Key takeaways
- Perplexity has no official brand report. Every "Perplexity brand report" comes from a third-party tracker that runs prompts and parses the answers, so the definitions vary by vendor.
- Read mention rate first, then citation share. A change in one usually explains the change in the other.
- Citation rank is a position in a list of roughly 10 slots, not a quality score. Perplexity publishes no ranking formula.
- Share of voice has at least three formulas (mention-based, citation-based, position-weighted), and the same brand can score 20%, 16.8% and 31.4% on the same data.
- Treat every number as directional. Retrieval is non-deterministic and cited domains drift heavily month to month.
First, what "Perplexity brand report" actually means
I went looking for an official report from Perplexity and found nothing. There is no brand dashboard on Perplexity's side that tells you how often you're named. What people call a Perplexity brand report is output from a tracking tool that sends a fixed set of prompts to Perplexity on a schedule, saves the answers, and counts what shows up.
That matters because every number in the report depends on choices the vendor made: which prompts, how many runs, how a mention is detected, how sentiment is scored. Two tools can show you different shares for the same brand on the same day, and neither is lying. So before you read a single chart, find the methodology page and answer three questions. Which prompts were used? How many times was each one run? Which competitors form the denominator?
If you're still choosing a tool, the directories at ai-rank-tools.com and bestgeosoftware.com list most of the options. Tools from our catalog that cover Perplexity include Promptwatch, LLM Pulse, Peec AI, Otterly.AI, Keyword.com and Rankability.
How Perplexity builds an answer (and why it shapes the report)
LLM Pulse describes Perplexity's pipeline in four steps: it rewrites the query, retrieves from the web, selects sources, and synthesizes an answer with numbered inline citations. Two consequences matter when you read a report.
First, citations are visible by default. That makes Perplexity easier to measure than most chat surfaces, because every cited URL is a data point.
Second, retrieval is non-deterministic. The same question can return slightly different sources on two runs. One snapshot is a sample, not a fact. Good reports run each prompt several times and show trends over weeks. The cloro team, for example, recommends at least five runs per prompt, and ten for weekly reporting.
The four metrics, one at a time
Mentions (mention rate)
Mention rate is the share of tracked prompts where your brand name appears anywhere in the answer text. It answers a basic question: are we in the answer at all?
It is the loosest metric. A mention can be glowing, lukewarm, or a warning. It also doesn't require a link, so a high mention rate with no citations usually means Perplexity knows of you from other people's pages but doesn't think yours is worth linking.
Watch out for branded prompts. If you include queries that already contain your name, you'll have a near-100% mention rate and learn nothing. Build the report around the questions a buyer asks before they know you exist.
Citation rank
This is the metric people misread most. Perplexity shows numbered sources next to the answer. A tracker records where your URL sits in that list and calls it citation rank or citation position.
It is a presentation order. Perplexity publishes no formula for it, and LLM Pulse explicitly warns against treating citation number as an authority score. Position 2 is not twice as good as position 4 in any way anyone can prove.
Context helps here. Promptwatch Data's average sources per response report finds Perplexity cites about 10 sources per answer, almost exactly, every day. It is the most predictable engine in that dataset. ChatGPT cites about 5 per web-search answer and Google AI Overviews about 10. (An older third-party figure of roughly 22 citations per Perplexity answer circulates, but Promptwatch's daily tracking doesn't support it, so I'd use roughly 10.) So read your rank against about 10 slots. Being third of ten is good. Being ninth of ten is barely there.
The stable citation count also makes Perplexity a decent test bench. Because the number of slots barely moves, a change in how often you get picked up is more likely to reflect your content than engine noise.
Citation share and citation rate
Citation rate is the percentage of prompts where your domain appears in the citation list. Citation share is the proportion of all cited sources in an answer that belong to your domain. Both are stricter than mentions, and they track real authority signals better, because the model had to choose your page.
The gap between mention rate and citation rate is where the insight is. Asva AI's guidance is a good rule of thumb:
- Mention rate falling while citation share holds steady: the framing changed. The engine now answers with a competitor.
- Mention rate steady while citation share falls: a source you relied on is gone, blocked to a retrieval crawler, or outranked by something newer.
Sentiment
Sentiment is how Perplexity frames you when it names you, usually scored positive, neutral or negative per answer. Ask how your tool computes it. Some score the whole answer, some only the sentence around your brand, and some report a 0-100 scale where 50 is neutral. Those aren't comparable.
A falling score often means a negative review or an unflattering comparison page has become a cited source. Go find that page. Sentiment rarely drops on its own. And a rising mention rate with collapsing sentiment is a reputation problem that looks like a win on the dashboard.
Be skeptical of vendor sample numbers too. One tracker's demo page showed 68% mentions and 92% positive sentiment. Those are illustrative UI figures, not benchmarks.
Share of voice: three formulas, three answers
Share of voice is where reports disagree most. Digital Applied walks through LLM Pulse's worked example: a brand with 60 mentions out of 300 total scores:
| Formula | How it's calculated | Result | Standing |
|---|---|---|---|
| Mention-based | Your mentions / all brand mentions | 20% | Third |
| Position-weighted | Harmonic decay: 1.0, 0.50, 0.33 for positions 1, 2, 3 | 16.8% | Fourth |
| Citation-based | Your domain's citations / all citations | 31.4% | First |
Same brand, same data, three competitive positions. A report that quotes one number without naming the formula tells you almost nothing.
Which should you trust? It depends on what you're trying to learn.
| Objective | Formula | Read it as |
|---|---|---|
| Brand reach | Mention-based | How much of the conversation is about us |
| Prominence | Position-weighted | Are we named first or buried |
| Source authority | Citation-based | Is our content the page the model links |
A simple mention-share calculation from cloro shows the mechanics. If you have 218 mentions and the tracked brands together have 1,779, your share is 218 / 1,779, about 12.3%. Notice how much the denominator controls the answer. If you only count the top four competitors, your share looks higher than it really is.
