AI Visibility Tracking

AI Visibility Tracking is Rankion's answer to the question: "Am I even mentioned in ChatGPT, Perplexity, Gemini, and Claude — and if so, how?" Instead of one-off spot checks, you create a tracking project, define brand, aliases, platforms, and prompts, and let Rankion query all four (plus Google AI Overviews) on a recurring schedule. You get an AVI score, six dimension scores, a list of cited sources with outreach status, and three automatic insight reports.

What it can do

• Tracking projects with wizard — Enter a domain or keyword, Rankion classifies, suggests keywords, competitors, prompts, and personas, and you activate with one click.
• Brand aliases — Coca-Cola, Coke, coca cola gmbh as aliases so mentions are merged.
• Platform selection — ChatGPT, Perplexity, Gemini, Claude, each with/without web-search mode, plus Google AI Overviews (AIO).
• Tracking runs — manual or on a frequency schedule. Each run queries every platform for every prompt.
• Score history — score per day, platform breakdown, compare mode across timeframes.
• Cited sources — every URL your platforms cited as a source, with outreach status (new, request_sent, linked, rejected, ignored).
• Reality Check / AVI score — six-dimensional rating (Visibility, Sentiment, Authority, Specificity, Freshness, Differentiation), tier red/yellow/green, optional PDF report.
• Insights — Low-Hanging Fruit (prompts that nearly cite you), Platform Gap (platform A cites you, B doesn't — diagnose per platform), Potential Hero (prompts with brand fit but low visibility, requires 20+ queries).
• Source whitelist — the up to 30 third-party domains your platforms cite most often as a source, one row per domain and never your own domain (details below).
• Data as CSV — a project's measurement data in ten topics as CSV or XLSX, with selectable delimiter, encoding, and line ending (details below).

When to use

• You want to prove measurably that your GEO efforts are working.
• Your brand is described "wrong" in ChatGPT (outdated, wrong market, wrong comparison).
• You want to find out which third-party sources your platforms cite — and run outreach there.
• You need a PDF to give to leadership/clients ("here's how visible you are in AI").

Workflow

1. Create project — POST /tracking-projects/analyze with {mode:"domain", domain:"…"}. Returns 202 + project_id.
2. Poll status — GET /tracking-projects/{id}/analysis-status until suggestions are ready.
3. Activate — POST /tracking-projects/{id}/activate with your final keywords, competitors, prompts, platforms, frequency.
4. Run — POST /tracking-projects/{id}/run triggers queries against all platforms.
5. Insights & outreach — GET /tracking-projects/{id}/insights/low-hanging-fruit, contact cited sources, pull the AVI report.

Full guide with UI screenshots: Track AI Visibility.

Choosing the date range

On the overview and on the project page, the button with the calendar opens the date range menu. At the very top, under Custom range, are the fields From and To: enter two dates or pick them in the calendar, then Apply range. You can go back at most 365 days. Below are the presets 7, 30 and 90 days, which apply with one click. On the project page the button is the first one in the row above the metrics.

Reading the source whitelist

The Deep Insights page of a tracking project (/ai-visibility/{id}/insights, tab Source-Whitelist) lists the third-party domains your platforms name most often as a source — becoming visible there earns you the most LLM citations.

• One row per domain. All cited URLs of a domain are merged; no domain appears twice. The individual URLs with their outreach status remain under Cited Sources.
• Citations = total since the project started — every citation of the domain across all its URLs and all tracking runs. Not to be confused with "Cited" in the project overview: that is a rate for the selected timeframe and only counts how often your own site was cited. The two numbers are therefore not directly comparable.
• Rank — domains are sorted by that total and numbered consecutively until together they reach at least 80 % of all citations. Only ranked domains appear in the list, 30 at most. For active projects the ranking is recalculated every night; a paused project keeps the ranking from its last nightly run.
• Value score (0–100) — per domain, the citation-weighted average of its scored URLs; empty as long as none of the domain's URLs has been scored.
• Your own domain never appears. It gets no rank and does not count toward the 80 % base, because the list shows the third-party sources where you want to become visible. How often you yourself are cited is shown by "Cited" in the overview.

