Key takeaways
- Perplexity does not rank pages. You track mentions, citations, and links across a fixed set of buyer prompts, so the prompt list is the foundation of everything else.
- Start with 20 to 50 prompts, run each several times, and average the results. Below 20 prompts, single-response noise swamps the signal.
- Manual tracking in a spreadsheet works for a month or two. After that, automate with a tracker that records mentions, citations, position, and sentiment against your competitors.
- Perplexity is a good test bench: Promptwatch Data shows it cites almost exactly 10 sources per answer, so changes in your citation pickup usually come from your content, not engine noise.
- Tracking only matters if it feeds action: refresh pages, fix crawler access, and get named on third-party sources Perplexity trusts.
Why Perplexity needs its own tracking setup
Perplexity is not a list of blue links. It retrieves live web content, synthesizes an answer, and cites the pages it used. That means two things for tracking. First, a page has to be indexed and retrievable before it can be cited at all. Second, a top-three Google ranking does not guarantee a mention. Google rewards keywords, backlinks, and technical signals. Perplexity's synthesis step rewards direct answers, specific facts, clear entity definitions, and passages that are easy to lift.
There is also no "position 1" to check. So you measure three different things:
| Metric | What it is | Why it matters |
|---|---|---|
| Mention | Your brand name appears in the answer text, with or without a link | Builds entity awareness, rarely drives clicks on its own |
| Citation | One of your pages appears in the source list | Shows which content Perplexity actually trusts and uses |
| Link or referral | A user clicks through to your site | The only one that shows up in your analytics |
Keep these separate in your reporting. A brand mentioned in 40% of answers but cited in 5% has a content or crawlability problem, not an awareness problem.
One useful fact from Promptwatch's average sources per response data: Perplexity cites almost exactly 10 sources per answer, day after day, with barely a decimal of movement. That report dates from September 2025, so treat the exact figure as a baseline rather than gospel. But the stability is the point. ChatGPT cites about 5 sources per web-search answer, and Copilot swung from under 2 to nearly 17 within weeks. Perplexity's steady 10 slots make it the cleanest place to see whether your content changes actually moved something.
Step 1: Build your brand and competitor inventory
Before you track anything, write down what you are looking for.
Brand terms
List your company name, product names, abbreviations, legacy names if you rebranded, and the misspellings customers actually use. If you are "DataSync Pro," you also want "DataSync," "Data Sync," and "DataSyncPro." Most teams end up with 8 to 12 variations. Whatever tool you use later, this list decides what counts as a mention.
Competitors
Pick 3 to 5 direct competitors. These are the brands that appear alongside, or instead of, yours when someone asks Perplexity for recommendations. Run a few category queries yourself and see who shows up. Those are your benchmarks.
Check that Perplexity can reach you at all
Before measuring anything, audit robots.txt, CDN, and WAF rules. If you block PerplexityBot or Perplexity-User, no amount of content work will fix your numbers. Promptwatch Data's AI crawler traffic report (June 8 to September 6, 2026) counts only IP-verified requests, and its advice is simple: compare your own crawler logs against the provider mix. If a provider is a large share of the benchmark but near zero in your logs, you are probably blocked. In that period OpenAI made up 79.8% of verified AI crawler requests, down from 94.8% in early June, with the rest split among Anthropic, Google, Perplexity, and Mistral.
Step 2: Create your prompt library
This is where most teams underinvest. Do not pull prompts from keyword volume tools. Perplexity users type full, conversational questions, and a bare keyword like "compliance software" gets interpreted very differently than it would on Google.
Source prompts from:
- Sales call recordings and CRM notes
- Support tickets
- Competitor comparison searches
- Question-style queries in your Search Console data
Good prompts sound like "What is the best compliance tool for a Series A fintech?" or "Compare [Your Brand] vs. [Competitor] for mid-market sales teams."
