How Does Perplexity AI Work, Step by Step, From Search to Citation?

Perplexity AI step by step: search, read, answer and cite

Perplexity AI handles a question in four stages. It searches its own index and the live web, picks the pages most likely to answer, reads them, then writes a reply that cites each source with a number. Clicking a number opens the page behind that part of the answer.

Two kinds of retrieval are involved. Pages already crawled by PerplexityBot sit in Perplexity’s search index, ready to be matched.

When a question needs a page right now, Perplexity-User fetches it on the user’s behalf. Perplexity’s crawler documentation describes both agents.

The answer is short by design. Most replies are a few paragraphs, with the citations doing the job of a results page.

Follow-ups build a thread. Asking “and the cost in India?” after a product question reuses the earlier context for the next search.

The company has not published its ranking method. The mechanics above come from its bot documentation and from how answers appear to users.

Digital Magma tests this the plain way: we ask the same buyer question several times and record which pages are cited each time. The list can shift between runs, so one screenshot proves little.

Perplexity search modes and what changes between them

Perplexity search modes trade speed for depth. A standard search returns a quick answer from a handful of sources. Pro searches and deep research are built to go further: more searches, more pages read and a longer wait for a fuller reply.

Airtel’s July 2025 announcement lists what Pro adds. It mentions more daily Pro searches, a choice of advanced models, deep research, file uploads and Perplexity Labs.

The free tier still searches and cites. Pro changes how much work goes into each answer, not whether sources are shown.

Mode names change often. Check the app itself for the current list before you write them into a client report.

For citation tracking, Digital Magma runs the standard mode by default and notes it in the report. Mixing modes in one panel makes month-to-month numbers meaningless.

PerplexityBot vs Perplexity-User: how Perplexity reaches your pages

PerplexityBot and Perplexity-User reach your pages for different reasons, and robots.txt treats them differently. PerplexityBot crawls to surface and link sites in search results. Perplexity-User visits a page only when someone’s question needs it, and because a person asked, it generally ignores robots.txt rules.

Both agents publish IP lists, at perplexitybot.json and perplexity-user.json. Use them to tell the real bots from fakes in your logs.

The documentation says changes may take up to 24 hours to reach its systems.

Firewalls are the hidden blocker. The same guide gives rules for Cloudflare WAF and AWS WAF that match both the user agent and the IP range.

A note on history. In June 2024, Wired reported that Perplexity did not respect robots.txt, as Wikipedia records. The bot split above is the company’s current documented answer.

Perplexity source selection: what we can and cannot see

Perplexity source selection is not documented in detail, so anyone claiming to know the formula is guessing. What is visible: it cites pages it can crawl, pages that answer the exact question, and pages whose facts are easy to quote. Academic work on generative engines adds some evidence.

The GEO paper by Aggarwal and colleagues, accepted at KDD 2024, tested content changes on Perplexity.ai as well as on its own benchmark.

Its best methods were citing sources, adding quotations and adding statistics. The authors report visibility gains of up to 40% on their benchmark.

Treat that as a research result, not a promise. It tested rewritten passages, not live business pages over months.

Perplexity for a business in India: why it matters

Perplexity matters for Indian businesses largely because of one deal. On 17 July 2025, Bharti Airtel announced 12 months of Perplexity Pro free for all its 360 million customers. That put a citation-first answer engine on a very large number of Indian phones almost overnight.

Airtel’s press release valued the subscription at about ₹17,000 a year. Customers claimed it through the Airtel Thanks app.

Check the current terms before quoting the offer. It was announced as a one-year benefit.

For B2B sellers, the effect is quieter. Anyone researching suppliers on a bundled Pro account sees named sources in every answer, so a cited spec sheet is a direct route to a buyer.

Digital Magma now asks Indian clients to check GA4 for perplexity.ai referrals before deciding how much effort it deserves.

Perplexity citations: 5 steps to get cited

Perplexity citations go to pages it can reach and quote with confidence. Digital Magma uses these five steps, starting with access because a blocked page cannot be cited at all. The later steps follow the GEO paper’s findings and our own reading of what gets linked.

1. Allow PerplexityBot. Keep it out of any disallow group, and let its published IPs through your WAF.
2. Answer the question in the first lines. Put the direct answer under each heading before the background.
3. Add numbers with sources. Prices, sizes, dates and test results, each with where it came from.
4. Quote named experts. A short quote from a real, named person gives the model a clean passage to lift.
5. Keep pages current. Show an update date and refresh the facts buyers check most.

Check results by asking your buyers’ questions each month. Note which pages are cited, and which rivals appear instead.

If nothing from your site is ever cited, Digital Magma rules out a WAF block first. A robots.txt file can look fine while every request from the bot gets a challenge page.

Perplexity in our client work

Perplexity gets the same treatment as any answer engine in Digital Magma’s client work: access first, then quotable facts. We do not run a separate Perplexity playbook, because the pages that earn citations elsewhere are the pages it can use too.

For a Toronto ENT clinic, the lesson from AI answers generally was precision. Service and doctor names had to be written plainly before any assistant described them correctly.

A Canadian welding equipment manufacturer we work with already appears in ChatGPT and AI Overviews.

We expect its specification pages to suit Perplexity as well, since they are dense with quotable facts. That is our inference, not a measured result.

Perplexity AI: frequently asked questions

Perplexity AI questions from marketers and founders, answered briefly.

Perplexity AI: related terms, sources and author

Perplexity AI connects to these entries in the Digital Magma knowledge base:

• PerplexityBot: Perplexity’s search crawler in detail
• AI citation: what a citation means and how it is counted
• ChatGPT search: OpenAI’s comparable product
• Retrieval-augmented generation: the method answer engines share
• How to get cited by ChatGPT and Perplexity: the practical follow-up

Sources: Perplexity’s crawler documentation, Bharti Airtel’s press release of 17 July 2025, the GEO paper on arXiv, and Wikipedia’s Perplexity AI article. All were checked on 22 October 2026.