GEO INSIGHT
Perplexity Optimization — Which Sources Are Cited and What Should You Verify?
This guide uses Perplexity's official documentation to explain access paths and controls, what to inspect in the answer interface, and what to do after each result.
Perplexity is easier to observe—and therefore easy to misread
Perplexity often displays sources with its answers. Because you can see which documents support a response, the observation cost is lower than on many other platforms. This is why brands often begin citation checks with Perplexity.
The problem comes next. Seeing your URL in one answer does not mean the result will persist. A small wording change or a different run time can change the source set. If a single screenshot is reported as performance, the next month's variation may be impossible to explain.
The two paths Perplexity uses to reach a page
Perplexity documents two agents with different purposes. Blocking both as if they were one can produce a result you did not intend.
PerplexityBot and Perplexity-User (verified 2026-08-23)
| Category | PerplexityBot | Perplexity-User |
|---|---|---|
| Purpose | Indexing websites for display in Perplexity search results | Visiting a page when a user asks a question |
| Model training | Documented as not used for training | Documented as not used for training |
| robots.txt | Observed. If blocked, full or partial page text is not indexed | Developer documentation states that rules generally do not apply because the visit is user initiated |
| If blocked | The site does not appear in Perplexity search results | Perplexity cannot access the page while answering |
| IP list | perplexity.com/perplexitybot.json | perplexity.com/perplexity-user.json |
(1)
A blocked page may not disappear completely
Blocking through robots.txt applies to page-body indexing, not necessarily to the page's existence. If a brand needs to hide a page completely, robots.txt alone is not sufficient. Conversely, allowing access is not sufficient for citation if the body text cannot be read. (2)
The two official documents use different wording
Perplexity's documents differ on whether robots.txt applies to a user-requested visit. The developer documentation says Perplexity-User generally ignores robots.txt because the request is initiated by a user. The Help Center page, updated 2026-07-16, says a feature that summarized previously blocked URLs was discontinued to prevent abuse. (1, 2)
User-agent: PerplexityBot
Allow: /
User-agent: Perplexity-User
Allow: /
Sitemap: https://example.com/sitemap.xmlA user-agent string can be spoofed. If a request appears as PerplexityBot in server logs, compare its IP with the range published in the JSON file before treating it as a verified visit. ChatGPT uses a similar separation: OpenAI documents different agents for search visibility and model training and notes that user-initiated visits may not follow robots.txt. See How Does GPT Choose Businesses to Cite?. (1, 3)
Five interface checks and the next action for each
After access conditions are documented, inspect the actual answer. Check all five items together and define the next action for each result.
Observation items and next actions
| No. | What to check | What to observe | What to do next |
|---|---|---|---|
| 01 | Answer body | How the brand and core facts are described; whether the context is positive, neutral, or negative | If the explanation is inaccurate, find the document behind the claim and publish corrective information on an official page |
| 02 | Source URL | Whether the source is an official domain, a third party, or absent | If only third-party sources appear, first check whether an official page covers the same information |
| 03 | Linked statement | Which claim in the answer is connected to the source | If your page is connected to the wrong claim, revise the heading and first sentence of that paragraph |
| 04 | Question conditions | Exact wording, language, time, and search conditions | Discard observations that cannot be compared because the conditions were not recorded |
| 05 | Competitors | Brands that appear or are cited together, and in which order | If a competitor document is repeatedly cited, list the information it contains that yours does not |
Record source results in four states
A binary appeared/did-not-appear record does not tell you what to do next. Four source states connect each outcome to a different action.
Four source states
| State | Classification | What it means | What to do |
|---|---|---|---|
| Official source | A URL on the brand's official domain is linked as a source | The official page is being used as evidence | Identify the cited page and check whether it provides a useful next step |
| Third-party source | A media, specialist publication, or other external URL is linked | An external document about the brand is being used as evidence | Check whether an official original exists on the same topic and whether it is indexed |
| Mention only | The brand name appears but no source is linked | The name is known, but there may be no suitable evidence document | Create a citable document with a definition, figures, a process, or a comparison table |
| Not observed | Neither the brand name nor an official source appears | A document connecting the brand to the question may be missing | Check access first; if access is healthy, consider a new document |
Review the counts before combining them into a rate. If 20 questions produce 3 official sources, 6 third-party sources, 4 mentions only, and 7 not observed, the immediate task is not reporting a 15% citation rate. It is checking whether documents exist for the seven unanswered questions.