Map them across the buyer journey so you can see where visibility breaks down:
| Stage | Example prompt pattern | What it tells you |
|---|---|---|
| Awareness | "What is [category]?" | Whether you are treated as a thought leader |
| Consideration | "Best [tools] for [use case]" | Competitive positioning |
| Decision | "[Competitor] alternatives" or "[A] vs [B]" | Whether you make the shortlist |
| Problem-solving | "How to [fix problem]" | Whether your content answers real questions |
Start with 20 to 50 prompts. Keep branded and non-branded prompts in separate groups, because mixing them in one KPI hides the real story. Branded prompts will always look healthy.
Step 3: Run a manual baseline
Before paying for anything, do one honest manual pass. It takes an afternoon and teaches you what the data looks like.
- Create a fresh Perplexity account used only for tracking, so personal history does not contaminate results.
- Build a spreadsheet with columns for date, prompt, brand mentioned (Y/N), cited (Y/N), cited URLs, sentiment, competitors mentioned, and notes.
- Run each prompt 3 to 5 times in fresh sessions and average the results. Single responses vary.
- Open the cited pages for each prompt, especially where competitors appear and you do not.
From this you can calculate your answer share of voice. If you track 50 prompts and appear in 8 responses, that is 16%. Run the same set weekly and you have a trend.
The honest problem: 30 prompts run 3 times each, weekly, is 360 manual checks a month. Most people quit by month three. Use the baseline to learn, then automate.
Step 4: Automate with a tracker
Automated tracking runs the same prompts on the same schedule, parses the answers into structured data, and compares you to competitors in the same run. That consistency is what turns noise into a trend.
What to look for in a tool that covers Perplexity:
- Real interface data rather than only API output, since user-facing answers and citations can differ
- Competitor benchmarking on the same prompts
- Citation-level data (which URLs, which domains), not just mention counts
- Alerts for meaningful changes, not every single mention
- Scheduled reports you can send to stakeholders
- Some way to act on the findings
Promptwatch is the one I would put at the top for teams that want to go beyond a dashboard. It monitors Perplexity's actual interface alongside ChatGPT, Google AI Overviews, Gemini, Claude, and others, and adds IP-verified crawler logs, so you can see whether PerplexityBot and Perplexity-User are actually reading your pages. It also has a Content Agent and an Action Board that turn visibility gaps into publishable work, then track whether the new pages get crawled and cited.
Other trackers that cover Perplexity and are worth a look:
| Tool | Best for | Notes |
|---|---|---|
| Promptwatch | Tracking plus fixing: content gaps, crawler logs, content generation | Starts free (ChatGPT only), paid plans from $95/mo; check promptwatch.com/pricing |
| Profound | Enterprise monitoring | Strong data, higher price point |
| Peec AI | Multi-language tracking | Prompt tracker, no content tooling |
| Otterly.AI | Budget monitoring | Simple setup, monitoring only |
| AthenaHQ | Multi-engine tracking | Monitoring focused |
| Scrunch AI | Agency workflows | Multi-client reporting |
If you want to browse more options, the GEO software directory at bestgeosoftware.com lists many more.
Set frequency to match your capacity
Daily tracking makes sense in news-heavy categories. For most B2B teams, a few runs a week is plenty. The real rule: match tracking cadence to how fast you can publish. If you ship one article a month, daily data just gives you more than you can act on.
Set alerts that matter
Alert on new prompts where you appear, prompts where a competitor starts displacing you, sentiment shifts, and big visibility score changes. Do not alert on every mention. You will learn to ignore it.
Step 5: Read the numbers properly
Data is only useful if you know what to conclude from it. Read these together:
- Visibility rate: share of tracked prompts where you appear. 15 of 50 is 30%.
- Share of voice: your appearances compared to competitors across the same prompts. If you appear in 23 of 100 and your closest rival in 41, they hold nearly twice your presence.
- Position: brands named first usually carry stronger authority signals than brands named fourth in a list.
- Citation quality: "according to [Your Brand]" is a strong citation. A bare name in a list of five is a weak one.
- Sentiment: positive, neutral, negative, plus a fourth bucket for mentioned incorrectly. Outdated pricing or a wrong category in an answer points to a page that needs updating.
