Perplexity optimization

The short version

Perplexity is the most transparent of the assistants: it cites its sources on screen, every time. That makes it the easiest surface to work on — you can see exactly which pages decided the answer — and the least representative, because the others do not behave like it.

Optimise for it, measure it separately, and do not treat what works there as proof of what works in ChatGPT, Gemini or Claude.

Why Perplexity is different

Perplexity retrieves and cites by design. Ask it who to use for a service and you get a short answer with numbered sources beside it, each one clickable.

Two consequences, and both change the work.

The deciding pages are visible. No inference needed. The citation list is the list of pages that produced the answer. For every other assistant you are reasoning about what was probably read; here you are reading it.

Fresh, well-structured pages do disproportionately well. Retrieval happens at question time rather than from what a model remembers during training, so a page published last week can be cited this week.

That second point cuts both ways: it is easier to appear quickly, and easier to disappear when someone publishes something better.

What the citation list tells you

More than any other single artefact in this field. For each buying question, the sources shown are:

A to-do list. Those are the pages where your business needs to be described correctly. Not “similar sites” — those exact pages.

A map of your category’s authority. If two directories and one review platform decide every price question in your market, that is the shape of the market’s information supply, and it will be broadly true for the other assistants too even though you cannot see their working.

A competitor shortlist. The businesses named in those sources are the ones the surface currently trusts.

A gap list. Sources cited for your questions where your business does not appear at all are the highest-value work available, and they are usually duller than anyone hopes: a regional listing nobody has updated since 2021.

What works

Be in the sources it already cites

Ask your buying questions, write down every cited domain, and check whether your business appears on those pages, correctly and completely.

For most local and professional categories the same handful of directories and roundups appears again and again. Four to six sources typically account for most of the citations in a category.

Answer the question in the first two sentences

Retrieval favours passages that stand alone. A page whose answer arrives in paragraph four rarely supplies the quote, however good paragraph four is.

Practically: use the buyer’s question as the heading, and answer that heading completely in the two sentences beneath it.

Publish specifics

Numbers, dates, method, named examples, prices. A page that says something checkable gets cited over a page that says something impressive. This is not a stylistic preference — an extraction system has nothing to extract from a claim with no content.

Keep the page reachable and quick

This is retrieval, not memory. A page that times out is simply not in the running. Ordinary technical hygiene — speed, no crawler blocks, no JavaScript-only content — is the floor rather than the lever.

Keep facts consistent across the cited sources

Where your own site and three directories describe your services differently, there is less reason to rely on any of them. Consistency matters more here than polish, and it is most of the work.

What does not transfer

Do not assume a Perplexity result predicts ChatGPT, Gemini or Claude.

When four assistants were asked the same questions in one market, they named 63 businesses between them, and only four were named by all four. Forty-six appeared in a single assistant only. In the same measurement, shares for one business ranged from 100% of answers on one assistant down to 50% on another — identical questions, asked the same way, at the same time.

Reporting one engine as “AI visibility” is wrong about most of the picture, which is why every figure we publish is per assistant and none are averaged.

There is a practical trap in this for suppliers, worth naming: Perplexity is the easiest surface to show progress on, because citations are visible and fresh pages move quickly. A report that leads with Perplexity results and mentions the others briefly is telling you about the easiest quarter of the problem.

How to measure Perplexity visibility

Ask the buying questions, written the way buyers ask, grouped by intent — general recommendation, service, price, urgency, trust, comparison.

Ask each one several times. Answers vary between identical asks. Five repeats is our default; fewer and you cannot distinguish “usually named” from “named once”.

Record four things each time:

1. Whether you were named.
2. Who was named instead.
3. Every source cited, by domain and URL.
4. The full answer text, kept verbatim.

The third is the one unique to this surface and the most useful. The fourth is what makes any percentage checkable later.

Report as a share, not a yes or no, and separately from the other assistants.

The method is published in full, and the sample report shows the output with no form in front of it.

A worked example

From a measurement on one local market, business anonymised.

Baseline on Perplexity. Named in 50% of general recommendation answers, and 0% of price answers.

What the citations showed. Across the price questions, four sources accounted for almost every citation: two local directories, a review platform, and a regional “best in the city” roundup. The business appeared on one of the four, with an out-of-date service list and no prices.

