# How Perplexity works: a plain-English guide for marketers

> How Perplexity searches, reads and cites the web, what Search, Research and Computer do, and what that means for marketers. Checked Nov 2026.

Source: https://mentionx.ai/learn/perplexity/perplexity-how-it-works-marketers  
Author: Dipuck Jones, founder of MentionX  
Updated: 5 November 2026  
Level: Practitioner

**Perplexity is an answer engine. When you ask a question, it searches the web (and any files or connected apps you allow), picks the most relevant passages, and writes an answer with numbered citations that link to its sources. For marketers this has two consequences. Perplexity is a fast research assistant whose sources you can check. And it is a new place where buyers find your brand, but only if your pages are reachable and worth quoting.**

> **Key takeaways**
>
> - Every answer starts with retrieval: a live search over Perplexity's index plus pages fetched on demand.
> - Citations are the product. Each numbered source is a link a buyer can click, which is why share of citations matters.
> - Search, Pro Search, Research, Model Council and Computer use the same retrieval idea at different depths and costs.
> - PerplexityBot (indexing) and Perplexity-User (on-demand fetches) are separate agents with separate rules.

*Checked 5 November 2026 (CET) against the Perplexity Help Center, Perplexity blog, Perplexity API docs and the outlets cited in Sources. Perplexity renames and moves features often, so confirm labels and limits in your own account. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.*

## What happens when someone asks Perplexity a question?

Here is a simplified version of the pipeline. It is enough to make good decisions as a marketer.

1. **Understand the question.** Perplexity rewrites the query, often into several sub-queries. "Best AP automation for mid-market" might become searches for comparisons, pricing, reviews and integration lists.
2. **Retrieve.** It searches its own web index. Perplexity says the index behind its Search API covers more than 200 billion URLs. It can also fetch specific pages live, for example a URL the user pasted. That on-demand fetch uses the Perplexity-User agent.
3. **Rank passages, not just pages.** Perplexity's engineering write-up on its Search API describes hybrid lexical and semantic retrieval, multi-stage ranking, and pages split into self-contained sub-document spans that are ranked one by one. In practice, a single well-written paragraph can win a citation even if the rest of the page is mediocre.
4. **Write the answer.** A language model writes a response grounded in the top passages and adds numbered citations.
5. **Offer next steps.** Related questions, follow-ups and, for some queries, shopping cards, maps or finance widgets.

*Our view:* steps 2 and 3 are where marketers win or lose. If your page isn't retrievable, it can't be cited. If it doesn't contain a clean, quotable passage that answers the sub-query, a competitor's page will be cited instead.

## How is Perplexity different from Google and ChatGPT?

| | Google Search | ChatGPT (with search) | Perplexity |
|---|---|---|---|
| Default output | Ranked links, AI Overviews on some queries | Conversational answer, searches when needed | Cited answer on every query |
| Citations | Links and AI Overview sources | Inline sources when search is used | Numbered sources by default |
| Crawler for search | Googlebot | OAI-SearchBot | PerplexityBot |
| On-demand fetch agent | Not applicable | ChatGPT-User | Perplexity-User |
| Ads in answers | Yes (Search, AI Mode, AI Overviews) | Testing for free users | No. Perplexity stopped testing ads in 2026 |

The last row matters. Search Engine Land reported in February 2026 that Perplexity had phased out the ads it began testing in 2024. That means the only way to appear in Perplexity's answers is to earn the citation (see [lesson 17](https://mentionx.ai/learn/perplexity/perplexity-shopping-publishers-ads)).

## What are the main product surfaces in 2026?

Perplexity has changed a lot. Here is the current map with old names in brackets.

| Surface | What it does | Who has it |
|---|---|---|
| Search (Ask) | Fast cited answers. Doesn't use Computer credits | All plans |
| Pro Search | Deeper multi-step search with more sources and model choice | Limited on Free, broad on paid plans |
| Research (Deep Research) | Dozens of searches, reads many sources, writes a report in minutes | All plans with limits, higher on Max |
| Model Council | Sends one question to several models and shows where they agree and disagree | Max and Enterprise Max on web and iOS. Uses credits in Computer on paid plans |
| Computer | An agent that does multi-step work such as reports, dashboards and files, and pays for it with credits | Paid plans |
| Projects (Spaces) | Shared workspace with instructions, files, connectors and threads | All plans, with sharing and limits by plan |
| Comet | Perplexity's free browser with an assistant built in | Free to download. Agent features on paid tiers |
| API | Agent API, Search API, Embeddings | Pay as you go |

Lesson 3 covers which mode to use when. Lesson 2 covers which plan to buy.

