How to get cited by Perplexity

Perplexity tells you exactly why it said what it said. Every claim carries a numbered citation, and you can click each one. No other major engine is this transparent, which makes it the single best place to learn how AI retrieval actually behaves before you spend money on the ones that hide their working.

It's also, in our scan data, the engine where a small site can win fastest. Not because it's easy, but because it's live. Perplexity searches on nearly every query rather than answering from memory, and that means the door is open right now instead of at the next model refresh.

How does Perplexity's retrieval actually work?

It decomposes your question into sub-queries, runs a live search for each one, retrieves a set of pages, then generates an answer that's constrained to what those pages say and cites them inline.

Roughly:

1. Query understanding. "What's the best AI visibility tracker for a small SaaS?" becomes several sub-questions: what is an AI visibility tracker, which tools exist, what do they cost, which suit small teams.
2. Retrieval. Each sub-query hits Perplexity's index (built by PerplexityBot) and other retrieval sources. It's pulling live pages, not recalling training data.
3. Reranking. Retrieved pages get scored on relevance, authority and freshness, and the top handful go into context.
4. Grounded generation. The model writes an answer from those pages, attaching a numbered citation to each claim.

Step 2 is where you win or lose, and step 1 is why. You are not competing for the user's question. You are competing for the sub-questions. A page that comprehensively answers "how much do AI visibility tools cost" can get cited in an answer about which tool to buy, because that was one of the decomposed queries. This is the same fan-out logic Google uses in AI Mode, and it rewards the same thing: coverage of a topic, not one page targeting one keyword.

Why do freshness and crawlability dominate here?

Because Perplexity is a search engine with a language model on top, not a language model that occasionally searches. Retrieval is the default path, so anything that blocks or deprioritises retrieval is fatal, and anything that boosts it pays off immediately.

Compare the three engines on what they weight:

The bottom row is the one people get wrong. Three different crawlers. Allowing Googlebot does nothing for Perplexity.

Allow PerplexityBot

Perplexity runs two agents and they do different jobs:

• PerplexityBot — the indexing crawler. Blocked means unindexed means never cited.
• Perplexity-User — fetches a page live when a user's specific request requires it.

Allow them explicitly:

# robots.txt
User-agent: PerplexityBot
Allow: /

User-agent: Perplexity-User
Allow: /

Sitemap: https://example.com/sitemap.xml

Then go check your CDN. This is the step people skip. Cloudflare, Fastly and Akamai all ship AI-crawler blocking that operates above robots.txt, and a lot of sites turned it on in the 2024 anti-scraping panic and never reviewed the decision. Your robots.txt can say "come in" while your edge returns 403 to every AI crawler that knocks. Fetch a page with PerplexityBot in the user-agent string and see what you actually get back:

curl -A "PerplexityBot/1.0" -sI https://example.com/your-key-page

A 200 is what you want. A 403 or a challenge page means you've been invisible and didn't know.

Then be fresh, honestly

Perplexity's reranker likes recent pages, and it likes them for a good reason: for most questions, a 2026 answer beats a 2023 answer. That's exploitable in a cheap way and a real way, and only one of them works.

The cheap way is bumping dateModified and changing three words. Systems get better at spotting that, and it does nothing for the actual relevance score, because the passage the reranker is judging didn't change.

The real way is a maintenance cycle on the pages that matter. Every quarter: update the prices, add whatever launched, remove whatever died, add a "last verified" line with an actual date. That improves the passage and the freshness signal, and it's about two hours per page.

What kind of page gets cited?

The kind that answers a specific question directly, in plain language, near the top. Perplexity cites Reddit threads and Stack Overflow answers constantly, and it's worth being honest about why: those pages answer the question in the first sentence, without a value proposition in the way.

Your marketing pages usually don't. That's the gap.

The four "yes" categories you can actually create: comparison pages, documentation, original data, and honest listicles. Original data is the strongest because it's uncopyable. If you publish a number nobody else has, and someone asks a question that number answers, there is exactly one source and it's you. We publish scan-derived stats for this reason, and they get cited back to us more than anything else we write.

Comparison pages are the fastest. A page that lays out competitors fairly, with real prices and real limitations, is retrievable and useful, and it will get cited even in answers where you're not the recommendation. Being in the answer at all is the point. (Ours are at /compare — including honest ones like Spottlo vs Profound, where the relevant fact is that Profound's $99 Starter tier tracks ChatGPT only and you need the $399 Growth tier for three engines.)

Is Perplexity big enough to bother with?

Smaller than ChatGPT, and a lot more valuable per user than the headcount implies.

The last official figure was 780 million queries in May 2025, growing 20%+ month over month. Against ChatGPT's 900 million weekly actives that looks modest.

But the user is different. People go to Perplexity to research, deliberately, instead of to chat. The queries are commercial, the answers carry citations, and the citations get clicked, because that's the whole product. Compare that to Google, where only 1% of users click a link inside an AI summary. Perplexity citations are a functioning traffic channel, not just a brand impression.

And it's the best diagnostic tool you have. Ask Perplexity your top buyer question, open the citation list, and read every page it pulled. That list is a plain-language brief for what you need to publish. No other engine hands you that.

What to do next

1. Curl your own site as PerplexityBot using the command above. If you get anything other than a 200, fix that before doing anything else on this list.
2. Ask Perplexity your five biggest buyer questions and read every cited source. Note which are forums, which are listicles, which are competitors. That's your content gap, itemised.
3. Publish or fix one comparison page covering the real alternatives with real prices, including the ones where you lose.
4. Set a quarterly refresh cycle on your top ten pages. Real updates, real dates, no dateModified theatre.
5. Track it over time. Perplexity answers vary run to run, so one check is an anecdote. Spottlo runs a fixed 25-prompt set weekly across Perplexity, ChatGPT, Gemini and Google AI Overviews and reports mention rate and share of voice. Free report, no signup, if you just want the baseline.