Automate Perplexity Brand Visibility Trend Reports

Automate reports brand visibility trends perplexity is a search for one job: a repeatable report showing whether a brand is named, cited and recommended in Perplexity answers over time. Fix the questions, run them on a fixed schedule, save every answer with its source links, and report mentions, citations and recommendations as three separate lines.

A trend report is only as good as its questions. If the prompt set changes between runs, the chart compares different questions and says nothing about visibility. Prime AI Visibility saves answers for chosen questions, and this guide shows the same method for a spreadsheet or a script, with a labelled illustrative dataset.

Who is this guide for?

It is written for brand analysts, SEO managers, agency reporting leads and marketing heads who have been asked for a monthly Perplexity number. It works for a small team with a sheet and for an enterprise team with a data warehouse.

Key takeaways:

• Measuring existing visibility is a different job from improving it. Report first, then act.
• A mention, a citation and a recommendation are three different facts.
• Fix the prompt set, repeat each prompt, and write down the denominator.
• Keep failed runs out of the rates and show how many there were.

What does a Perplexity brand trend report actually measure?

It answers four questions: which buyer questions should name this brand, how often they do, which sources the answer links when they do, and whether the change since last period is reliable. It describes what answers displayed on the dates you ran them. It does not explain why an answer chose a source, and it does not predict what a buyer will see tomorrow.

The three events must stay apart:

Event	What counts	What it does not prove
Brand mention	The answer text names the brand	That any page was linked
Owned citation	A source link points to your domain	That the brand was recommended
Recommendation	The answer clearly suggests the brand as a choice	That the suggestion is accurate

Prime AI Visibility's measurement metrics page uses the same separation: Brand named, Website cited, Source links and Explicit recommendation. Keep these as columns, not one blended score.

What is the Perplexity Evidence Ledger?

This is the framework in this guide, seven steps in order:

1. Prompt cohort. A frozen list of buyer questions with tags.
2. Fixed-run configuration. The same settings, account type, country and repeats every time.
3. Answer snapshots. The full text of each answer, saved with its time.
4. Citation extraction. Every source link, split into owned, third-party and competitor.
5. Denominator and variance. Valid runs counted, errors excluded, repeats compared.
6. Trend interpretation. Rates with a range, annotated with what changed.
7. Remediation. Only now decide which pages or sources to work on.

How do you set a stable category and buyer-prompt universe?

Choose 20 to 40 questions a real buyer asks and freeze the wording. Tag each one so the report can slice the results:

• Category: the product or service area.
• Persona: who is asking, such as a buyer, an analyst or a technical reviewer.
• Region: the country the answer is requested for.
• Buying stage: learning, comparing or deciding.
• Brand or non-brand: whether the question names you.

Do not edit the list during a reporting period. If you must add a question, add it to a new cohort and report the old cohort separately, so a trend is never a comparison of different lists. Our guide to monitoring brand mentions in ChatGPT uses the same cohort rule for another engine.

Which run settings, repeats and sample window should you use?

Answers vary between runs, so a single run per question is an anecdote. Run each prompt the same number of times each period. Three runs is a practical minimum for a monthly report, and more help where answers swing. How many repeats you need depends on how unstable your own answers are, and the volatility measure below tells you.

Fix what you can and record what you cannot:

• Same account type, country and language on every run.
• Same day window each period, such as the first week of the month.
• Same time of day if your tool allows it.
• The model or mode shown, if the product displays it, noted as shown.

If you use Perplexity's API instead of the app, remember the two can differ, so never mix them in one trend line. Perplexity's developer documentation describes an Agent API that returns web-grounded answers with built-in citations, and a search results item listing the sources. That makes a script possible, and you should test the output on your own questions before trusting it.

What should you store for every answer?

One row per run. These fields are the minimum:

Field	Why you need it
Prompt ID and exact text	To prove the question never changed
Run date and time	To place the run in a period
Run number	To compare repeats
Full answer text	To recheck mention and accuracy later
Brands named	For mention rate and share of voice
Source URLs in order	For citations and first appearance
Recommendation (yes or no) and extract	For the recommendation line
Accuracy note	To flag a wrong claim about you
Run status	To exclude errors

Keep the source URL ledger as its own table: one row per link, with the prompt ID, run, position, domain and whether it is yours, a competitor's or neutral. A ledger lets you ask later which domains an answer relied on and whether they changed.

How do you calculate visibility and citation share separately?

Choose the denominator before you look at the data. The usual choice is valid runs: runs that completed and returned an answer.

Measure	Formula
Mention rate	valid runs naming the brand ÷ valid runs
Owned citation rate	valid runs linking your domain ÷ valid runs
Recommendation rate	valid runs recommending the brand ÷ valid runs
Source concentration	links from the top three domains ÷ all links
Share of voice	runs naming the brand ÷ runs naming any tracked brand
First appearance	average position of the brand in answer text
Citation decay	share of last period's cited sources missing now
Volatility	share of prompts with mixed results across repeats

For more on share of voice, read AI search share of voice, and for the decay measure, what is citation decay. Share of voice is an analyst's calculation from named brands, and the definition you choose belongs in the report.

What does an illustrative dataset look like?

The numbers below are invented to show the arithmetic. They are not customer data and not a Prime result. Assume 20 saved prompts, each run three times in three monthly periods, so 60 runs per period.