## What does Perplexity cite?

Perplexity cites the pages it used. Each citation appears as a numbered marker in the answer and a source in the sources panel. Users can open the sources list, hover over a citation to preview it, and click through. Some things are worth knowing:

- **Citations are passage-level.** Perplexity can cite one paragraph from a 5,000-word guide. Clear headings and self-contained paragraphs help (see [lesson 15](https://mentionx.ai/learn/perplexity/perplexity-content-formats-get-cited)).
- **Freshness matters for time-sensitive queries.** News, pricing and "2026" queries pull recent pages. Evergreen definitions can pull older ones.
- **Third-party pages count.** For "best X" and "X vs Y" prompts, review sites, communities and publisher roundups often appear alongside vendor pages. That is why earned media matters ([lesson 16](https://mentionx.ai/learn/perplexity/perplexity-citation-gap-earned-media)).
- **Sources can be steered by the user.** People can choose sources (Web, Academic, Social, Finance and others, depending on your plan and the current UI), upload files or connect apps. So the same brand can be cited in one setup and missing in another.

## Which crawlers touch your site?

Perplexity documents two user agents:

- **PerplexityBot** "is designed to surface and link websites in search results on Perplexity. It is not used to crawl content for AI foundation models." Perplexity recommends allowing it and permitting its published IP ranges.
- **Perplexity-User** supports user actions. When someone asks a question, it may visit a page to answer and link it. Perplexity says that "since a user requested the fetch, this fetcher generally ignores robots.txt rules."

Changes to robots.txt can take up to 24 hours to take effect. [Lesson 14](https://mentionx.ai/learn/perplexity/perplexity-perplexitybot-robots-waf) has the full configuration, including Cloudflare and AWS WAF rules. The reference page [PerplexityBot vs Perplexity-User](https://mentionx.ai/learn/technical/perplexitybot-perplexity-user) covers the technical details.

## Why do answers vary?

Expect variance. The same prompt can produce different citations because:

1. The live search results change daily.
2. The user's mode (Search, Pro Search, Research) and model differ.
3. Follow-up context changes the sub-queries.
4. Location and language affect results.
5. Project instructions and uploaded files change what is retrieved.

This is why one screenshot proves nothing. Measure with a fixed prompt set, run on a schedule, as covered in [lesson 19](https://mentionx.ai/learn/perplexity/perplexity-share-of-citations-tracking).

## Worked example: Ledgerline

*Ledgerline is a fictional accounts-payable software company. Details are illustrative.*

Ledgerline's head of content asked Perplexity "best accounts payable automation software for mid-market companies". The answer cited Competitor A's pricing page, two review-site roundups, a Reddit thread and Competitor C's integration list. Ledgerline didn't appear.

The team traced the answer back through the pipeline:

| Pipeline step | Ledgerline finding |
|---|---|
| Retrieval | robots.txt allowed PerplexityBot, but the CDN's bot protection was challenging unknown bots |
| Passage ranking | The comparison page buried pricing in a PDF, so there was no quotable passage |
| Third-party sources | Ledgerline was missing from both cited review roundups |
| Freshness | The "2025 buyer's guide" hadn't been updated |

### Walkthrough

1. Write the buyer question exactly as a buyer would type it.
2. Run it in Search, then Pro Search, and save both source lists.
3. For each cited page, note its type: vendor, review site, community, publisher or documentation.
4. Check whether your own equivalent page exists, is reachable, and has a passage that answers the question directly.
5. Turn each gap into a ticket: technical (lesson 14), content (lesson 15) or earned media (lesson 16).

> **Pro move:** open the sources panel before you read the answer. The sources show you what Perplexity trusts for this question. The prose is only a summary of them.