Check the denominator
Before you celebrate or panic, confirm:
- Which competitors are in the set, and who chose them?
- Is it a fixed set of prompts, or does it change?
- Does the share include branded prompts?
- Is it Perplexity alone, or blended across engines?
Blended numbers hide a lot. Only 11% of domains cited by ChatGPT overlap with Perplexity's (a per-engine audit cited by Digital Applied), and Rankability reports that no pair of AI platforms shared more than 24.1% of cited pages. A strong blended score can be a weak Perplexity score.
A worked reading, step by step
Suppose your monthly report shows these Perplexity numbers over 150 prompts:
- Mention rate: 34%
- Citation rate: 12%
- Average citation position: 6.2
- Sentiment: 71/100
- Citation-based share of voice vs three competitors: 14%
Here is how I'd read it.
- Mention rate is decent but citation rate is much lower. Perplexity knows who you are and doesn't link you. Look at which prompts you're mentioned in without a citation, and what it cites instead.
- Average position of 6.2 out of roughly 10 slots means that when you do get cited, you're in the back half. Check whether the cited page is a thin one.
- Sentiment of 71 is healthy, so the problem is coverage and source authority, not reputation.
- A 14% citation share against three rivals is worth comparing prompt by prompt. Averages hide the clusters where you're absent entirely.
EPR's worked example makes the same point at the engine level. A vendor with 16% composite presence was at 22% on Perplexity but 6% on Claude, and 31% on comparison prompts but 4% on broad category prompts. The headline said "fine". The breakdown said "invisible to anyone who doesn't already know you."
Where citations come from: sources matter more than your site
A lot of brands assume Perplexity cites their own pages. Often it cites other people's pages that talk about them. Offsite coverage, reviews and comparison articles supply many of the citations.
Social is a small slice. In Promptwatch Data's social media citations by AI model report (Sep 28 to Oct 4, 2026), social platforms made up 2.7% of Perplexity's citations. LinkedIn and Reddit tied at 0.86%, and YouTube was 0.79%, down from 2.67%. For comparison, ChatGPT got 0.46% of its citations from social and Google AI Overviews got 12.2%. So don't over-invest in Reddit for Perplexity alone, and consider LinkedIn.
When the report lists your cited URLs and competitors' cited URLs, sort by source type. If competitors win on review sites and you only appear on your own blog, that's your gap.
Common ways to misread the report
- Treating one week as a trend. LLM Pulse and Digital Applied both flag drift: cited domain sets can shift 40 to 60% month over month in active categories. Use rolling averages.
- Reading citation rank as quality. It's list position.
- Tracking only your own brand. Without a competitor set there's no share of voice.
- Mixing branded and non-branded prompts.
- Ignoring the prompt list. Dan Taylor of SALT criticizes vendors for using small static prompt sets that create a "contrived and artificial environment". A 50 to 200 prompt panel built from real buyer questions (sales calls, Search Console, support tickets) is a sounder base.
- Comparing numbers across tools. Different prompt sets and sentiment models mean different outputs.
What to do with each metric
| Signal in the report | Likely meaning | First move |
|---|---|---|
| Low mention rate | Coverage problem | Find which prompts exclude you and who is named instead |
| Mentions but few citations | Model knows you, doesn't trust your pages | Publish citable, structured pages with original data |
| Late citation position | Prominence problem | Compare cited competitor pages for freshness and structure |
| Falling sentiment | Reputation problem | Find the negative source being cited |
| Falling share of voice | Competitor gained ground | Check what they published or earned on those prompts |
LLM Pulse's observations on what Perplexity tends to cite (recent content for time-sensitive queries, authoritative domains for ambiguous ones, structured pages like lists and tables) are tendencies, not rules. Test them against your own report rather than taking them on trust.
Choosing a tool to produce the report
Since the report is only as good as the tool, here's a quick map of options from our catalog. I haven't verified pricing for all of them, so confirm on each vendor's site.
| Tool | Angle | Worth knowing |
|---|---|---|
| Promptwatch | Tracks real Perplexity UI, plus crawler logs, AI traffic and content actions | Goes past monitoring into fixing gaps |
| LLM Pulse | Perplexity citation lists, share of voice, sentiment per prompt | Sentiment starts on its Growth plan, per TopCited |
| Keyword.com | Scraping-based Perplexity tracker | Scrapes the web UI rather than the API |
| Rankability | Citation analytics tied to content updates | Shows cited pages |
| Otterly.AI | Low-cost mention monitoring | Directional checks |
| Peec AI | Prompt-based visibility tracking | Marketing teams |
Otterly.AI
Two questions separate the good ones from the rest. Does it track the actual Perplexity interface rather than only an API? User-facing answers and citations can differ from API output, and Promptwatch monitors the real UI for that reason. And does it connect the report to action, such as which pages to update and whether the fix worked? A dashboard that only shows the score leaves you to work out the rest.
A minimum viable routine
- Build 50 to 200 non-branded prompts from real buyer questions and freeze the list.
- Pick three to five real competitors, the ones buyers actually compare you with.
- Run weekly, with multiple runs per prompt.
- Report four numbers: mention rate, citation rate, sentiment trend, and citation-based share of voice against the top three competitors.
- Name the formula in every report.
- Review monthly, and only act on moves that persist across several weeks.
The report won't tell you what Perplexity believes about your brand in some absolute sense. It tells you how you compare to rivals on a fixed question set, week over week. Used that way, it's useful. Used as a verdict, it will mislead you.
For more on how AI engines pick sources, browse the research at Promptwatch Data.