A Perplexity SEO audit you can run in an afternoon

What a Perplexity SEO audit checks

A Perplexity SEO audit checks whether Perplexity can read your site, whether it cites your pages in its answers, and why it cites other pages when it does not cite yours. It is a record you can repeat, not a score.

Perplexity answers each question by searching the web and linking the pages it used. Those numbered links are its citations. There is no ranking report to download and, as far as we could find, no Perplexity equivalent of Google’s Search Console. So the audit has to be built by hand from what Perplexity documents and what it shows in its answers.

The same shape works for any assistant. This version goes deeper on what is specific to Perplexity: its two user agents, its published IP addresses, and the way it shows its sources.

Part one: check that Perplexity can reach your site

Check access first, because a blocked crawler makes every other finding meaningless. Perplexity’s own documentation asks you to allow PerplexityBot in robots.txt and to permit requests from its published IP addresses.

Perplexity’s crawler documentation names two user agents, the names a program gives your server when it asks for a page. PerplexityBot builds the search results and, Perplexity says, is not used to train AI models. Perplexity-User fetches a page when a person asks a question, and Perplexity says it “generally ignores robots.txt rules” because a person asked for the fetch. The audit is mostly about PerplexityBot. Our page on the robots.txt check for AI crawlers explains how a block is read from the file.

The firewall row catches more sites than the robots.txt row. On 1 July 2025 Cloudflare said it was changing its default to block AI crawlers unless they pay creators. A site can have a perfect robots.txt and still turn Perplexity away at the door.

The IP row is there because a name can be faked. In August 2025 Cloudflare reported seeing Perplexity reach sites through an undeclared crawler that posed as Chrome on a Mac and used addresses outside Perplexity’s published ranges. Perplexity disputed that account. For the audit, the practical point is narrow: trust a request as PerplexityBot only when the name and the address both match.

Part two: write the buyer questions

Write 20 to 50 questions a buyer would type into Perplexity, in full sentences, and keep the wording fixed. The questions decide everything the audit finds, so this is the part to spend time on.

• Take the questions from real sources: sales calls, support emails, and the questions people ask in forums about your market.
• Write them the way people talk to Perplexity, with context. “Best invoicing tool for a two-person design studio” is a buyer question. “Invoicing software” is a keyword.
• Mix the stages. Include questions that compare options, questions about price, and questions about one specific job the product does.
• Keep questions that name your brand on a separate list. Perplexity cites you far more often when the buyer has already typed your name, and blending those in hides your losses.
• Give each question a topic, so the results can be read by topic as well as one by one.

Our guide to finding the questions buyers ask covers the sources in more depth, and our method page explains why branded and non-branded questions are always reported apart, and why every figure is a count over a stated total.

Part three: ask every question and log every citation

Ask each question in a fresh Perplexity conversation, signed out, and write down every page it cites. The log is the audit’s raw evidence, so record the answer itself, not just a yes or a no.

1. Open a new private window and go to Perplexity, so your past conversations do not shape the answer.
2. Paste the first question exactly as written.
3. Copy the full answer text into your log, with the date and time.
4. Open the list of sources and copy every cited address in order.
5. Note whether your brand is named in the answer text, and whether one of your pages is cited. These are two different results.
6. Note every other brand named, and which brand owns each cited page.
7. Repeat for every question, then ask the whole list again on another day before you trust any single result.

Expect some answers to be wrong. In March 2025 the Tow Center for Digital Journalism gave eight AI search tools excerpts from news articles and asked them to name the source. Perplexity answered 37% of those queries wrongly. That test was about news attribution, not product questions, but it means a citation to your page tells you Perplexity used it, not that it described you correctly. Read the sentence beside the link.

Part four: study the pages Perplexity cited instead

For every question you lost, open the pages Perplexity cited and write down what they have that your page does not. This is where the fix list comes from.

Look for concrete things you can count, not a general sense that the other page is better. Does it answer the question in its first lines? Does it give a price, a date, a number with a source? Does it compare named options? Is it a third-party page, such as a review site or a forum, that you cannot edit but might be listed on?

The only published test of what helps a page in Perplexity is the GEO paper, accepted at the KDD 2024 conference. In a test of 200 questions on Perplexity, adding quotations to a page raised the paper’s main measure of how much the answer drew on it by about 22%. Adding statistics raised a second measure by up to 37%. Keyword stuffing did about 10% worse on the main measure. The source text in that test was uploaded to Perplexity as files rather than found on the web, so treat the figures as a direction rather than a forecast.

The pages Perplexity cites instead of yours are the most useful thing the audit produces. They show what Perplexity trusts on that question.

Part five: turn the findings into a short fix list

End the audit with a short list ordered by what blocks the most, starting with access. A long list of small fixes is a list nobody finishes.

Our guide on how to get cited by Perplexity goes further into writing the passage a page needs.

What this audit cannot tell you

One round of the audit cannot tell you whether a change worked, and it cannot tell you why Perplexity ranks pages the way it does. Both need more than one afternoon.

• Answers vary. The same question can cite different pages on different days, so a single round mixes real gaps with chance.
• Your view is not every view. A signed-in person with a history, a location or a different model setting may see a different answer.
• A small list is noisy. With 20 questions, one changed answer moves your rate by five points.
• The live fetcher is outside your control. Perplexity-User generally ignores robots.txt, so blocking it there may not stop it.
• Causes stay hidden. You can see which page won, never the scores that made it win.

Agentcites runs parts two to five every week for a fee. It asks up to 50 buyer questions to Perplexity, ChatGPT, Gemini and DeepSeek, keeps every answer word for word, records every page cited, checks your robots.txt for five AI crawlers including PerplexityBot, and explains each lost question. It asks through each assistant’s API, so an answer can differ from what a signed-in person sees. It does not read your firewall or your server logs, so part one stays a job for you or your host. The plan is on the pricing page, at $149 a month per website.