How to track brand mentions in Perplexity: four checks that work (2026)

To track brand mentions in Perplexity, you need four separate readings, because Perplexity shows its sources more openly than any other major AI engine but never tells you how often it names you. The answer text and its numbered citations show whether you were mentioned and cited for one prompt in one session. The API repeats that check at scale. Your server logs show when Perplexity's two crawlers fetch your pages, and analytics shows the clicks that follow. Ahrefs' analysis of 15,000 queries found that 28.6% of the URLs Perplexity cites also rank in Google's top 10 for the same query, far above other assistants, so it is the one engine where classic SEO data still predicts part of the result.

What counts as a mention in Perplexity, and what counts as a citation

Perplexity puts two different signals on screen at once, and a tracking sheet needs a column for each. Record them separately: a brand can be named on the strength of a review site's page while its own domain is never cited at all.

In Perplexity, a mention is your brand name inside the answer text; a citation is one of your URLs among the numbered sources attached to that answer. One often appears without the other.

Each claim in a Perplexity answer carries an inline citation number that points to a source listed with the answer, which makes the evidence behind a mention easy to inspect. The way those sources get picked is covered in how Perplexity decides which sources to cite. A third column matters too: whether the citation is a clickable link to your site or a page about you on someone else's domain.

The Ahrefs figure explains why this engine behaves differently. Across the same 15,000 queries, only about 8% of ChatGPT's in-text citations ranked in Google's top 10, against 28.6% for Perplexity. If your page ranks on page one for a question, its odds of being cited by Perplexity are better than for any other assistant, and your rank tracking data becomes an early warning signal. Most tools to track brand mentions record all three columns for Perplexity, but the method below lets you collect them yourself first and judge whether a tool is worth paying for.

How to check Perplexity by hand without fooling yourself

A manual check is a fixed list of prompts, typed the same way, in the same conditions, on a fixed schedule, with every answer saved. It costs nothing but time, and it is the only method that shows exactly what a buyer sees on screen, sources included.

Start with 20 to 30 prompts that a buyer would type before knowing your name: category questions ("best payroll software for a 50-person company") and problem questions, with no brand in them. Branded prompts tell you what Perplexity says about you, which matters, but they say nothing about whether you get recommended.

Then control the conditions. Use a private window or a dedicated account with no history, write down the country you are connecting from, and note which model was selected, since paid plans let users switch models and the choice can change the answer. Save the full answer and the list of sources, not just a yes or no.

Run each prompt more than once. A study by SparkToro and Gumshoe.ai, with 600 volunteers and about 3,000 runs, put the chance of getting the same brand list twice at less than 1 in 100. A single run is an anecdote. Ten runs of the same prompt give you a rate, and rates are what prompt monitoring is built on. Divide the runs that name you by the total runs, do the same for each competitor, and you have the inputs to measure AI Share of Voice for Perplexity alone.

Running the same prompts through Perplexity's API

Perplexity's Agent API returns, with each answer, a list of search results giving the title, URL and date of every source used, which turns the manual spreadsheet into a script. You send the prompt, store the answer and the sources, and search both for your brand name and your domain.

The developer setup changed in September 2026. Perplexity's documentation says support for the older Sonar Chat Completions ended on 27 September 2026, with synchronous requests now reformulated as Agent API requests, and it recommends the Agent API for all new projects. Any tracking script written against the old endpoint should be moved.

Two parameters make the API useful for brand tracking. The web search tool accepts a user_location object with country, region and city, and the documentation notes that city and region improve location accuracy, so you can test how answers shift between markets. A domain filter accepts up to 20 domains to include or exclude, which helps isolate whether a competitor's page or a review site is carrying the answer. Each standard web search costs $2.50 per 1,000 calls, or $1.00 on the faster path, plus model tokens (prices checked on 27 October 2026). A weekly run of 300 prompts means at least 1,200 searches a month, about $3 in search fees before tokens.

Treat API results as a close proxy, not a copy of the app. A logged-in user brings history and a model choice that your script does not reproduce, which is why manual spot checks stay useful even once the script runs.

What your server logs say about Perplexity

Perplexity identifies two user agents. PerplexityBot crawls pages to surface and link them in Perplexity's search results, and Perplexity-User fetches a page when a person's question needs it. Perplexity's crawler documentation says Perplexity-User generally ignores robots.txt because a user triggered the request, and it publishes both IP ranges as JSON files so you can verify each visit.

The split is useful for tracking, and reading it only takes your raw server logs. PerplexityBot visits tell you which pages sit in Perplexity's index. Perplexity-User visits, grouped by URL, tell you which pages were pulled into live answers, the closest thing to an impression report Perplexity offers. A page with steady Perplexity-User hits and no mention in your prompt checks is being read for questions you are not tracking yet.

Filter by user agent and by IP range together. On 4 August 2025, Cloudflare reported that besides the declared Perplexity-User agent, at 20 to 25 million requests a day, it saw an undeclared crawler impersonating Chrome on macOS at 3 to 6 million requests a day, from IPs outside Perplexity's published ranges, and it removed Perplexity from its verified bots list. Perplexity answered the same day that user-driven fetching is not crawling. Whatever side you take, some Perplexity traffic in your logs may not announce itself, and the method for separating real bots from look-alikes is the same one used in AI crawler log analysis.

Perplexity referrals in GA4, and why the numbers look small

Clicks from Perplexity arrive with perplexity.ai as the referrer, so in Google Analytics 4 they appear as a referral source you can isolate with a filter or a custom channel group. This is the only one of the four readings that ties a Perplexity answer to a visit and a conversion.

Expect modest volumes. SE Ranking's study of 101,574 websites, tracked from January 2025 to April 2026, found that Perplexity sent 7.23% of all AI referral traffic in 2026, behind ChatGPT at 74.78% and Gemini at 11.56%. In the United States, Perplexity's share fell from 11.42% in 2025 to 6.85% in 2026. A low referral count therefore does not mean you are absent from Perplexity answers; it may only mean few readers click through.

Set up the channel once with the steps in our guide to measure AI traffic in GA4, then compare the landing pages that receive Perplexity sessions with the pages Perplexity-User fetches in your logs. Pages that appear in both lists are the ones Perplexity both reads and sends people to.

Which method answers which question

Each method answers a different question, and none of them covers everything. The table below sums up what each one shows and where it goes blind.

Manual checks and logs are enough to start. A tracker becomes worth its price once you follow more than about 50 prompts, several countries, or several competitors, because the repeat runs that make a rate reliable quickly become a full day of work each week.

What to do when Perplexity gets your brand wrong

Because Perplexity shows its sources, a wrong answer about your brand usually traces back to a page you can open and sometimes correct. Click each cited source for the wrong claim and note who owns it.

If the page is yours, fix the facts there first and make the correct version easy to quote: one clear sentence with the right price, feature or date. If the page belongs to a review site, a directory or a forum, the work is outreach or a fresh review, since citations that don't come from your site often weigh more than your own pages. The full playbook for fixing wrong AI answers about your brand covers each engine's feedback options.

Perplexity's documentation says Perplexity-User may visit a page when a user asks a question, so a corrected page can be read on the very next relevant question rather than after a model retraining cycle. Whether it is cited again depends on retrieval, which is exactly what the next round of prompt checks will show.