What Is Perplexity and How Does It Pick Sources?
Perplexity is an AI answer engine that searches the web as you ask and shows numbered citations 2. Its engineering post on its search API says it splits pages into self-contained passages, retrieves them by keyword and meaning, then reranks them 3. Only 28.6% of its cited links ranked in Google's top 10 in one study 4. Allowing PerplexityBot makes your site eligible 1.
- Perplexity searches the web in real time and attaches numbered citations to every answer, according to its help centre 2.
- Perplexity's search API post says it retrieves self-contained passages with a mix of keyword and meaning-based search, then reranks them 3.
- Ahrefs found 28.6% of Perplexity's cited links ranked in Google's top 10, the highest of the assistants it tested 4.
- Perplexity recommends allowing PerplexityBot in robots.txt. Its user-triggered fetcher generally ignores robots.txt 1.
- In our test, two AI tools ran 9 searches for this question and listed 15 sites, with one site on both lists.
What is Perplexity?
Perplexity is an AI search product that answers a question in plain language and shows where each part came from. Its help centre describes four steps: it interprets the question, searches the internet in real time, compiles the most relevant insights, and attaches numbered citations that link to the original sources 2.
That design makes it different from a chat tool that sometimes looks things up. A lookup is the normal path here. Perplexity also offers a Research mode that runs many searches, reads many sources and writes a longer report in about two to four minutes 2.
How big is it? Perplexity's post says its search system processes 200 million queries a day, including API use, so it is not a count of Perplexity app users 3. It is self-reported. We hold no independently verified figure for its share of all AI searches, so we do not give one.
How does Perplexity choose which websites to cite?
Perplexity does not publish a ranking formula, but it has described its retrieval pipeline. In a September 2025 engineering post it says its index tracks over 200 billion unique URLs, and that it breaks documents into self-contained spans that are retrieved and ranked on their own 3. The post describes its Search API, and we assume, without confirmation, that the consumer product works the same way.
Candidates come from two kinds of search merged together: lexical (matching words) and semantic (matching meaning). Basic filters then remove clearly stale or non-responsive (off-topic) content, and several ranking stages follow, ending with cross-encoder rerankers on a narrowed set 3. A cross-encoder is a model that reads the question and a passage together and scores how well they fit.
The post names no weights. It mentions human feedback and LLM checks of parsing quality in improving ranking and parsing, and says domain authority and topic coverage influence which documents are kept readily available in the index. It does not say these decide which sites are cited 3. Freshness is handled by a model that decides when a URL should be re-fetched, based on its importance and how often it changes 3.
Ahrefs links this design to results. It says Perplexity is the outlier among assistants because it has its own index and is built to cite sources 4. That is Ahrefs' explanation, not a tested finding.
| Stage | What Perplexity says happens | What it means for your page |
|---|---|---|
| Index | Tracks over 200 billion URLs and decides when to re-fetch each one | Be crawlable, and update pages you want kept fresh |
| Split | Documents become self-contained spans | Make each paragraph stand alone under its heading |
| Retrieve | Lexical and semantic search merged into one candidate set | Use the buyer's own words and the meaning behind them |
| Filter | Stale or off-topic content is removed early | Keep pages current and on-topic |
| Rerank | Fast scorers first, then cross-encoder rerankers | Answer the question directly near the top of the passage |
Which bots does Perplexity use, and should you allow them?
Perplexity documents two. PerplexityBot surfaces and links websites in Perplexity search results, and Perplexity says it is not used to crawl content for AI foundation models. Perplexity-User visits a page when a person asks a question, and Perplexity says that fetcher generally ignores robots.txt because the user requested it 1.
Perplexity recommends allowing PerplexityBot and its published IP ranges, and says changes can take up to 24 hours to apply 1. It treats each robots.txt setting as independent. If you want to appear, do not block PerplexityBot.
There is a dispute you should know about. In August 2025 Cloudflare alleged that Perplexity used an undeclared crawler with a generic Chrome user agent to reach sites that had blocked it, at an estimated 3 to 6 million requests a day 8. Cloudflare's post is its own account and test results; we have not checked whether Perplexity has since replied, and we have not verified the claim. If control matters to you, block at the firewall as well as in robots.txt.
Do you need to rank on Google to be cited by Perplexity?
It helps more here than elsewhere, but it is not required. Ahrefs ran 15,000 long-tail prompts and found 28.6% of Perplexity's cited URLs were in Google's top 10 for the same prompt, against 8.0% for ChatGPT's in-text citations 4.
