How to Verify PerplexityBot (2026): IP Lists, Log Check

Perplexity publishes two IP lists, 13 and 10 addresses. Check PerplexityBot requests in your log with a script, and see what 46 requests showed.

Published
10 min read
  • perplexitybot
  • perplexity-user
  • verify crawler
  • ip ranges
  • user agent
  • server logs

To verify PerplexityBot, compare the requesting IP address with the two JSON files Perplexity publishes: 13 addresses for PerplexityBot and 10 for Perplexity-User. In our server logs from 19 to 31 October 2026, 12 of 46 requests naming PerplexityBot (26 percent) came from outside both files, but both files were created on 28 October, so that count includes days before the lists existed. In our fake bot study for 2 to 8 November 2026, 358 of 359 requests naming PerplexityBot (99.7 percent) came from inside the published ranges. The user agent is only a claim, and the address decides.

As of 9 November 2026 for the lists and the study; the log counts are from 19 to 31 October. The numbers describe one small site, not the web, and the lists are a snapshot (limits below).

Picture two lines in your access log, minutes apart. Both say PerplexityBot. Only one comes from an address Perplexity lists:

203.0.113.5 - - [01/Nov/2026:10:00:02 +0000] "GET /_next/static/app.js HTTP/1.1" 404 0 "-" "Mozilla/5.0 (compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot)"
192.0.2.10 - - [01/Nov/2026:10:00:04 +0000] "GET /docs HTTP/1.1" 200 900 "-" "Mozilla/5.0 (compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot)"

Both lines are invented examples in the real log format, with documentation addresses. The user-agent strings are identical, so nothing in the line tells you which one to trust. This guide shows what Perplexity documents, what the 46 requests looked like, and a script that sorts your own log.

What does Perplexity document for PerplexityBot and Perplexity-User?

Perplexity documents two user agents, PerplexityBot and Perplexity-User, each with its own published IP file. The details come from Perplexity's crawler page, read on 9 November 2026.

PerplexityBot Perplexity-User
What Perplexity says it does "designed to surface and link websites in search results on Perplexity"; not used to crawl content for AI foundation models "supports user actions within Perplexity"; "it might visit a web page" when a user asks a question
Example user agent in the docs Mozilla/5.0 (compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot) Mozilla/5.0 (compatible; Perplexity-User/1.0; +https://perplexity.ai/perplexity-user)
robots.txt Controlled by robots.txt tags; changes "may take up to 24 hours" "this fetcher generally ignores robots.txt rules"
IP file perplexitybot.json: 13 entries perplexity-user.json: 10 entries

Perplexity adds that "the version number may change" in both strings. The page does not mention reverse DNS for either bot, so the published JSON files are the only documented way to verify an address. For a Cloudflare or AWS firewall rule, the page advises combining the user-agent match with the IP check and matching each user agent against its own IP file.

The JSON files themselves are plain. On 9 November 2026 both carried the creation time 2026-10-28T16:48:51, contained only IPv4 entries, and every entry was a single address (a /32 prefix). The two files do not overlap, so a match tells you which of the two bots the address belongs to. The per-bot pages cover the rest: PerplexityBot, Perplexity-User and the Perplexity operator page.

So what does a request outside those 23 addresses look like in practice?

How many "PerplexityBot" requests came from outside the lists?

In our raw access archive for 19 to 31 October 2026, 46 requests named PerplexityBot, 34 came from inside the two lists and 12 from outside, which is 26.1 percent. Both lists were created on 28 October, nine days into the window, so this counts requests against lists that did not exist for the first part of it. We recorded no Perplexity-User requests in that period.

Method and limits. Period: 19 to 31 October 2026 (13 days). Source: this site's raw web-server access archive, every GET request whose user agent contains the token PerplexityBot, our own test tools excluded. "Inside" means the client address was in perplexitybot.json or perplexity-user.json, both downloaded on 1 November 2026 at 02:49 UTC. The lists were created on 28 October, so an address that was listed earlier and later dropped would count as outside here. We did not keep earlier copies and cannot rule that out. The sample is small: read the shares as counts.

