# Track your Perplexity share of citations and mentions

> Build a prompt set, measure mention and citation share in Perplexity against competitors, run repeat tests and automate with the API.

Source: https://mentionx.ai/learn/perplexity/perplexity-share-of-citations-tracking  
Author: Dipuck Jones, founder of MentionX  
Updated: 5 November 2026  
Level: Practitioner

**Share of citations measures how often Perplexity cites your domain for a fixed set of prompts that buyers ask, compared with competitors. To track it, build a prompt set of 30 to 100 questions, run each one several times per period, and record mentions, citations and position. Then report shares with ranges, not single runs. Automate the runs with the Agent API or Search API once the method is stable, and spot-check the app by hand, because API results and app results can differ.**

> **Key takeaways**
>
> - Track mention share (brand named in the answer) and citation share (your domain cited) separately.
> - Use a stable prompt set grouped by intent, and change it only on a schedule.
> - Run each prompt several times per period. Answers vary.
> - Report trends and ranges, not single numbers.
> - Automate with the API, but spot-check the app.

*Checked 5 November 2026 (CET) against the Perplexity Help Center, Perplexity blog, Perplexity API docs and the outlets cited in Sources. Perplexity renames and moves features often, so confirm labels and limits in your own account. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.*

## What should you measure?

| Metric | Definition | Why it matters |
|---|---|---|
| Mention share | Share of prompt runs where your brand is named in the answer | Recommendation visibility |
| Citation share (prompts) | Share of prompt runs citing at least one of your URLs | Source authority |
| Citation share (citations) | Your citations ÷ all citations across runs | Share of the source pool |
| Position | Order of first mention, or rank of first citation | Prominence |
| Sentiment or framing | Positive, neutral, negative or inaccurate | Message quality |
| Third-party citations about you | Citations of reviews or articles that mention you | Earned media impact |
| Competitor shares | The same metrics for competitors A-D | Context |

See [Measuring citation share](https://mentionx.ai/learn/measurement/measuring-citation-share), [Measure AI share of voice](https://mentionx.ai/learn/measurement/measure-ai-share-voice) and the glossary entry for [citation share](https://mentionx.ai/glossary/citation-share).

## How do you build a prompt set?

1. **Start from buyer questions.** Use sales calls, support tickets, search queries and your research from [lesson 6](https://mentionx.ai/learn/perplexity/perplexity-market-audience-research).
2. **Group them by intent:**
   - Category ("best AP automation software")
   - Problem ("how to stop duplicate invoice payments")
   - Comparison ("Ledgerline vs Competitor B")
   - Pricing
   - Integration
3. **Write prompts the way buyers ask them.** Mix short and conversational phrasings.
4. **Size the set at 30 to 100 prompts.** Fewer is noisy. More is costly to review.
5. **Freeze it.** Change prompts only quarterly and log changes, so trends stay comparable.

> **Pro move:** add branded prompts ("Is Ledgerline good?", "Ledgerline pricing") as a separate group. They show accuracy problems, not share, and should never be mixed into category share or they inflate it.

## Why run prompts more than once?

Perplexity answers change with live search results, model choice and phrasing. One run per prompt mixes real change with noise.

- Run each prompt **three to five times** per period. *Our view:* more for prompts that matter most.
- Report **the mean and the range**, such as "cited in 40% of runs (range 30-55% by week)".
- **Treat small changes as noise** until they persist across periods.

See [Statistical confidence in AI visibility tracking](https://mentionx.ai/learn/measurement/statistical-confidence-ai-visibility-tracking).

## Should you track by hand or automate?

| Approach | Good for | Limits |
|---|---|---|
| Manual in the app | Small sets, spotting framing and accuracy | Slow, inconsistent, personalised by account |
| Agent API | Answer text plus cited URLs, repeatable | Can differ from the app. Costs per call |
| Search API | Which pages rank for a query in Perplexity's index | Ranked results, not answers. A proxy only |
| Third-party trackers | Scale, dashboards, multi-engine | Method varies, so ask vendors how they sample |

## How do you automate with the Agent API?

A minimal Python sketch, using the SDK from [lesson 11](https://mentionx.ai/learn/perplexity/perplexity-api-agent-search-marketers):

```python
from perplexity import Perplexity
import csv, re

client = Perplexity()  # reads PERPLEXITY_API_KEY
brands = {"Ledgerline": "ledgerline.example", "Competitor A": "competitor-a.example"}

def cited_urls(response):
    # web_search adds a "search_results" output item; treat its URLs as the citations
    urls = []
    for item in response.output:
        if getattr(item, "type", "") == "search_results":
            urls += [res.url for res in item.results]
    return urls

rows = []
for prompt in open("prompts.txt").read().splitlines():
    for run in range(3):
        r = client.responses.create(input=prompt, effort="low", tools=[{"type": "web_search"}])
        text, urls = r.output_text, cited_urls(r)
        for brand, domain in brands.items():
            rows.append({
                "prompt": prompt, "run": run, "brand": brand,
                "mentioned": bool(re.search(re.escape(brand), text, re.I)),
                "cited": any(domain in u for u in urls),
            })

with open("results.csv", "w", newline="") as f:
    w = csv.DictWriter(f, fieldnames=rows[0].keys())
    w.writeheader()
    w.writerows(rows)
```

Notes on this sketch:

- **Citations come from the search_results output item.** Perplexity's docs say to treat each result's id and url as the source of truth for citations. Field access can vary by SDK version, so test on one prompt first.
- **Keep the model and effort settings fixed** across periods.
- **Estimate costs before running.** Prompts × runs × periods add up. Check the pricing page.
- **Store raw responses** so you can re-score them later.