One answer is one observation
Perplexity can attach different sources to the same question across runs. A screenshot alone is not enough; save the result in a form that can be compared with the next run.
- Full question — preserve the exact wording and the question-set version
- Full response — save it without summarizing
- Every source URL and its reachability result
- Source state — official / third party / mention only / not observed
- Co-occurring brands and their order
- Measurement time plus language and region conditions
- Response status — completed / error / interrupted; do not count an error as non-visibility
The formulas and small-sample limits behind citation rate are explained in Five GEO Performance Metrics. The platform-level measurement procedure is in How to Measure AI Citation Rates.
Do not treat a Perplexity result as the whole AI search market
Strong citation in Perplexity does not imply the same outcome elsewhere. This is not a matter of preference; official documentation describes different access paths and source-display behavior by platform.
- Access paths differ — Perplexity separates PerplexityBot and Perplexity-User, while OpenAI separates OAI-SearchBot, GPTBot, and ChatGPT-User. Their controls differ.
- Source displays differ — inline citations, separate source areas, and answers with no displayed source are mixed across platforms, so the denominator behind the same 'citation rate' label may differ.
- Blocking has different effects — Perplexity states that a blocked page may still have its domain, title, and summary indexed. (1, 2, 3)
Compare platform-level checks in GEO Strategy by AI Platform. A single blended average hides platform differences and makes the next action harder to choose.
Conclusion — separate what is visible from what is proven
A visible source in Perplexity means the result is easy to observe; it does not mean performance has been proven. A URL can appear in one answer and disappear under another question or at another time.
Follow a sequence. First, verify access: confirm that PerplexityBot is not blocked, look for real server-log visits, and compare the IP with the published range. Next, inspect the five interface items and classify the source into one of four states. Finally, repeat the run under the same conditions and document the differences.
This sequence lets you distinguish a content problem from an access problem when results are weak. Without that distinction, the next measurement cannot explain why a change worked or failed.
Frequently asked questions
Which sources does Perplexity cite?
Perplexity does not publish an official rule describing which documents it selects or in what order. It does document the access path: PerplexityBot indexes pages for Perplexity search, and Perplexity-User visits pages for user questions. You can verify whether your page is accessible, but actual citation must be observed repeatedly with a fixed question set.
Does blocking PerplexityBot remove a page from Perplexity completely?
The page body is not indexed, but the page may not disappear completely. Perplexity states that it may still index the domain, headline, and a brief factual summary of a blocked page. To appear in Perplexity search results, do not block PerplexityBot.
Can I opt out of AI training and still appear in Perplexity search?
Perplexity documents both agents as not collecting data for generative foundation-model training, so it does not provide a separate Perplexity training-versus-search setting. ChatGPT differs: OpenAI separates training agent GPTBot from search agent OAI-SearchBot, allowing one to be blocked while the other remains allowed.
Does robots.txt apply when a user pastes a URL and asks Perplexity to summarize it?
Perplexity's two official documents use different wording. The developer documentation says Perplexity-User generally ignores robots.txt for user-initiated requests, while a Help Center page updated 2026-07-16 says a feature that summarized blocked URLs was discontinued to prevent abuse. Use access control, not robots.txt alone, for pages that must remain private.
How can I verify a PerplexityBot visit in server logs?
Find requests using the PerplexityBot user agent and compare the request IP with the ranges in perplexity.com/perplexitybot.json. User-agent strings can be spoofed, so the IP check is required. Verify Perplexity-User against perplexity.com/perplexity-user.json in the same way.
Does a high Perplexity citation rate imply strong visibility in other AI search platforms?
No. Platforms document different access routes, source-display behavior, and blocking effects. Even the denominator behind a citation rate changes when a platform does not display sources. Report each platform separately instead of combining them into one average.
Conclusion
What should you do to earn citations in Perplexity?
Check access for PerplexityBot and Perplexity-User in robots.txt, and record brand mentions separately from official-URL citations. Save the exact question, full response, source URLs, and measurement time so each result can be compared with the next one.
If only third-party sources appear, strengthen the official page covering the same topic. If the result is mention only, revise the relevant heading and first sentence. Then rerun the same question under the same conditions and check whether an official source becomes connected.