DeepSmith

Nov 26 · AEO & AI Visibility

17 min read

How to Track Your Brand Visibility in Perplexity

Avinash Saurabh
Avinash Saurabh · CO-Founder & CEO
Dark monochrome illustration of stacked answer cards linked by thin lines to source nodes and a small bar chart, behind the words Track Your Perplexity Visibility.

If you want to know whether Perplexity names your brand when buyers ask about your category, a few searches here and there won't tell you much. This guide is for marketing leads who need a repeatable way to monitor visibility in Perplexity, so you can say how often your brand is mentioned, how often your pages are cited, and how that compares with your competitors. By the end you'll have a prompt set, a run log, a few simple formulas, and a way to read the trend without fooling yourself.

You don't need special software to start. A spreadsheet and a bit of discipline will get you there. This guide is about measuring Perplexity brand visibility, not improving it, so you won't find tactics for earning more citations here.

Step 1: Define what you are measuring

Before you run a single prompt, write down what counts. Pick the brand or product you're measuring and list the names buyers actually use for it, including shorthand and any alternate spellings. Then list the domains and subdomains that belong to you. If your company has several products, decide whether this report covers one product or the whole company.

Next, set the market and language. Location and language can change the sources and answers Perplexity shows, so treat them as part of the measurement, not a side detail.

Last, choose three to seven competitors and keep that list stable for the whole reporting period. If you're not sure who belongs on it, our guide to choosing which competitors to track walks through it. If you also watch other engines, the Google AI Overviews tracking tools roundup covers that side.

You're done when another person could read your notes and know exactly which brand, which domains, which locale, and which competitors are in the report.

Common mistake: changing the competitor list or the brand aliases from one period to the next without saying so. A shift can then look like a change in visibility when part of it is just a change in how you measured.

Step 2: Build a fixed, versioned prompt set

Your prompt set is your measuring instrument, so it needs to stay put. Write down the questions your buyers really ask, across the stages they go through: learning about the category, comparing options, and evaluating a product. Give each prompt a stable ID, the exact wording, and a label for its intent or buyer stage, so you can read results by topic as well as in total. The article on the four types of prompts worth tracking is a good way to make sure the set isn't lopsided.

Then version the set. Call it something like core-buyer-prompts-v1 and note the date you froze it. If a question has to change in a real way, make a new version instead of quietly editing the old one, and only compare results against the same version. If you want more on building one, see how to build a repeatable prompt set.

You're done when every prompt has an ID, exact wording, a version, and an intent or stage label.

Common mistake: rewording a prompt for clarity halfway through a trendline. Even a small change can alter what Perplexity retrieves and writes, and then you can't tell whether the answer moved or your question did.

Step 3: Set a consistent Perplexity search protocol

Perplexity gives you a few choices that change what you get back, so pick them once and stick with them.

First, use the same source setting every time. Perplexity offers source choices like Web and Academic, and custom sources can limit or favor certain sites. A search that's limited to specific sites isn't a fair stand-in for an ordinary web search, so don't let a custom-source or file-only answer slip into a report about public-web visibility.

Second, use the same search mode. For a routine buyer question, ordinary Search is the usual pick. Research mode is built for in-depth research, runs many searches, and picks its own models, so it isn't a controlled match for a standard Search answer. Write down the mode and source mode shown on each answer.

Third, cut down on account history. Perplexity's help center says Memory and Search history can each be turned off in Settings, and that both are off in Incognito mode. If you're signed in, note whether you ran with history and memory off or used Incognito. The exact clicks to get into Incognito can differ by interface, so check what you see on screen rather than copying someone else's steps.

Finally, start a fresh session for every run. A follow-up question in an old thread is shaped by the earlier turns, so it isn't an independent repeat of your prompt.

For every run, write down the date and time, the time zone, the locale, the exact prompt, the search mode, the source mode, and the account state. Capture what you actually used, because you can't assume every user or interface behaves the same way.

You're done when someone else could recreate your conditions and tell whether a result came from Web search, a limited source setup, or Research mode.

Pro tip: treat each prompt-run as its own clean session. It's the easiest habit to build and the one that most protects your data.

Step 4: Repeat each prompt and keep the sample counts visible

One answer is a snapshot. It's not a good estimate of how often your brand shows up, because Perplexity answers and the sources behind them can differ from one run to the next. So run each prompt more than once.

A practical floor is at least three runs per prompt in each collection window, then record the majority outcome. DeepSmith's own prompt-set guidance recommends this, and it's a floor for repeatability, not a promise of statistical precision. More runs help if you have the time, as long as the prompt, settings, and window stay the same.

