How does Perplexity decide who to recommend?
The official docs do not say. Perplexity's Help Center (checked 2026-11-07) says only two things: it searches the web in real time, and every answer carries numbered citations linking to the original sources. How sources are chosen and ranked is not described on that page or in the crawler docs. Third-party lists of “ranking factors” have no verifiable basis that we could find, so we do not repeat them. What you can do is work from both ends: make sure its crawlers can reach you, then ask the questions yourself and read the sources.
| Question | Does the official documentation say? | What we do |
|---|---|---|
| Is the answer based on a live search? | Yes: it searches the web in real time | Your page must be readable at the moment of the request |
| Does each answer have citations? | Yes: numbered citations linking to the original sources | Use the source list to find gaps |
| How are sources chosen and ranked? | Not stated | Do not guess; ask 3 times and read the sources |
| Will being allowed in robots.txt get me recommended? | Not stated | Being allowed is a precondition, not a guarantee |
What is the difference between PerplexityBot and Perplexity-User?
Both are Perplexity crawlers, but they are triggered differently and call for different actions. The table below is from Perplexity's official crawler page (checked 2026-11-07; the page carries no date):
| Item | PerplexityBot | Perplexity-User |
|---|---|---|
| What the docs say it is for | Surfaces and links websites in Perplexity's search results | When a user asks a question, it may visit a page to answer and link to it |
| robots.txt | Can be managed with robots.txt; allowing it is recommended | Because the user requested the fetch, it generally ignores robots.txt rules |
| Training foundation models | Officially not used to crawl content for training foundation models | Officially not used for crawling or for collecting training content |
| IP list published by Perplexity | perplexity.ai/perplexitybot.json | perplexity.ai/perplexity-user.json |
| The list we read (2026-11-07) | creationTime 2026-10-28, 13 entries | creationTime 2026-10-28, 10 entries |
Two conclusions. First, PerplexityBot is the one you can control with robots.txt; it decides whether your pages can appear in Perplexity's search results, which is the precondition for being recommended. Second, whatever you write in robots.txt has little effect on Perplexity-User; the way to let it in is to not block it at your firewall. A side-by-side of all crawlers is in AI crawler user agents: a side-by-side table.
How should robots.txt be written so Perplexity can get in?
If you have no rules about them and no site-wide block, you do not need to write anything. To make it explicit, allow both names. Perplexity says changes can take up to 24 hours to take effect.
User-agent: PerplexityBot Allow: / User-agent: Perplexity-User Allow: /
To hide one directory and allow the rest:
# Perplexity can read everything except the private area User-agent: PerplexityBot Disallow: /private/ Allow: /
Do not block PerplexityBot because you “don't want Perplexity using my content”: its purpose is to put you in the search results, and blocking it keeps you out. Full templates and the trade-offs for each crawler are in How to write robots.txt for AI crawlers.
How do I confirm Perplexity can really reach my site?
A correct robots.txt does not mean it gets in. Firewalls, CDN bot blocking and login walls can all stop it regardless of robots.txt. Check in four steps:
1. Check the robots.txt layer with a tool
Enter your URL in AI crawler check and see whether PerplexityBot shows as allowed.
2. See whether the firewall or CDN blocks it separately
Perplexity's official page describes the Cloudflare approach: create a custom rule that matches the User-Agent plus the official IP list, set the action to Skip, and place it above every blocking or challenging rule. The page also warns that Bot Fight Mode cannot be skipped by custom rules and may still challenge these crawlers. For AWS WAF, create an IP set per crawler and an allow rule that requires both the IP set and the User-Agent to match, with a lower priority number than your blocking rules. Cloudflare's own AI crawler switches are covered in Does Cloudflare block AI crawlers by default? How to check and allow them.
3. Count requests in your server logs
Only requests from PerplexityBot or Perplexity-User in your logs prove it has been there. This line counts source IPs:
# Count requests from Perplexity in an nginx access log grep -E "PerplexityBot|Perplexity-User" /var/log/nginx/access.log | awk '{print $1}' | sort | uniq -c | sort -rn | head4. Check that the IP is an official one
Anyone can fake a User-Agent, so compare the IPs in your logs with the official JSON. The official page advises using the current IP lists and automating updates, because rule changes take time to propagate:
# Is this IP on Perplexity's published list? (replace the IP) curl -s https://www.perplexity.ai/perplexitybot.json | grep -c "203.0.113.5"
The list contains single addresses and ranges; ranges must be matched as ranges, so the line above only suits checking a single address.
