- Home
- Skills
- Research & Analysis
- AI Citation Tracker
Works with the AI Agents you already use
AI Citation Tracker
**Know whether AI assistants actually recommend you, and whether that is changing.** Most "AI visibility" checks are one lucky screenshot.
Free
AI Citation Tracker
Example session with this skill installed
I need to track how often ChatGPT and Perplexity cite my brand, 'Solaris Energy', compared to 'SunRun' and 'Tesla' for home solar questions in the US. Produce a baseline report from my log.csv.
- Read your context and instructions
- Compiled the ai citation tracker
- Generated the UI component
I have analyzed the provided run logs for Solaris Energy. The baseline reveals a significant gap between mention rate and citation rate, particularly in comparison prompts.
AI Visibility Baseline Report: Solaris Energy
Period: October 2023 | Panel: 30 Prompts | Runs: 180 total (3 per prompt/engine)
Core Metrics
| Metric | ChatGPT (GPT-4o) | Perplexity (Pro/Online) | Aggregate |
|---|---|---|---|
| Mention Rate | 42% (38/90) | 58% (52/90) | 50% [42.7% - 57.3%] |
| Citation Rate | 12% (11/90) | 44% (40/90) | 28% [22.0% - 35.3%] |
| Share of Voice | 15% | 22% | 18.5% |
| Accuracy Score | 94% | 88% | 91% |
Source Map (Top 5 Cited Domains)
- energysage.com (Informational/Aggregator) - 34% of citations
- solarreviews.com (Review Site) - 22% of citations
- sunrun.com (Competitor) - 15% of citations
- tesla.com (Competitor) - 12% of citations
- solaris-energy.net (Owned) - 10% of citations
Diagnostic Summary
Solaris Energy is frequently mentioned by name but rarely cited as a primary source. Perplexity shows a higher citation rate due to its search-first architecture, but it often favors aggregator sites like EnergySage for technical specs. Two "Brand" intent prompts in Perplexity resulted in hallucinations regarding your warranty period (reported as 10 years instead of 25).
Next steps
- Review the hallucinated warranty responses in
logs/audit_errors.csvfor specific prompt triggers. - Focus SEO efforts on EnergySage and SolarReviews, as these are the primary "authority" bridges for both engines.
- Schedule the next run for November 15th to establish the first trend line.
audit_errors.csv
TSX · React component
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Know whether AI assistants actually recommend you, and whether that is changing.
Most "AI visibility" checks are one lucky screenshot. AI answers change between runs, so a single query proves nothing. This skill gives your agent a disciplined, repeatable method and honest statistics.
What it does
- Builds a frozen prompt panel of 20 to 40 realistic prompts across category, comparison, how-to, problem and brand intents, with holdout prompts to catch overfitting. A script generates starter prompts from your brand, category and competitors.
- Defines exactly how to run and log each engine, with coding rules so two people log the same answer the same way: what counts as a mention, a citation, a position, an accuracy error.
- Computes the metrics: mention rate, citation rate, share of voice, average position, accuracy and sentiment, per engine and per intent, with 95% confidence intervals.
- Compares rounds honestly. Uses a significance test, refuses to call a result with too few runs, and flags prompts whose result flips between runs.
- Builds the source map: which domains engines cite most in your category and what those pages have that yours lack. This is usually the most actionable output.
- Logs wrong facts about your brand with the exact wording, engine and date.
- Writes the monthly report from a ready template.
How it works
- Collects brand, domains, category and competitors.
- Builds and freezes the panel.
- Fixes the run method and measures noise.
- Captures the baseline, logging every run.
- Runs the analyzer on your CSV.
- Diagnoses the failing stage (eligibility, retrieval, selection, recommendation).
- Produces the report and schedules the next round.
What makes it different
- Statistics included: Wilson intervals and a two-proportion test, in a standard-library Python script. Verified on a synthetic dataset where a real improvement and a noise-level change were correctly told apart.
- Evidence you can audit: every figure traces to log rows.
- Integrity rules built in: no invented results, no fake accounts, no scraping that violates an engine's terms. If an engine cannot be queried, it is marked "not run".
- Pairs with the GEO Citation Playbook (optimization), but works on its own.
Where it stops
- It measures visibility. It does not guarantee rankings, citations, traffic or revenue.
- It does not query AI engines itself; your agent or you collect the runs.
- Results are valid for the dates, modes and locations logged.
- Referral traffic from AI is poorly tracked by analytics; the skill says so rather than inventing attribution.
See it in action
Example output on synthetic data, illustrating the report format.
Request: "Set up AI citation tracking for Acme (acme.com), invoicing software for freelancers. Competitors: FreshBooks, Wave. Engines: ChatGPT, Perplexity, Google AI."
Result (excerpt):
Round 1 -> Round 2 (90 runs each; 30 prompts x 3 engines, 1 run sampled per prompt-engine-day)
Mention rate: 15 of 90 (17%, 95% CI 10% to 26%) -> 41 of 90 (46%, CI 36% to 56%)
Difference unlikely to be chance alone (p<0.001); runs share prompts, so treat the size as approximate.
Citation rate: 8 of 90 (9%) -> 14 of 90 (16%): inconclusive (p=0.172)
Most cited domains: g2.com 90, reddit.com 76, acme.com 14
Unstable prompt/engine pairs: P01/ChatGPT 2 of 3, P04/Perplexity 1 of 3 ...
Diagnosis: the brand is now named more often, but its own pages are rarely cited. Recommendations come from review roundups on g2.com and Reddit threads. Next: a legitimate, disclosed presence in those sources, plus a comparison page answering the "Acme vs FreshBooks" sub-questions.
Limitations
- Does not guarantee any outcome.
- Needs engine access (browser, API or manual runs) for data collection.
- Heuristic scripts on small panels give wide intervals; this is stated in every report.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
Requires manual or external data collection for engine runs. Does not provide optimization instructions, only measurement and diagnostics. Python scripts use standard library only.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 9 days ago
- Free with an Agensi account