So roughly seven in ten Perplexity citations came from outside that top 10. The data was collected in 2025 and covers long-tail prompts only 4. Treat it as a reason to publish focused pages for sub-questions, not as a promise that ranking is optional.
Which websites does Perplexity cite most?
In Semrush's 13-week tracking of more than 230,000 prompts to 12 October 2025, Perplexity's most-cited domains were Reddit, LinkedIn, NIH, Microsoft and Google 5. Wikipedia, so prominent in ChatGPT, held at about 0.8% of Perplexity responses 5.
A second Semrush study of about 217,000 prompts found Reddit was the most-cited domain on Perplexity, at about 4% (3.5% in its summary), on a base the article does not define clearly. Its average position in the answer was about 3.4, earlier than on ChatGPT Search or Google AI Mode 6. The article's own figures differ slightly between its summary and body, and Semrush sells AI visibility tools 6.
The practical reading is about types of source, not single domains. Community threads, professional networks and public-interest sites all earn citations. A business page competes with them, so it needs a clear, specific answer they do not give.
What did two AI tools search for on this question?
On 20 October 2026 we gave this page's question to Gemini 2.5 Flash with Google Search grounding and to a Claude agent with a web search tool. We did not run Perplexity itself. Gemini ran 2 searches and the Claude agent ran 7 7. The Claude agent was told to research thoroughly and Gemini was not, so the counts do not compare the tools.
The runs listed 15 sites, and only one site, llmpulse.ai, appeared on both lists. Only the Claude run listed a perplexity.ai domain, which is part of why we cite Perplexity's own documents directly. Two runs on one day show a pattern at best.
| Tool | Search it ran | What it seems to be after |
|---|---|---|
| Gemini | how Perplexity chooses websites to cite | The selection question |
| Gemini | Perplexity citation methodology | How its citations are built |
| Claude | how Perplexity chooses sources to cite | The selection question, worded differently |
| Claude | Perplexity citation study most cited domains | Data on which domains win |
| Claude | Perplexity ranking factors freshness authority | Which signals might matter |
| Claude | optimize content to get cited by Perplexity | Tactics |
| Claude | Perplexity vs ChatGPT vs Google AI Overviews citations comparison | Differences between engines |
| Claude | PerplexityBot robots.txt crawling | The access rule |
| Claude | Perplexity Reddit YouTube citation share 2026 | Current share of community and video sources |
What helps a page get cited by Perplexity?
The best published test is academic and dates from 2023 and 2024. The GEO paper reported that methods such as adding quotations and statistics could raise visibility in generative engine responses by up to 40%, and that the effect varied by subject area 9. In a test on Perplexity.ai, the authors uploaded source text for a 200-query subset instead of letting it search the web. Quotation addition improved Position-Adjusted Word Count by about 22%, and keyword stuffing scored about 10% below baseline 9.
Those tests are older than the engines now in use, and the paper's authors ran them on test queries. Use them as a hint, not a plan. What follows from Perplexity's own description is simpler: because it ranks spans, each paragraph has to make sense alone.
Fix this paragraph by paragraph. Name the subject, state the answer, then give the condition. Avoid openers such as 'as noted above', which strip meaning from a passage that is quoted alone.
- Weak: It also costs more in winter, as noted earlier.
- Strong: A custom cake from Example Bakery costs about 15% more in December, because holiday orders fill the ovens.
- Why: the strong version names the business, the product, the change and the reason, so it still works when quoted alone.
Is Perplexity worth optimising for?
It is worth a light, cheap effort, because the basics overlap with every other engine. Cloudflare measured how many crawl requests each platform made per referred visit across its network in 2025. Perplexity stayed below 200 to 1 from September, while OpenAI peaked near 3,700 to 1 in March 10. A lower ratio suggests more visits sent back per page read, though the figure comes from one network.
We hold no verified figure for how many of your own visitors will arrive from Perplexity, and rates differ widely by industry. Measure it with your analytics, as in how to track AI traffic in GA4.
What should you check on Perplexity first?
Check access first, then wording. The steps are free and take an afternoon.
- Confirm PerplexityBot is not disallowed in your robots.txt, and that no firewall rule blocks its published IP ranges 1.
- Run the free AI Crawler Checker on your five most important pages.
- Rewrite each key paragraph so it stands alone: subject, answer, condition 3.
- Put the direct answer in the first sentence under every heading. The Answer Block Checker tests this.
- Ask Perplexity your ten buyer questions and record which sources it cites, in what position, and whether you appear.
- Read how to rank in Perplexity for the full routine.
What an AI assistant searches when it answers this
We gave the question below to AI search tools and recorded the web searches they ran behind the scenes. This is called query fan-out. Each phrase is a page you could publish to be found and cited.
