indexes and Cites Your Site

Perplexity reaches the web through two documented agents, and they do different jobs. Mixing them up is the most common source of bad Perplexity advice.

What this means in practice:

• PerplexityBot is the gate into the index. If it can't fetch and store your page, the page can't be a stored candidate for Perplexity answers.
• PerplexityBot is not a training crawler. Allowing it does not hand your content to a model's training set, per Perplexity's documentation. That makes it a different decision from GPTBot or ClaudeBot.
• Perplexity-User is a separate path. Because it acts for a person in real time, robots.txt is not the control for it. If you need to keep a page away from it, robots.txt alone won't do that.

The next move is simple: decide on PerplexityBot as a search-visibility question, not a training question, and stop expecting robots.txt to govern user-triggered fetches.

Perplexity cites whole sources, each referenced by an integer index. A number in the answer points to a source in the list. Claude works differently: its API cites text spans, with a short quoted snippet attached to each citation. Our page on why Claude and Perplexity cite sources differently covers the contrast in detail.

The developer side shows the same structure. As of October 2026, Perplexity's Agent API returns search_results with fields including id, url, title, snippet, date and last_updated, and the answer refers back to those results by index. For a field-by-field walkthrough, see Perplexity citations explained.

Three things follow for a site owner:

• A citation credits the page, not a sentence. You learn that a URL was used, not which line mattered.
• Citation is not the same as influence. Research on attribution found that citation output can be changed without changing the sources, which is covered in our note on how LLMs cite. Count citations, but don't read them as a precise measure of use.
• Web search is opt-in on the Agent API. Code that calls Perplexity without enabling web search gets an answer with no live sources. Perplexity's Sonar Chat Completions API was scheduled to retire on 2026-09-27, and unmigrated code can silently return ungrounded answers. If you test Perplexity through the API, confirm search is on before trusting the result.

It helps to split Perplexity's process into four stages.

1. Parametric. What the underlying model knows with no web access. You can't change this in the short term.
2. Access. Whether PerplexityBot may fetch and store your page. This is yours to fix.
3. Retrieval. Whether your page enters the context for a given question. Perplexity decides this, and its retrieval queries are not visible to you.
4. Generation and citation. Which retrieved pages get used and cited. The model decides.

The model behind an answer also varies. Secondary reports as of September 2026 say Perplexity's free tier uses an auto-router called "Best" and paid tiers offer a menu of models. So "what Perplexity says" is not one fixed thing.

Where the evidence is strong, it's about stage 2. A critical survey of 45 GEO studies (arXiv, July 2026) concluded that already-retrieved content can change its citation, but no reviewed technique shows a stable, cross-platform causal effect on discoverability. C-SEO Bench found most conversational SEO rewrites "largely ineffective" and often harmful to ranking in context.

The honest split:

• In your control: PerplexityBot permission, firewall and CDN reachability, server-rendered facts, and clear factual statements on pages that can be fetched.
• Partly in your control: being mentioned on the third-party pages Perplexity already cites. The evidence shows an association here, not a guaranteed effect.
• Not in your control: which retrieved page Perplexity cites, the queries it runs, and the model a user is routed to.

Work through access first. Each step removes a possible block, and a later step is wasted if an earlier one fails.

1. Check robots.txt for PerplexityBot. Make sure no blanket AI block or wildcard rule disallows it. A minimal explicit allow looks like this:

User-agent: PerplexityBot
Allow: /

2. Test real reachability. A firewall, CDN or bot-management rule can return a 403 or a challenge page to a crawler that robots.txt allows. Request a few key URLs with the PerplexityBot user agent and check the status code. The AI Crawler Checker does this per crawler. Cloudflare changed its AI crawler defaults on 2026-09-15 for new domains and free-tier sites, so review your bot settings even if you edited nothing.
3. Put load-bearing facts in the HTML. Vercel's crawler study found AI crawlers fetch JavaScript files but don't execute them, and independent testing in June 2026 found PerplexityBot does not render JavaScript. Fetch your page with JavaScript off and confirm prices, product names and key claims are there.
4. State specific facts plainly. A price, a number or a one-line definition on a fetchable page gives the engine a sentence it can use. This is the only on-page step with any positive experimental signal.
5. Fix contradictions on your own site. If your pricing page and your about page disagree, any engine reading both can repeat the wrong one.
6. Find the third-party pages Perplexity cites for your category. Our playbook on finding the pages AI engines cite shows how. Being accurately named on those pages is the route into answers where Perplexity prefers other sites.
7. Then check what Perplexity actually says. If an answer is wrong, correct the sources Perplexity cites rather than rewriting your own page for the model.

Most errors come from advice built for a different crawler or a different engine.

• Treating PerplexityBot as a training bot. Blocking it to protect content from training removes you from Perplexity's index while doing nothing about training, because Perplexity says it isn't used for that.
• Relying on robots.txt to stop Perplexity-User. Perplexity documents that it generally ignores robots.txt. Robots.txt is necessary for the index, not sufficient as a block.
• Checking robots.txt and stopping there. A 403 from your CDN is a silent, complete failure that robots.txt won't show.
• Shipping key content only through client-side JavaScript. If the price appears only after hydration, PerplexityBot doesn't see it.
• Rewriting pages in an "AI-friendly" style. C-SEO Bench and a 252,000-trial factorial found formatting and citation-oriented rewrites had little or negative effect.
• Adding schema, an llms.txt file or chunked content to get cited. No engine, Perplexity included, documents using llms.txt, and no evidence shows these steps earn citations. Existing schema does no harm, but it isn't a Perplexity lever.
• Cloaking or hidden instructions aimed at AI crawlers. Research on search agents shows manipulated content can sway answers once retrieved. It is illegitimate, it is a reputational risk, and you shouldn't do it.

Perplexity doesn't publish a ranking position for your site, so measure presence and accuracy, not rank.

• Ask the questions your buyers ask. Run them logged out and note which sources Perplexity numbers. The free Perplexity brand check shows what it says about you for a few buyer questions.
• Compare engines separately. Perplexity draws on its own index, so results can diverge from ChatGPT. In Otterly's 2026 experiment with two AI-generated sites, Perplexity cited them while Google's AI features almost never did. That was one pair of low-stakes sites, so read it as a reminder to measure per engine, not as a tactic. Our page on why Perplexity cites a site while ChatGPT doesn't goes further.
• Treat any prompt sample as a sample. Perplexity rewrites and routes queries internally. A tracker only sees answers to the prompts it asks.

If a check shows Perplexity never cites you, go back to the access steps before changing any content.