Measure	Period 1	Period 2	Period 3
Valid runs (of 60)	55	57	56
Mention rate	30.9%	35.1%	44.6%
Owned citation rate	14.5%	17.5%	30.4%
Recommendation rate	7.3%	8.8%	12.5%

The valid-run counts show how many runs were excluded as errors: 5, 3 and 4. Each rate divides by valid runs, not by 60.

Now read the table carefully. A rough 95 percent range around the mention rate is about plus or minus 12 points in Period 1 and plus or minus 13 in Period 3, so the two ranges overlap. The owned citation rate moved from 14.5 to 30.4 percent with ranges that only just overlap. Repeats of the same prompt are not independent, so these ranges are loose, and the honest label is direction, not proof. In the same data, assume the top three cited domains held 54 of 90 links in Period 1 (60 percent) and 62 of 110 links in Period 3 (56.4 percent), so the sources stayed concentrated. For volatility, assume that in Period 3, 6 prompts named the brand in all three runs, 7 in some and 7 in none. A third of the prompts are unstable, and that is why one run per prompt would mislead.

How would you read a before-and-after source improvement?

Suppose that between Period 2 and Period 3 the team published one clear page that answers a question the brand was missing from, and added it to its own site's internal links. The illustrative data above show the owned citation rate rising. A careful report states what was done, on which date, and what the rates did. It does not state that the page caused the change.

A short quality check protects the claim:

1. Open the answers in which the owned citation appeared and confirm the link points to the new page.
2. Confirm the prompt set and settings did not change.
3. Compare against questions that were not touched, which act as a rough control.
4. Check that competitors did not lose a source in the same window.
5. Read a sample of answers for accuracy.

Repeated sampling does not guarantee what a real buyer sees, and no edit guarantees an answer will cite you.

How do you detect missing runs and unstable sources?

Mark every run with a status: completed, empty, error or timed out. Exclude non-completed runs from the rates and show their count in the chart footnote. If more than a small share of runs fail in one period, say so, because the denominator is thinner.

Source links can also be unstable. Save each URL as displayed, then check whether it resolves, redirects or has changed. Record the domain separately from the full URL, so a page move does not look like a source change. A rise in redirected or dead cited links is a finding of its own.

How should you build trend views and show confidence?

Use three views: the three event rates by period, a source table showing the top domains and what changed, and a prompt table showing which questions gained or lost the brand. On every chart show the valid-run count and a range, and annotate dates when pages, tags or campaigns changed. Our marketing dashboard for AI search guide covers where each number comes from, and the AI visibility KPIs article defines the answer-side measures.

Label each trend with a confidence word: clear when ranges do not overlap across several periods, directional when they overlap but the movement is consistent, and unclear when runs are few or results swing. Do not call a single-period change a trend.

Can you automate reports brand visibility trends perplexity needs, and what is native versus user-run?

Separate the layers:

• Manual sheet. You run the questions by hand and paste answers into the table above. It is free and slow, and it is the best way to learn the method. See tracking AI visibility manually vs with a tool.
• User-run script. If you call Perplexity's API on a scheduler you control, you own the schedule, the storage and the checks. Perplexity's documentation describes an API with citations. Test a script on your own questions before trusting it.
• Product-run schedule. A monitoring product runs the saved questions for you. Prime AI Visibility's pages describe monthly checks as the default, with optional weekly and temporary daily schedules and on-demand checks that use credits, and saved PDF and CSV exports at no credit cost. Its pages do not describe a public API.
• Perplexity's own recurring reports. This guide does not assume that Perplexity can email a recurring brand report, and nothing here relies on it. Check Perplexity's current documentation for your plan before you design around one.

Whichever route you use, a person should read a sample of the answers each period. Automated extraction can misread a name or a link.

Where does Prime AI Visibility fit?

Prime AI Visibility is an AI Search Visibility Platform that saves answers for the questions and AI tools you choose, and records Brand named, Website cited, Source links and Explicit recommendation as separate facts. Read how it works for the collection method and the platform features page for what a workspace includes. Its AI Overview-style preview is built from Google-grounded answers and is not a captured live Google AI Overview. The Primey AI strategist page describes the planning layer built on saved answers.

To compare monitoring products before you pick one, use compare AI search trackers. To start a baseline, use Prime as an AI visibility checker for five questions, or as an AI visibility tracker once the question set is stable. This is a disclosed affiliated product and it describes what answers displayed without explaining why. Prime AI Visibility has a working relationship with Percepture, which describes how to track Perplexity brand mentions its wider AI search visibility strategy and its generative engine optimization services for the implementation work.

How do you share a governance-friendly report?

• Put the prompt list, run settings and definitions on an appendix page.
• Show valid-run counts, excluded runs and ranges next to every rate.
• Keep the answer archive so any number can be traced to its text.
• Mark who reviewed the sample and on what date.
• Remove personal data and keep client records out of shared exports.

What are the common dashboard pitfalls?

• Changing the prompt list and still calling it a trend.
• Blending mention, citation and recommendation into one score.
• Dividing by all runs when some failed.
• Reading one lucky answer as growth.
• Comparing the app in one period with the API in another.
• Treating share of voice from a small list as market share.

See also AI brand visibility mistakes for a broader list.

What is the decision checklist?

• The prompt list is frozen and tagged.
• Each prompt runs the same number of times each period.
• The denominator is valid runs and errors are counted.
• Mentions, citations and recommendations are separate columns.
• Source URLs are stored as a ledger.
• Every chart shows a range and an annotation.
• A person reads a sample of answers.

What not to believe: a single-run chart, a score without its answers behind it, a causal claim from one edit, and any promise of a fixed placement in Perplexity or any other engine.