If you want to optimise content for Perplexity, start by accepting that it is not a ranking board you climb. Perplexity is a retrieval-and-synthesis pipeline: it turns one typed question into several searches, pulls back a shortlist of documents, ranks the passages inside them against the original question, and writes a short answer with numbered citations pointing back at what it used. Optimising for it means handing that pipeline something clean to lift.
What actually happens when someone asks Perplexity a question?
Perplexity rewrites the question into multiple search queries, retrieves candidate documents, reranks the passages inside those documents against the original question, and then generates a cited answer from the highest-scoring passages. There are four stages, and you can influence three of them.
The reason this matters is that each stage throws work away. Retrieval discards most of the web. Reranking discards most of what retrieval returned. Synthesis quotes only a handful of the surviving passages. A page that is technically excellent but written as one continuous argument tends to die at the reranking stage, because no single chunk of it stands on its own.
Where your leverage sits, stage by stage
- Query expansion — you cannot touch it, but you can anticipate it. A question like "best CRM for a small Indian business" plausibly expands into pricing, integration, GST-invoicing and migration sub-questions. Pages that cover only the headline question tend to lose the expanded ones.
- Retrieval — this is where crawl access, index coverage and plain topical match decide everything. If you have pages that Google is refusing to index, the same underlying problems usually block AI retrieval too; our guide to indexing problems covers the usual causes.
- Reranking — this is passage design, and it is the part most sites ignore completely.
- Synthesis — specific, attributable sentences survive summarisation. Vague ones get paraphrased into nothing and the citation goes to whoever was specific.
Why does Perplexity cite some pages and skip others that rank well?
Because the unit of selection is the passage, not the page. A page can hold the top organic position for a query and still contribute nothing quotable, if its answer is spread over six paragraphs, three subheadings and a video.
Test your own pages this way: copy a single paragraph out of the article, paste it into a blank document, and read it as a stranger. If it still answers a real question without the surrounding context — no dangling "this", no "as we mentioned above", no pronoun pointing at a heading — it is a candidate. If it does not, it is invisible to the reranker no matter how well the page performs.
Does Perplexity need to crawl your site directly?
Yes, and it uses named agents to do it. Perplexity documents a crawler for building its search index and a separate user-triggered fetcher that pulls a page because someone's question needed it right then. Blocking the crawler in robots.txt removes you from the index it retrieves against, which is the fastest way to disappear from a surface you were hoping to appear on.
Two practical checks follow from that:
- Read your own robots.txt. Blanket rules written years ago to keep scrapers out now block the retrieval systems you want. Decide deliberately, agent by agent, rather than inheriting somebody's template.
- Serve real HTML. An answer-time fetch is not a patient rendering pipeline. If your headings, tables and body copy only exist after a client-side framework hydrates, assume they are not there. Server-side rendering is the fix, and it is worth raising with whoever handles your web development before you rewrite a word of copy.
The related question of whether a machine-readable summary file helps is genuinely unsettled — we work through the evidence in our post on llms.txt.
How is Perplexity different from AI Overviews and ChatGPT?
They differ mainly in how the candidate source set is built, which changes what you should optimise first.
| Surface | How the source set is built | What you can influence |
|---|---|---|
| Perplexity | Its own crawl plus search partners, retrieved fresh for each question after the query is rewritten | Crawl access, passage clarity, topical depth on the sub-questions |
| Google AI Overviews | Grounded in Google Search's own results for the query and its variations | Classic Search eligibility, snippet directives, passage structure |
| ChatGPT with search | A web index plus its own fetcher, invoked only when the model decides a search is needed | Crawl access for its named agents, plus how clearly your brand is defined as an entity |
The overlap is large enough that one body of work serves all three, which is the argument made in our answer engine optimisation guide. The differences are large enough that you should not assume a page cited by one will be cited by the others.
Free consultation
Want to know whether Perplexity can even see your site?
We will check your robots rules, your rendering and how your key pages break into passages, and tell you what is blocking retrieval. No obligation, and you keep the findings either way.
What does a quotable passage actually look like?
A quotable passage answers one specific question completely in two or three sentences and names something concrete — a mechanism, a setting, a number that comes from a real source, a trade-off. Everything else is elaboration and belongs underneath.
Compare these two openings for a section on GST invoicing inside a billing tool:
- Not liftable: "There are several considerations when it comes to compliance, and businesses should evaluate their options carefully before deciding."
- Liftable: "A billing tool is usable for Indian GST invoicing if it can store an HSN or SAC code against every line item, split tax into CGST, SGST and IGST based on the place of supply, and export in a format your accountant can reconcile. Anything missing one of those three has to be patched manually every month."
The second one names the actual requirement. It can be lifted whole, and the citation has to come with it because the specifics are yours. Writing at that density across a whole site is slow, which is why teams often pair a subject-matter expert with an assisted drafting process — the approach behind our AI content services, where the model does structure and the human supplies the specifics.
How do you tell whether Perplexity is citing you?
There is no rank tracker for it, so you measure with three cruder instruments instead.
- Prompt testing. Keep a fixed list of twenty questions a real buyer would type, run them monthly, and record which domains get cited. It is manual and it is the most honest signal you have.
- Referral traffic. Perplexity sends real referrals. Segment them in analytics and treat them as a separate acquisition channel rather than letting them sit inside "Other".
- Server logs. Filter for the named agents. If they are not hitting you at all, no amount of content work will help until that is fixed.
Do not expect these numbers to be large early on. The point of watching them is direction, not volume — and the same discipline applies to being cited by other assistants, which we cover in how to get your business cited by ChatGPT.
Where to start if you only have a week
Pick your five highest-intent pages. Confirm the crawlers can reach them and that the content exists in the raw HTML. Rewrite the first paragraph under every heading so it answers the heading directly. Add one concrete specific to each section that nobody else on page one has bothered to state. Then leave it alone and re-run your prompt list in a month.
That sequence is deliberately small. Most sites do not have a Perplexity problem — they have a clarity problem that Perplexity happens to expose, and the same fixes lift ordinary organic performance too. If you would rather have it done as one structured pass, that work sits inside our SEO services.
Free consultation
Get your top pages rewritten for passage retrieval
Send us five URLs and the questions you want to own. We will come back with the specific rewrites that make each section liftable — no obligation to take the work further.