The two groups differ on every axis we split by.

Inside the lists Outside the lists
Requests 34 12
Status 200 34 7
Status 404 0 5
robots.txt requests 10 1
Page requests 24 5
/_next asset requests 0 6
Long user agent (with AppleWebKit) 16 2
Short user agent (Mozilla/5.0 (compatible; ...)) 18 10

Three differences stand out. Every request from a listed address was a robots.txt or page request that returned 200, while all 5 of the 404 answers and all 6 requests for bundled /_next files came from outside. Listed addresses fetched robots.txt and pages; only unlisted ones asked for script files, and some of those requests missed.

The user-agent form is a weaker signal than it looks. Ten of the 12 outside requests used the short form, but 18 of the 28 short-form requests came from listed addresses, so the short form does not mark a request as foreign. That fits the lead: a string cannot verify a request in either direction. The docs show the short form for PerplexityBot, and the long form with AppleWebKit/537.36 (KHTML, like Gecko; compatible; ...) appears here too, from listed addresses.

Our weekly fake-bot study counted 359 requests claiming PerplexityBot in the week of 2 to 8 November 2026, with 358 from the published ranges (99.7 percent), and 96.8 percent over its newest three weeks. That is a later window and a different list snapshot (the study judges a week with the lists of the following Monday). We did not re-run the October window against the current lists. The two figures are not meant to match. What they share is the lesson: a request that names PerplexityBot can still come from an unlisted address, and a user-agent allow list lets it through.

An outside request is not proof of a forgery. The next question is what you can prove.

What does a request outside the list prove?

A request from an address outside Perplexity's lists proves only that the address is not on the current lists. It does not show who sent it, and it does not show bad intent.

Several explanations are possible, and a log cannot separate them. The list may lag behind new addresses. The sender may be someone testing with a copied string. Or the sender may be a client that Perplexity does not list. In August 2025 Cloudflare published a different case: its blog post said it observed a crawler that "utilized multiple IPs not listed in Perplexity's official IP range" and sent a generic Chrome-on-macOS string instead of a Perplexity one. Perplexity's post of the same day, which addresses the Cloudflare report, says: "User-driven agents only act when users make specific requests." It also says: "This is the fundamental difference between a user agent and a bot." We make no judgement between the two accounts. Both were published on 4 August 2025.

For your logs the consequence is narrow. A request with a browser-like string and no Perplexity token cannot be tied to Perplexity by any user-agent rule. Perplexity's crawler page lists PerplexityBot and Perplexity-User and does not describe any other Perplexity user agent, so a browser or agent product is not documented there. The IP check applies only where the string names Perplexity.

How do you run that check on your own log?

How do you check your own log against the lists?

Extract the addresses of requests that name a Perplexity bot, then compare each one with both JSON files using Python's ipaddress module. It takes three commands and no packages.

# 1. Requests per bot token
grep -ioE 'perplexitybot|perplexity-user' access.log | tr 'A-Z' 'a-z' | sort | uniq -c

# 2. Unique client addresses that used a Perplexity name
grep -E 'PerplexityBot|Perplexity-User' access.log | awk '{print $1}' | sort -u > ips.txt

# 3. Compare them with both published lists
python3 verify_perplexity.py < ips.txt

Behind a CDN or proxy, field 1 is the proxy's address unless your log format records the client address. Fix that first or every line will come back "NOT in list". Save this as verify_perplexity.py:

# verify_perplexity.py: which Perplexity list, if any, contains each IP address read from stdin?
import ipaddress, json, sys, urllib.request