Requesting your own site with a PerplexityBot User-Agent via curl only shows whether rules that match on User-Agent are in effect; it cannot show rules that verify the source IP (for example, allowing only official IPs). So treat your logs as the final word.
How should pages be written to be more likely cited?
The official docs do not state ranking rules, so below are basics that hold for all AI search, not conclusions specific to Perplexity:
| What to do | Why | Which guide |
|---|---|---|
| Body text is written directly in the HTML | Crawlers read the page at request time; content that appears only after scripts run is missed | See: Do AI crawlers run JavaScript? |
| One page answers one specific question, with the answer first | The sources people open are pages that answer directly | See: How do I measure AI visibility? |
| Get included in “best X tools” articles | Sources in answers are often list and review articles | See: How do I get into “best tools” articles? |
| Answer honestly in forums and say you are the author | Community posts are one common type of source | See: Why do AI answers love citing Reddit? |
The matching articles: Do AI crawlers render JavaScript?, How do you measure AI visibility? Four metrics and a table you can follow, How do I get into “best X tools” list articles?, Why do AI answers love to cite Reddit? What product teams can and shouldn't do.
How do I run my own test round in Perplexity?
1. Write down 5 questions buyers ask
Leave out your brand name and use the buyer's wording, for example “AI tool that turns a PDF into an editable PPT”.
2. Ask each question 3 times in new chats
Note three things: whether you are mentioned, who is recommended, and which URLs are in the sources. Perplexity's answers differ each time, so count how many of 3 runs include you.
3. Close gaps by source type
If the source is a list article, contact its author; if it is a comparison page, write an honest comparison on your own site; if it is your own page but you were not recommended, rewrite that page's first paragraph to answer the question directly. The full 6-step process is in How to get cited by Perplexity.
4. Re-test with the same questions after two to four weeks
There is no fixed time for a page to be read and cited. How to keep records is in How do you measure AI visibility? Four metrics and a table you can follow.
How is this different from ChatGPT, Claude, Gemini and Copilot?
Each has different crawlers, sources and tactics, and each has its own guide:
| Assistant | What this guide covers | Link |
|---|---|---|
| Perplexity | What each of the two crawlers does, and how to confirm they can get in | This guide; for the overall process see How to get Perplexity to cite your website |
| ChatGPT | OpenAI's crawlers and the pages it cites | How to get ChatGPT to recommend my product |
| Claude | Anthropic's crawlers and the pages it cites | How to get Claude to recommend my product |
| Gemini | Google's ecosystem | How to get Google Gemini to recommend my product |
| Copilot | Microsoft and Bing's ecosystem | How to get Microsoft Copilot to recommend my product |
Entry points to the four guides: How to get ChatGPT to recommend my product, How to get Claude to recommend my product, How to get Google Gemini to recommend my product, How to get Microsoft Copilot to recommend my product.
FAQ
What does Perplexity SEO involve?
Perplexity has not published ranking rules. What can be confirmed: allow PerplexityBot, make sure your firewall is not blocking either crawler, put the main text directly in the HTML, then ask buyer questions yourself and read the sources.
What happens if I block Perplexity-User?
The official page says that because the request comes from a user, it generally ignores robots.txt, so blocking it there does little. Blocking it requires a firewall, but then users who paste your link into Perplexity will not be able to open it either.
Will Perplexity train models on my pages?
The official page says PerplexityBot is not used to crawl content for training foundation models, and Perplexity-User is not used for crawling or for collecting training content. The page does not discuss other uses.
How do I know whether Perplexity has visited my site?
Look in your server logs for requests from PerplexityBot or Perplexity-User, then compare the source IPs with the official JSON lists, because a User-Agent can be faked.
Will allowing PerplexityBot guarantee a recommendation?
No. Being allowed is a precondition. The official docs do not say how ranking works after that, and we have no verifiable basis, so we promise nothing.
Sources
- Perplexity · Perplexity Crawlers (Checked 2026-11-07; the page carries no date)
- Perplexity · How does Perplexity work? (Checked 2026-11-07)
- Perplexity · PerplexityBot official IP list (Read on 2026-11-07, creationTime 2026-10-28)
- Perplexity · Perplexity-User official IP list (Read on 2026-11-07, creationTime 2026-10-28)
- kagentx evaluations: the data on the case study pages; each example gives its date.