For a no-code version, see [lesson 12](https://mentionx.ai/learn/perplexity/perplexity-n8n-automation). For a full tracker design, see [Build your own AI visibility tracker with APIs](https://mentionx.ai/learn/tools/build-own-ai-visibility-tracker-apis).

## How do you read the results?

1. **Share by intent group.** You may lead on problem prompts and trail on comparisons.
2. **The top cited domains** per group, which form your source map ([lesson 13](https://mentionx.ai/learn/perplexity/perplexity-how-sources-are-chosen)).
3. **Gaps:** prompts where competitors are cited and you aren't ([lesson 16](https://mentionx.ai/learn/perplexity/perplexity-citation-gap-earned-media)).
4. **Accuracy:** branded answers with wrong facts.
5. **Movement after changes,** such as a new page or a PR placement, with a lag of weeks.

## Worked example: Ledgerline

*Ledgerline is a fictional accounts-payable software company. Details are illustrative.*

Ledgerline built a 60-prompt set in four intent groups and ran each prompt three times a week through the Agent API, with monthly app spot checks.

| Intent group (illustrative) | Ledgerline mention share | Competitor B | Ledgerline citation share |
|---|---|---|---|
| Category | 25% | 60% | 15% |
| Problem | 45% | 30% | 40% |
| Comparison | 35% | 55% | 20% |
| Pricing | 50% | 50% | 45% |

### Walkthrough

1. Pulled 120 candidate questions from sales notes and Search Console, then cut them to 60.
2. Grouped them by intent and froze the set for the quarter.
3. Ran an n8n workflow calling the Agent API, saving answer text and citations to a sheet.
4. Scored mentions and citations, and averaged the runs per week.
5. Found comparison prompts were the weakest, with review sites and Competitor B pages dominating.
6. Prioritised a comparison page and review programme, then watched the comparison group for change.

## Common mistakes

- **Single runs.** Noise looks like a trend.
- **Mixing branded and category prompts.**
- **Changing prompts every month.**
- **Counting mentions as citations.** They're different signals.
- **Trusting one tool's numbers without knowing its method.**

## Try it yourself

Write 20 prompts in three intent groups, run each three times in Perplexity, and record mentions and citations for you and two competitors in a sheet. Calculate shares by group.

## Quick quiz

**1. What's the difference between mention share and citation share?**

*Answer:* Mentions count your brand named in the answer. Citations count your URLs in the sources.

**2. Why run each prompt several times?**

*Answer:* Answers vary between runs, so averages separate change from noise.

**3. Why keep branded prompts separate?**

*Answer:* They inflate share and measure accuracy, not competitive visibility.

## What should you check before the next lesson?

- Your prompt set is frozen and grouped by intent.
- The run count and settings are documented.
- A baseline period has been recorded.

Previous: [Track Perplexity referrals in GA4](https://mentionx.ai/learn/perplexity/perplexity-ga4-referral-tracking). Next: [A monthly Perplexity routine and governance](https://mentionx.ai/learn/perplexity/perplexity-monthly-routine-governance). Or return to the [Perplexity for Marketers track](https://mentionx.ai/learn/perplexity/perplexity-for-marketers). Related: [Competitor research](https://mentionx.ai/learn/perplexity/perplexity-competitor-research), [n8n AI visibility monitoring](https://mentionx.ai/learn/n8n/n8n-ai-visibility-monitoring), [Citation gap](https://mentionx.ai/glossary/citation-gap).

## Common questions

**What is share of citations in Perplexity?** It is the percentage of tracked prompts, or of all citations across those prompts, in which your domain is cited. Measure it alongside mention share, which counts brand names in the answer text.

**How many times should I run each prompt?** Answers vary between runs, so run each prompt several times per measurement period, for example three to five, and report the average with a range rather than a single result.

**Do API results match what users see in the Perplexity app?** Not exactly. The API and the app can use different models, settings and personalisation. Use the API for consistent trend tracking and spot-check important prompts in the app.

## Sources

- Perplexity Docs, [Agent API quickstart](https://docs.perplexity.ai/docs/agent-api/quickstart)
- Perplexity Docs, [Agent API web_search tool](https://docs.perplexity.ai/docs/agent-api/tools/web-search)
- Perplexity Docs, [Search API quickstart](https://docs.perplexity.ai/docs/search/quickstart)
- Perplexity Docs, [Pricing](https://docs.perplexity.ai/docs/getting-started/pricing)
- MentionX Learn, [Measuring citation share](https://mentionx.ai/learn/measurement/measuring-citation-share)
- All accessed 5 Nov 2026.

## Frequently asked questions

### What is share of citations in Perplexity?

It is the percentage of tracked prompts, or of all citations across those prompts, in which your domain is cited. Measure it alongside mention share, which counts brand names in the answer text.

### How many times should I run each prompt?

Answers vary between runs, so run each prompt several times per measurement period, for example three to five, and report the average with a range rather than a single result.

### Do API results match what users see in the Perplexity app?

Not exactly. The API and the app can use different models, settings and personalisation. Use the API for consistent trend tracking and spot-check important prompts in the app.

---

Written and reviewed by Dipuck Jones for MentionX Learn. The HTML version of this page: https://mentionx.ai/learn/perplexity/perplexity-share-of-citations-tracking