## What does this mean for your marketing plan?

Think of Perplexity work in three layers:

- **Use it.** Research faster with verifiable sources (lessons 3-12).
- **Be findable.** Crawler access, clean HTML, fast pages (lesson 14).
- **Be quotable and corroborated.** Answer-first content, original data, and presence on the third-party pages Perplexity already cites (lessons 15-16).

Then measure all three: referral sessions in GA4 (lesson 18) and share of citations (lesson 19).

## Common mistakes

- **Treating Perplexity like Google rankings.** There is no "position 1". There is a set of cited sources, and it changes with each sub-query.
- **Blocking PerplexityBot by accident.** Bot-fight settings or a blanket AI-crawler block can remove you from the index.
- **Judging by one prompt.** Use a prompt set.
- **Ignoring third parties.** Your own site is often a minority of the citations on commercial prompts.

## Try it yourself

Pick three of your ten buyer questions. For each one, run Perplexity Search and record every cited source in a sheet with columns for prompt, source URL, source type, and whether it is yours or a competitor's. Count how many of the citations are yours. That's your first rough share-of-citation number. You will make it rigorous in lesson 19.

## Quick quiz

**1. Which Perplexity agent indexes sites for search results?**

*Answer:* PerplexityBot.

**2. Why can a single paragraph win a citation?**

*Answer:* Perplexity ranks sub-document spans (passages), not only whole pages.

**3. Can you buy an ad placement in Perplexity answers in November 2026?**

*Answer:* No. Perplexity phased out its ad tests and says it isn't pursuing new ad deals.

## What should you check before the next lesson?

- You can name the difference between PerplexityBot and Perplexity-User.
- You have saved source lists for three buyer questions.
- You know which product surfaces your team already uses.

Next: [Perplexity plans and pricing for marketing teams](https://mentionx.ai/learn/perplexity/perplexity-plans-pricing-marketers). Or return to the [Perplexity for Marketers track](https://mentionx.ai/learn/perplexity/perplexity-for-marketers). Related: [How Perplexity finds and chooses sources](https://mentionx.ai/learn/perplexity/perplexity-how-sources-are-chosen), [Search modes and Research](https://mentionx.ai/learn/perplexity/perplexity-search-modes-research), [How to rank in Perplexity](https://mentionx.ai/learn/engines/perplexity/how-to-rank).

## Common questions

**Is Perplexity a search engine or a chatbot?** Both. It runs live web searches, reads the results and writes an answer with numbered citations, so it behaves like a search engine that replies in prose.

**Does Perplexity train its models on my website?** Perplexity says PerplexityBot is used to surface and link sites in search results and is not used to crawl content for AI foundation models.

**Why do two people get different answers to the same question?** Answers depend on live search results, the mode and model used, follow-up context, location and Project instructions, so small differences are normal.

## Sources

- Perplexity API docs, [Perplexity Crawlers](https://docs.perplexity.ai/docs/resources/perplexity-crawlers)
- Perplexity blog, [Architecting and Evaluating an AI-First Search API](https://www.perplexity.ai/hub/blog/architecting-and-evaluating-an-ai-first-search-api)
- Perplexity Help Center, [What is Model Council?](https://www.perplexity.ai/help-center/en/articles/13641704-what-is-model-council)
- Search Engine Land, [Perplexity stops testing advertising](https://searchengineland.com/perplexity-stops-testing-advertising-469452) (18 Feb 2026)
- All accessed 5 Nov 2026.

## Frequently asked questions

### Is Perplexity a search engine or a chatbot?

Both. It runs live web searches, reads the results and writes an answer with numbered citations, so it behaves like a search engine that replies in prose.

### Does Perplexity train its models on my website?

Perplexity says PerplexityBot is used to surface and link sites in search results and is not used to crawl content for AI foundation models.

### Why do two people get different answers to the same question?

Answers depend on live search results, the mode and model used, follow-up context, location and Project instructions, so small differences are normal.

---

Written and reviewed by Dipuck Jones for MentionX Learn. The HTML version of this page: https://mentionx.ai/learn/perplexity/perplexity-how-it-works-marketers