What was done. Nothing to the homepage. The listings were corrected and completed on all four, using the same service names and price ranges as the business’s own pages. One page was rewritten so the price question is answered in the first two sentences.

What moved, on Perplexity. Price answers went from zero to being named in roughly a third. This surface moved first, which is typical: retrieval at question time picks up corrected pages sooner than systems answering from memory.

What moved elsewhere. Less, and later. The same work showed up on the other assistants over the following weeks, unevenly.

What was not claimed. That the listings caused it. The report stated what changed, when, and what moved after, and noted the movement was concentrated in the targeted question group rather than spread evenly.

Should you optimise for Perplexity specifically?

Only in the sense that you should start there.

Start there because the citation list removes the guesswork. Two hours asking your buying questions in Perplexity and recording the sources gives you a work list you can act on immediately, and that same list is a reasonable first approximation of what matters on the other surfaces.

Do not stop there because Perplexity’s audience is smaller than ChatGPT’s, and the correlation between them is weak. Winning the easiest surface and declaring victory is the most common failure in this category.

Never report it alone. One number per assistant, always.

Frequently asked

Can I pay Perplexity to be cited? No. There is no placement product for appearing inside answers.

Does blocking PerplexityBot matter? If you block it, your pages cannot be retrieved when it answers. Given that retrieval is how this surface works, blocking it is close to opting out.

How fast do changes show? Faster than the other assistants — days to weeks rather than weeks to months — because retrieval happens at question time.

Does being cited bring traffic? More than the others, because citations are links and they are visible. It is still a small number in absolute terms for most local businesses.

Is a Perplexity citation worth more than a ChatGPT mention? Different things. A citation brings a click; a mention brings a name someone remembers. Neither substitutes for the other.

Turning a citation list into a work plan

The citation list is the most actionable artefact in this field, and most people look at it once and do nothing with it. Here is the version that produces work.

Step 1. Collect properly. Ask each buying question five times. Record every cited URL, not just the domain — the specific page matters, because being on a directory is not the same as being on the page that gets cited.

Step 2. Count. Which domains appear most often across your question set? Four to six usually account for most citations in a category.

Step 3. Check yourself on each. Three states, and they need different work:

Step 4. Compare against your own site. Every difference between what a cited source says and what you say is a contradiction a system can find.

Step 5. Re-measure after four weeks. Not two: listings take time to be re-read even on a retrieval-first surface.

What the sources tell you about your category

Beyond your own to-do list, the citation pattern is a free market study.

If two or three domains dominate, your category’s information supply is concentrated. Getting onto those pages is most of the available upside, and it is a short project.

If citations are scattered across twenty domains, no single source decides anything, and the lever moves towards your own publishing — being the page worth citing rather than being on someone else’s.

If the cited pages are mostly old, there is an opening. A current, specific page in a category served by 2021 roundups tends to get picked up quickly here, because retrieval happens at question time.

If your competitors publish and you do not, that shows up in the citation list before it shows up in your share.

Questions people actually ask

How fast does a new page get cited? Days to weeks, which is fast compared with the other assistants. Retrieval happens at question time, so there is no waiting for a model to be retrained.

Can I lose a citation as quickly? Yes. A better page published by someone else takes the slot, and nothing warns you. This is the trade-off for moving quickly in the first place.

Do I need to be on page one of Google to be cited here? No. Perplexity retrieves independently, and pages that rank poorly get cited regularly when they answer the question directly.

Should I write pages aimed at Perplexity? Write pages that answer buying questions in their first two sentences. That works here, and everywhere else.

Is a Perplexity citation worth more than a ChatGPT mention? They are different things. A citation is a link and a click; a mention is a name somebody remembers. Neither substitutes for the other, and a report treating them as equivalent is not measuring carefully.

Where to start

Ask your five most valuable buying questions in Perplexity today and write down every source it cites. That list is the most actionable thing you will get for free in this category.

Then measure the rest properly: the free check covers four assistants including Perplexity and emails you what came back. Related reading: Gemini and Claude visibility, how to get recommended by ChatGPT, and the full guide.

One thing to do tomorrow

Ask your five most valuable buying questions in Perplexity and write down every source it cites. That list is the most actionable thing available to you for free in this category: it is the set of pages where your business needs to be described correctly, named explicitly rather than inferred.