To monitor visibility in Perplexity well, pick your rhythm before you start. Weekly is a standard reporting window, monthly is the least you'd want, and daily tends to add noise that's hard for people to read. If you use DeepSmith, the platform can run a frozen prompt set on a schedule, and you can set how often collection happens in Settings.

Whatever you choose, put the number of runs and the dates covered in your report, so readers can see how much evidence sits behind each number. If a prompt got fewer valid runs than you planned, say so with the real count, and don't treat a missing answer as a "not mentioned."

You're done when each prompt has repeated runs inside one consistent window and your report shows how many runs there were.

Common mistake: reading one odd answer as a lasting change. Check it against repeated runs and the next comparable window before you tell anyone it moved.

Step 5: Capture the full answer and its sources

For each run, save the exact prompt and the whole answer. Then log whether your brand appears, where in the answer if that's easy to tell, and the sentiment if your program tracks it. Record every source the answer cites, with the page title and URL, and note whether each one belongs to you or to a competitor. Also log which competitors were mentioned and which were cited. If you notice a mode change, an outage, or a model change, add that to the notes too.

A row per prompt-run works well. Here's a layout you can copy into a sheet.

FieldWhat to record
Prompt ID and exact promptThe precise question you submitted, not a paraphrase
Prompt-set versionThe version used for the run
Intent or buyer-stage labelThe category you gave the prompt
Engine and modePerplexity and the mode shown for the answer
Source modeFor example, Web or another displayed setting
Run timestamp and localeOne consistent time zone, plus the market and language
Account stateSigned in or not, and whether history and memory were off or Incognito was used
Full answer textThe answer word for word
Brand mentionYes or no, and optionally position and sentiment
Brand citationYes or no, and the exact cited page or pages
CompetitorsNames mentioned, and which of their pages were cited
NotesMode, model, outage, or process changes

When you decide what an answer cited, use the sources linked in the answer itself, not a general list of search results. If your team wants an audit trail, Perplexity documents exporting an answer to PDF, Markdown, or DOCX, which can sit alongside your log. The row-level log is still the record you analyze.

You're done when someone else can open a row and see what was asked, what Perplexity said, and which sources the answer linked to.

Step 6: Label mentions, citations, and appearances separately

Mentions and citations are two different things, and mixing them is the quickest way to get a confusing report. A brand can be named in an answer without any link to its site. Use one simple coding rule and apply it the same way every time:

  • Mentioned: the answer text names your brand, product, or a defined alias.
  • Cited: the answer links to a page on a domain you've defined as yours.
  • Competitor mentioned: the answer names a competitor on your list.
  • Competitor cited: the answer links to a tracked competitor's page.
  • No answer citation: no source link supports the answer, even if a brand name appears in it.

Decide ahead of time how you'll handle aliases, product names, parent-company names, and domains. Count an answer once for the mention rate even if your name shows up in it five times. If you want page-level analysis, keep a separate count of distinct cited pages. That way a long answer with repeated references doesn't skew the numbers.

There's one more detail worth knowing if you ever pull data from Perplexity's API. Its documentation separates search results from answer citations: search results carry source details, while the numbered citation markers in the generated text point back to those sources. A search result that never turned into a citation in the answer shouldn't count as a cited source. The documentation also says API and interface outputs can differ in configuration and possibly in model. Since this guide is about what buyers see in Perplexity, treat the consumer interface as your reference. If you do collect API data, keep it as its own labeled stream and don't blend it into the same trendline.

You're done when two reviewers could look at the same run and give it the same labels.

Step 7: Calculate rates and competitor citation share

Now turn the log into a few numbers. DeepSmith reports Mention Rate, Citation Rate, Share of Voice, competitor scores, and cited pages for Perplexity, but the exact denominators aren't spelled out in the product materials. So treat the formulas below as plain reporting conventions you can use and explain yourself, not as DeepSmith's own formula.

  • Mention rate: valid prompt-runs where your brand is mentioned, divided by all valid prompt-runs, times 100.
  • Citation rate: valid prompt-runs where at least one of your pages is cited, divided by all valid prompt-runs, times 100.
  • Competitive citation share: your citation events divided by all citation events earned by you and your chosen competitors, times 100. Say whether an event means one answer citing a brand or one individual cited page, and keep that unit the same.
  • Page-level citation share: citations of one of your pages, divided by all citations of your pages in the same scope, times 100. This shows which of your pages carry your citations. It isn't the same thing as competitor share.

Here's an example, with made-up numbers, just to show the arithmetic. Say you have 72 valid prompt-runs. Your brand is mentioned in 30 of them and your pages are cited in 18. Your mention rate is 30 divided by 72, which is 41.7%, and your citation rate is 18 divided by 72, which is 25%. If two competitors have 24 and 12 citation events, your competitive citation share is 18 divided by 18 plus 24 plus 12, which is 33.3%. These figures are invented to show the math. They aren't DeepSmith customer results or industry benchmarks.