A quick diagnostic table:
| Symptom | Likely problem | First fix |
|---|---|---|
| Low share of voice | Coverage | Publish content for missing prompts |
| Mentioned late in answers | Prominence | Tighten answer-first structure, add comparisons |
| Mentioned, never cited | Content or crawlability | Check bot access, restructure pages |
| Weak sentiment | Reputation | Fix third-party reviews and sources |
| Wrong facts | Stale sources | Update pages and directory listings |
Also look at which third-party domains Perplexity leans on for your category. If it keeps citing the same review site or roundup, that is where your brand needs to be.
Step 6: Turn gaps into content and off-site work
Every prompt where competitors appear and you do not is a content brief waiting to be written. Before writing, run the prompt in Perplexity, open the pages it cites, and study their format, depth, and how directly they answer the question.
Fix your highest-value pages first
Do not rewrite everything. Pick the ten pages tied to revenue and:
- Make the first two sentences answer the question in the title
- Rewrite H2s as the questions buyers actually type
- Add a comparison table or numbered process, since tables get pulled intact
- Put real figures in prose: prices, timeframes, versions, dates
- Show a visible last-updated date and set dateModified in your Article schema
- Keep each page on one canonical URL that returns 200
Freshness matters. A 2024 "best X" guide with an outdated comparison table is exactly the kind of page that gets displaced.
Get named elsewhere
Perplexity looks for consensus. A product mentioned on Reddit, a review platform, a directory, and two independent roundups reads as established. A product mentioned only on its own site reads as a claim. Priorities, in order: category review platforms, genuine participation in the few subreddits where your category is discussed, independent roundups, and customer reviews that name specific benefits.
On social, be realistic. Promptwatch's October 2026 social citations data (Sep 28 to Oct 4, 2026) shows Perplexity gets only 2.7% of its citations from social platforms: LinkedIn and Reddit tie at 0.86%, YouTube is at 0.79%, and X peaks around 0.1%. The takeaway is to publish on LinkedIn as well as Reddit and leave X out of your AI search plans. For comparison, ChatGPT got 0.46% of citations from social platforms in that week and Google AI Overviews got 12.2%.
Sell products? Check Perplexity Shopping
If you are a US-based e-commerce brand, Perplexity's merchant program takes structured catalog data, so prices and stock status are accurate instead of scraped. Missing GTINs and empty attribute fields are common reasons competitors get recommended instead of you. Confirm current eligibility on Perplexity's own site.
Step 7: Measure impact and set a review rhythm
Tracking without a review habit is a screenshot collection. A workable 90-day plan:
| Window | Work | Expected signal |
|---|---|---|
| Week 1 | Robots.txt and WAF audit, set up a GA4 channel group for AI referrals, log 20 or more buyer prompts, record baseline | Baseline citation count |
| Weeks 2 to 5 | Restructure top pages, fix factual errors, publish gap content | Early movement on prompts you targeted |
| Weeks 6 to 10 | Reddit and review platform presence, a couple of roundup placements | First citation movement on tracked prompts |
| Weeks 10 to 13 | Re-run the full prompt set, compare against baseline, refresh pages that missed | Referral sessions and conversion rate |
Book a 30-minute weekly review. Look at what gained, what lost, which pages got cited, and what you published since the last check. Annotate your trend charts with publishing dates so a visibility change can be tied to a cause.
If you also want to see whether Perplexity visits your site and what it reads, a crawler log tool is worth adding. Promptwatch's Agent Analytics covers this, and DarkVisitors is a lighter option focused on bots and agents.
Common mistakes
- Tracking fewer than 20 prompts, then reacting to one odd answer
- Running each prompt once per cycle instead of averaging several runs
- Mixing branded and non-branded prompts in a single score
- Building the prompt list from keyword tools rather than real customer language
- Setting up tracking, getting excited, and never opening the dashboard again
- Optimizing only your own site and ignoring the third-party sources Perplexity cites
What to do this week
Write down your brand variations and four competitors. Draft 25 buyer prompts from sales and support conversations. Run them manually, three times each, and fill in the spreadsheet. Check your robots.txt and CDN rules for PerplexityBot. Then pick a tracker and move the prompt list into it, so next month you are comparing numbers instead of guessing.
A tight list of 20 well-chosen prompts, reviewed every week, will teach you more than 100 prompts nobody looks at.