“how does Perplexity choose which websites to cite”
The searches it ran- how Perplexity chooses sources to cite
- Perplexity citation study most cited domains
- Perplexity ranking factors freshness authority
- optimize content to get cited by Perplexity
- Perplexity vs ChatGPT vs Google AI Overviews citations comparison
- PerplexityBot robots.txt crawling
- Perplexity Reddit YouTube citation share 2026
- how Perplexity chooses websites to cite
- Perplexity citation methodology
Frequently asked questions
How does Perplexity choose which websites to cite?
Perplexity publishes no formula. Its engineering post for its Search API says it retrieves self-contained passages with lexical and semantic search, removes stale or non-responsive content, and reranks with cross-encoder models. Domain authority and topic coverage are mentioned for indexing priority, not as citation factors 3.
Should I allow PerplexityBot in robots.txt?
If you want to appear in Perplexity results, yes. Perplexity says PerplexityBot surfaces and links sites in its search results and is not used to crawl content for AI foundation models. It recommends allowing it and its IP ranges, and changes can take up to 24 hours 1.
Does Perplexity only cite pages that rank on Google?
No. Ahrefs found 28.6% of Perplexity's cited URLs were in Google's top 10 for the same long-tail prompt, which is higher than ChatGPT at 8.0% but still a minority 4. Its own index is one likely reason.
Which sites does Perplexity cite most often?
Sources
- Perplexity Crawlers · PerplexityOpened October 20, 2026. Supports: PerplexityBot surfaces and links sites and is not used for foundation model crawling; Perplexity-User generally ignores robots.txt; recommends allowing PerplexityBot and IP ranges; up to 24 hours for changes; each setting independent.
- How does Perplexity work? · Perplexity Help CenterOpened October 20, 2026. Supports: Four-step process: interpret, search in real time, compile, numbered citations; Research mode runs many searches and takes about 2 to 4 minutes.
- Architecting and Evaluating an AI-First Search API · PerplexityOpened October 20, 2026. Supports: 25 September 2025: index tracks over 200 billion URLs; 200 million daily queries (self-reported); self-contained spans; hybrid lexical and semantic retrieval; prefiltering; multi-stage reranking with cross-encoders; freshness model; human feedback and LLM parsing checks; domain authority and topic coverage for indexing priority.
- Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt · AhrefsOpened October 20, 2026. Supports: Published 11 August 2025; 15,000 long-tail prompts; Perplexity 28.6% overlap with Google's top 10 versus 8.0% for ChatGPT in-text citations; Perplexity has its own index.
- The Most-Cited Domains in AI: A 3-Month Study · SemrushOpened October 20, 2026. Supports: Published 10 November 2025; 230,000+ prompts over 13 weeks to 12 October 2025; Perplexity's top sources Reddit, LinkedIn, NIH, Microsoft, Google; Wikipedia about 0.8% of Perplexity responses.
- Reddit Visibility in AI Search · SemrushOpened October 20, 2026. Supports: Published 10 November 2025; about 217,000 prompts; Reddit the top-ranked domain on Perplexity at about 4% of citations; average position about 3.4; internal inconsistencies and vendor interest noted.
- Query Fan-Out Explorer (observed examples) · Chalk LabsOpened October 20, 2026. Supports: From Chalk Labs' research log: the searches recorded on 20 October 2026 for the prompt 'how does Perplexity choose which websites to cite': Gemini 2.5 Flash with Google Search grounding (2 searches) and a Claude agent with a web search tool (7 searches), and the sites each run listed.
- Perplexity is using stealth, undeclared crawlers to evade website no-crawl directives · CloudflareOpened October 20, 2026. Supports: 4 August 2025 allegation of undeclared crawling at 3 to 6 million daily requests; Cloudflare de-listed Perplexity as a verified bot; we have not checked whether Perplexity has since replied.
- GEO: Generative Engine Optimization · arXiv (Aggarwal et al.)Opened October 20, 2026. Supports: Visibility in generative engine responses can be boosted by up to 40% and effects vary by domain (abstract); in a test on Perplexity.ai using uploaded source text for a 200-query subset, quotation addition improved Position-Adjusted Word Count by about 22% and keyword stuffing scored about 10% below baseline (full paper, Table 5 and Appendix C.1).
- Cloudflare Radar 2025 Year in Review · CloudflareOpened October 20, 2026. Supports: Crawl-to-refer ratios on Cloudflare's network in 2025: Perplexity below 200:1 from September; OpenAI peak 3,700:1 in March.