Always report the prompt-level results next to the total, because a mix of prompts can hide big differences. If you want a deeper walk-through of the share side, see how to calculate AI share of voice. The roundup of tools that track competitor share of voice in Perplexity shows what other options report.

You're done when every percentage in your report has a stated numerator, denominator, time window, and competitor set.

Common mistake: calling page-level citation share "share of voice" without saying what the denominator covers. Competitors, all citations, and your own pages only are three very different answers.

Step 8: Compare like with like and report the uncertainty

When you compare two periods, check that the prompt-set version, locale, search mode, source mode, account state, reporting window, coding rules, and competitor list all match. If any of them changed, mark the change on your chart or start a new baseline. Show the overall result and the prompt-level changes together. For citations, name which of your pages earned them and which competitor pages showed up, not just a single percentage.

It also helps to know how much Perplexity brand visibility can move on its own. A 2026 study of repeated AI-search runs included Perplexity and compared same-prompt results within 24 hours. It reported a mean source Jaccard similarity of 0.282 and a brand Jaccard similarity of 0.492 for Perplexity. In plain terms, the cited sources overlapped less between runs than the brands named did. Those are measures from that study only. The brand analysis covered three campaigns that met its detection threshold, the source analysis left out runs with no extracted citations, and the data came from web interfaces in a Swiss locale. They're not a benchmark for your brand and not a suggested sample size.

What you can take from it is a habit: report the number of runs, the dates, and any changed conditions beside every rate. A small gap between two short windows may be ordinary run-to-run variation. If you haven't calculated statistical intervals, call the report descriptive, and don't present small swings as conclusions.

Referral traffic is a separate view. In GA4, open Reports, then Acquisition, then Traffic acquisition, and look at Session source or Session source / medium to find Perplexity. Sessions are visits that started on your site. They don't count how often Perplexity named or cited you, so they sit beside your visibility numbers and don't replace them.

You're done when a reader can compare two periods without mistaking a changed setup, a different sample size, or referral traffic for a change in answer visibility.

Where a Perplexity tracking tool fits

Everything above works by hand, and for a small prompt set that's a perfectly good place to start. The manual route gets slow once you have dozens of prompts, several competitors, and a weekly rhythm, because each run needs a fresh session and a logged row.

That's the part a Perplexity tracking tool can take over. DeepSmith's AI Visibility module runs your tracked prompts on a schedule, keeps the full answer history, tracks mentions and citations separately, and shows Perplexity-specific views, competitor comparisons, and the pages that get cited along with the prompts behind them. You still define the prompt set and the scope yourself, since the tool measures what you tell it to. It saves you the collecting and the log keeping.

Perplexity tracking is available on the Grow plan at $199 a month, or $160 a month billed annually, with 100 tracked prompts and Perplexity alongside ChatGPT. Scale is $399 a month, or $299 billed annually, with 200 prompts and Gemini added. Enterprise is custom. Pro doesn't include Perplexity. Every plan starts with a 7-day free trial and has no long-term contract. If you're comparing options, the best Perplexity optimization tools roundup is a decent place to see what's out there, and the best AI visibility tools roundup covers the wider field.

What to do next

Start small this week. Write your scope notes, freeze a first prompt set of ten or so questions, and run each one three times in fresh sessions. Log everything in the table above. After a month you'll have a baseline you can trust, and a clearer picture of where Perplexity names you and where it doesn't.

Once you know where you stand, the next question is what to do about it, and that's covered in how to improve your visibility in Perplexity. If you'd rather skip the manual logging and have the runs collected on a schedule, you can start a free DeepSmith trial and set up your prompts there.

Frequently asked questions

How many times should I run each Perplexity prompt?

Run each prompt at least three times in a collection window as a practical floor, and report the run count. More runs can show you how much the answers vary, but no number guarantees a stable result. Keep the wording and the run conditions fixed.

Does a brand mention count as a citation?

No. A mention means the answer text names your brand. A citation means the answer links to a page on a domain you've defined as yours. Record the two separately.

Can I use Perplexity API results to stand in for what people see in the interface?

Not without a label. Perplexity says API and interface outputs can differ because of configuration and possibly the model. Keep API runs and interface runs apart, or start a new baseline when you change the method.

Does Perplexity referral traffic show my brand visibility?

No. Referral sessions show visits that reached your site. They don't measure how often Perplexity named or cited your brand in answers, so use them as a separate, downstream view.