The same list is part of the Deep Insights PDF report (button "Generate insights report", then "Download PDF"; the weekly insights report email carries it as an attachment). Via API: GET /v1/data-mining/insights/{project}/section/source_whitelist — per domain with citation_count (total), citation_value_score, url_count (number of merged URLs) and the most-cited url.

Data as CSV

On the project page, directly below the Analysis report as PDF tile, you'll find the Data as CSV tile. It is collapsed when the page loads. Expanded, it shows the format row and ten topics, each with one button for CSV and one for XLSX.

• The ten topics: Visibility per platform (time series), Prompt results, Cited Sources, Topic cluster, Potential gaps, Competitor mentions, Competitor sources, Measurement coverage, Keyword gaps, Actions per platform.
• Date range: The range selected at the top of the page applies. Keyword gaps and Measurement coverage are snapshots and have no date range.
• Presets: Windows (semicolon, UTF-8 with BOM, CRLF), Linux (comma, UTF-8 without BOM, LF), and Apple (semicolon, UTF-8 with BOM, LF).
• Individual overrides: After picking a preset, you can change the delimiter (semicolon, comma, tab), encoding, and line ending one by one.
• Remembered per user: Your choice is saved to your account and applies again on your next export.
• Numbers and XLSX: Numbers always use a decimal comma. The format row only applies to CSV; any program reads XLSX files without it.
• Large exports run in the background and then appear in the export list.
• Other perspectives: If you are viewing the page from a competitor's perspective, the export is locked, because the file always contains your own project's data.

Via API: GET /v1/tracking-projects/{id}/exports returns the topics with their columns and the allowed format values. GET /v1/tracking-projects/{id}/exports/{thema} returns one topic, with these parameters: format (csv, xlsx), preset (windows, linux, apple), delimiter (semicolon, comma, tab), encoding (utf8-bom, utf8), eol (crlf, lf), from, to. from and to use the form YYYY-MM-DD; without them, the last 30 days apply. An unknown format value falls back to the preset. A large export answers with 202 and export_id; the finished file is then listed under GET /v1/exports.

Values for {thema}: kpi-verlauf, prompt-ergebnisse, zitierte-quellen, themencluster, potenzialluecken, wettbewerber-nennungen, wettbewerber-quellen, mess-abdeckung, luecken, massnahmen.

API

Credits & Limits

• Platform-Gap diagnose: 1 credit per platform diagnose.
• Content brief from prompt: 3 credits.
• Tracking-Project report: 10 credits, rate limit 1 request / 30 min / project.
• Reality Check / AVI / Dimensions: no credit cost (run cost comes from the platform calls).
• Potential Hero: requires 20+ queries, otherwise {eligible:false}.
• Prompts per run: Every run checks all active prompts of your project, on every plan. Each answer from an AI system costs 1 credit from your quota. Your plan only limits how many AI systems a project queries. If your balance runs out during a run, the run stops at that point.
• When the next run happens: The schedule follows the frequency you chose. A run you start yourself does not move it. If a finished run already happened on the scheduled day, no second run starts and the schedule moves on.
• Fact check »Check now«: rate limit 1 request / 60 min / project. While a run is in progress the button is greyed out and shows a spinner — the page refreshes itself, you do not need to reload. A further click within the same hour is declined and states the reason.
• On your first visit the fact check guides you through three steps: facts imported · facts confirmed · now check. The first two are done as soon as your website has been read and approved; the block disappears once you have rated your first statement.
• The "AI factual accuracy" rate only appears from ten pages read onwards. Below that you see a dash and the reason underneath. It used to show a greyed-out percentage — that looked like a measurement but was not one.

Reports and diagnose calls run async — you receive 202 + a status endpoint to poll.

Related modules

• AI Content Editor — turn low-visibility prompts into content briefs straight in the editor.
• Backlinks Module — the cited-sources list maps 1:1 to outreach targets.
• Agentic Chat — query AVI trends in natural language.
• Guide: Track AI Visibility
• Guide: Source profile of your competitors: which sources carry your competitors' mentions.
• Guide: Period comparison: two periods side by side, with export.
• Guide: Record measures and measure the change: measures in the trend curve and the change afterwards.
• Guide: Manage personas: create, rename and delete personas.