SOURCES = sys.argv[1:] or [
    "https://www.perplexity.ai/perplexitybot.json",
    "https://www.perplexity.ai/perplexity-user.json",
]
nets = []
for url in SOURCES:
    req = urllib.request.Request(url, headers={"User-Agent": "verify-crawler"})
    for p in json.load(urllib.request.urlopen(req))["prefixes"]:
        nets.append((url.rsplit("/", 1)[-1], ipaddress.ip_network(p.get("ipv4Prefix") or p.get("ipv6Prefix"))))

for line in sys.stdin:
    text = line.strip()
    if not text:
        continue
    ip = ipaddress.ip_address(text)
    hits = sorted({name for name, n in nets if n.version == ip.version and ip in n})
    print(text, ", ".join(hits) if hits else "NOT in list")

We tested it without touching any real address. We wrote two local list files in the same format, one listing the address 192.0.2.10 and one listing 198.51.100.0/24, wrote a five-line sample log of our own with those addresses, and ran steps 2 and 3 with the files passed as arguments. Step 2 produced three addresses, and the script printed:

192.0.2.10 perplexitybot.json
198.51.100.77 perplexity-user.json
203.0.113.5 NOT in list

The address in the PerplexityBot file matched that file, the one in the Perplexity-User range matched the other, and the unlisted address was reported as not in any list. The script also loaded both real URLs on 1 November 2026 and reported 192.0.2.1 as not in any list. It fetches the lists on every run, which matters because Perplexity says the addresses are "updated regularly".

A string you cannot place goes into the user-agent lookup. Which result should you act on?

Should you block the requests that fail the check?

Blocking by user agent alone is the weak option. A rule that matches PerplexityBot and allows the request lets any client with the string in, and a rule that blocks the string stops only the honest senders. A rule that requires a matching address as well, as Perplexity's WAF advice describes, removes the cheap forgery.

To keep out only the requests that fail the check, block by address in your firewall or CDN. To control the real bots, use robots.txt: Perplexity documents robots.txt tags for PerplexityBot, while Perplexity-User generally ignores them. Both paths are covered in blocking AI crawlers and, for Cloudflare rules, in blocking AI bots on Cloudflare. Our robots.txt test for AI assistants shows how to confirm what actually happens on your site. The same method for Anthropic's crawlers is in the ClaudeBot detection guide, and for other operators' lists see GPTBot IP ranges and Bingbot IP ranges. Signature-based verification, which Perplexity's page does not document, is covered in Web Bot Auth.

Start with the counts: run step 3 on a week of your log and see how many lines say "NOT in list". The crawler IP checker gives the same answer for one address in a few seconds.

Common follow-up questions

Does PerplexityBot respect robots.txt?

Perplexity's crawler page, read on 9 November 2026, says PerplexityBot is controlled by robots.txt tags, with changes taking up to 24 hours. It says Perplexity-User generally ignores robots.txt rules. We did not test either claim beyond the log counts above.

What is the PerplexityBot user agent?

Perplexity's documentation gives Mozilla/5.0 (compatible; PerplexityBot/1.0; +https://perplexity.ai/perplexitybot) as an example and says the version number may change. In our logs a longer form with AppleWebKit/537.36 (KHTML, like Gecko; compatible; ...) also appeared, from listed addresses. Match on the token, not the whole string.

Where is the PerplexityBot IP range?

Perplexity publishes it as a JSON file at perplexity.ai/perplexitybot.json, with a separate file for Perplexity-User. On 9 November 2026 the PerplexityBot file held 13 single-address entries and the Perplexity-User file held 10. There is no CIDR block to copy, so compare against the file rather than a range.

Does Perplexity support reverse DNS verification?

Perplexity's crawler page does not mention reverse DNS, so we cannot say it works. The documented method is the IP lists. Our tool and script therefore use only the lists, and we did not test reverse lookups against Perplexity addresses.

Is Perplexity-User the same as PerplexityBot?

No. Perplexity describes PerplexityBot as the search crawler and Perplexity-User as the fetcher for questions people ask, with separate IP files and different robots.txt behavior. In our 19 to 31 October 2026 logs we recorded 46 PerplexityBot requests and none for Perplexity-User.

Sources

Checked 9 November 2026.