You track Perplexity visibility by writing down a fixed set of questions your customers ask, sending them to Perplexity on a schedule (by hand, through its API or with a tracker), and recording for each answer whether your brand is mentioned, whether your page is cited as a source, in which position, and who else is cited. Perplexity has no official ranking report, so sampling is the only method. Done consistently, it gives you a share of voice per topic that you can act on, and a list of questions where a competitor is the source instead of you.
Why Perplexity is the easiest AI engine to measure
Of all the AI answer engines, Perplexity is the most transparent. It searches the web for almost every question, it shows numbered citations next to its sentences, and it lists the sources it used. New or updated pages can show up in answers within days. That makes it a good place to learn the mechanics of AI visibility before you tackle ChatGPT or Google AI Overviews, where the picture is murkier.
Two things still make it harder than a Google ranking. Answers vary between runs, because the model rewrites each time and the search results shift. And there are two different kinds of visibility: a mention (your brand name appears in the text) and a citation (your page is one of the numbered sources). A brand can be mentioned without being cited, and cited without being named in the text. Track both, separately.
Step 1: Build the prompt set
Everything depends on the questions you measure. Write 20 to 50 prompts per topic that matter for your business, phrased the way a person types them into a chat box rather than the way they type them into Google. Cover three intents:
- Orientation: "what should I look for in an accounting tool as a freelancer"
- Comparison: "which accounting tools are best for freelancers in the Netherlands"
- Purchase: "which accounting tool integrates with Dutch banks and does VAT returns"
Include prompts that do not contain your brand name; those are the ones where you can win or lose. Add a few competitor-branded prompts ("is competitor X good for freelancers") to see how the engine talks about them. Do not fill the set with prompts about your own brand; those measure reputation, not discovery. Keep the list under version control: when you change the set, your history becomes harder to compare.
Step 2: Decide the sampling protocol
Perplexity personalises less than ChatGPT, but consistency still matters. Fix the protocol and write it down:
- Session: logged out or in a fresh thread, no earlier conversation in the same thread.
- Location and language: the country and language of your market. A prompt in Dutch from the Netherlands gets different sources than the same question in English.
- Mode: decide whether you measure the default search mode or the deeper research mode, and keep it the same. They pull different numbers of sources.
- Repetitions: run each prompt at least three times per measurement, because a single answer is a coin flip.
- Cadence: weekly. Answers change as pages get published and updated.
- Storage: keep the full answer text, the list of sources with their positions, the date and the mode, so every number can be checked later.
At scale, the API is the practical route: the Sonar models return citations with each answer, which makes it possible to run a few hundred prompts a week from a script. Manual sampling in a spreadsheet works fine up to a few dozen prompts.
Step 3: Record the right fields
| Field | Definition | Why it matters |
|---|---|---|
| Mentioned | Your brand name appears in the answer text | Awareness; the engine "knows" you |
| Cited | Your domain appears in the numbered sources | Traffic; the engine trusts your page |
| Position | The number of your citation (1 is first) | Early citations get more clicks |
| Cited URL | The exact page that was cited | Tells you which page format works |
| Competitors cited | Which other domains appear as sources | Your comparison set |
| Third-party sources | Review sites, forums, directories, Wikipedia | Where you may need a presence |
| Answer sentiment | Neutral, positive, negative, or wrong facts | Wrong facts are a content task |
Record these per prompt, per run. Aggregate afterwards, never during.
Step 4: Score it
Four numbers cover most of what you need:
- Mention rate: the share of runs in which your brand is named.
- Citation rate: the share of runs in which your domain is cited.
- Average citation position: across the runs where you are cited.
- Share of voice: your citations divided by all citations to you and your defined competitor set, per topic.
An example, with made-up numbers to show the shape of the table (this is illustrative data, not a measurement):
| Topic (example data) | Prompts | Mention rate | Citation rate | Share of voice |
|---|---|---|---|---|
| Accounting for freelancers | 30 | 40% | 23% | 18% |
| VAT returns | 20 | 10% | 5% | 6% |
| Invoicing software | 25 | 55% | 41% | 33% |
The table already tells a story: the second topic is where the work is.
Step 5: Analyse who gets cited and why
For every prompt where you are absent, look at the sources that were cited. Patterns show up quickly:
- Pages that answer the question directly in the first paragraph, then go deeper.
- Lists, tables and definitions the model can lift without rewriting.
- Recent pages, or pages with a visible updated date.
- Clear entities: who the company is, what the product does, for whom, where.
- Third-party sources such as comparison sites, forums and directories that describe your category, with or without you in them.
- Your own help centre or documentation, which is often cited before your marketing pages.
Turn each pattern into a task: a new page, a rewrite of an existing one, a fact correction, or a listing on a third-party source where the engine keeps looking.
Step 6: Act, then re-measure
Visibility work on Perplexity is content work. Write the page that answers the question you are missing, restructure the page that exists but is not cited, make your entity information consistent across the site, and check that your robots rules allow Perplexity's crawler if you want to be cited. Publish, wait a week or two, run the prompt set again and compare. Keep a log of what you changed and when, so that a shift in citation rate can be tied to an action.
Common mistakes
- Measuring once and calling it a baseline. One measurement is noise.
- Filling the prompt set with your own brand name.
- Measuring while logged in, with a thread full of earlier questions.
- Running each prompt once.
- Mixing mentions and citations into a single "visibility" number.
- Chasing prompts nobody asks because they are easy to win.
By hand, by script or with a tracker
A spreadsheet is fine to start: pick 30 prompts, run each three times a week, record the fields above. It takes an hour or two per week and teaches you more about your market than any dashboard. Once you pass a few topics, a script against the API or a dedicated tracker takes over the repetition and keeps the history. If you want a quick first reading before you build anything, the free AI Visibility Check shows whether Perplexity, ChatGPT and Gemini mention your brand or a competitor for a handful of questions.
Conclusion
Tracking Perplexity visibility comes down to discipline: a fixed prompt set, a fixed protocol, repeated sampling, and a clean separation between mentions and citations. The data tells you which questions you own, which a competitor owns, and which page formats the engine prefers, and each of those is a concrete content task. If you would rather have this measured across six engines every week and turned into a content plan automatically, see how AI visibility works in Traze.