Track AI visibility (ChatGPT, Perplexity, Gemini)

Classic SEO measures whether you're in Google's top 10. AI visibility measures whether you appear in answers from ChatGPT, Perplexity, Gemini and Claude at all — and if so, how. This guide walks you through the full tracking workflow.

What is AI visibility?

LLMs generate answers from their own training, from web searches or from retrieval-augmented sources. When a user asks "What is the best vegan protein source?", the model lists vendors, brands and URLs. Whoever gets named there wins traffic, trust and market share. Whoever doesn't is invisible — no matter how good the Google position is.

Rankion automatically and periodically asks your brand-relevant questions to several models and logs: are you mentioned? In what context? Which sources does the model cite?

Create a tracking project

In the menu: AI Visibility → New tracking. The wizard asks you:

1. Name — e.g. "Vegan protein DACH"
2. Mode — domain tracking (one brand/domain) or keyword tracking (topic, all relevant players)
3. Domain or keywords — what is being tracked

After the analysis (step 2), the wizard starts the sitemap scan on its own when you continue; within a few minutes the suggestions from your own pages appear under Keyword suggestions from the sitemap. The analysis costs 5 credits once, and the sitemap scan another 5 credits once.

Via the API:

curl -X POST https://rankion.ai/api/v1/tracking-projects \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -d '{"name": "Vegan protein DACH", "mode": "domain", "domain": "mydomain.com"}'

If your team already has an SEO project, the still-empty AI-Visibility page offers it to you directly. Clicking it starts the wizard with the name and domain already filled in, instead of typing everything from scratch. The "New tracking" path stays available alongside it.

Define brand aliases

Brands have spelling variants. "MyDomain", "My Domain Inc.", "MD24" — all the same company. Maintain aliases on the project so mentions are attributed correctly. Without aliases you miss hits.

Pick platforms

Pick which models are queried. Typical selection:

{
  "llm_platforms": ["chatgpt", "perplexity", "gemini", "claude"]
}

Each platform costs credits per run. You can also pick just one if your market is clear (e.g. only Perplexity for tech B2B).

Set the frequency

Five frequencies are available:

Before you save, the picker already shows what a frequency costs — for example "Weekly — up to approx. 200 credits/month". This per-frequency preview appears in three places: the setup wizard, an AI-visibility project's settings (project view), and the project's SEO settings. The Prompt monitoring page has no per-frequency picker; it shows the total for the current prompt selection at the set frequency, and a help icon explains the math. In the project view's settings and on the monitoring page, the calculation always uses the currently set value, even unsaved.

1 credit is counted per response received: a response is generated per active prompt × platform × search mode (with/without web search); Google AI Mode runs with search only, Grok without. With Reality Check on, both search modes count plus 15 credits per run; with Google AI Overviews on, add 1 credit per active keyword. The monthly total multiplies credits per run by the highest number of runs possible in a calendar month (table above) — hence "up to": many months have fewer runs, and only responses actually received are charged. Your plan limits how many prompts and platforms are included per run; the preview already accounts for that limit.

Example: 20 prompts × 2 platforms × 1 search mode = 40 credits per run — weekly up to approx. 200 credits/month, daily up to approx. 1,240.

Via the API: GET /api/v1/tracking-projects/{id}/prompts/cost-preview returns monthly_credits, credits_per_run and credit_breakdown; prompt_ids[]=… restricts to a subset.

For most brands, weekly is enough; fast-moving markets (news, crypto) benefit from daily.

Start a run + wait

The first run is automatically scheduled when you create the project. To trigger one manually:

curl -X POST https://rankion.ai/api/v1/tracking-projects/{id}/run \
  -H "Authorization: Bearer YOUR_TOKEN"

Response: HTTP 202 with a job ID. Duration depends on number of questions × platforms — usually 2–10 minutes. The UI polls automatically and shows live progress.

Interpret the score

After the run you see:

• Visibility score (0–100) — how often you were mentioned, in what position of the answer, with what sentiment.
• Reality check / AVI — compares the model's knowledge with the real top-10 SERPs. Big gap = LLM remembers you, Google doesn't (or vice versa).
• Platform breakdown — where you're strong, where you're weak.
• Trends — historical development per run.

When the question set has changed

The trend chart only compares like with like as long as the same questions ran on every measurement day. If questions are added or reworded, the number moves without your visibility having moved. New questions are often phrased more informationally and name a brand less often; the share then drops purely because the question set got bigger.

Two things show you this:

• A dashed stretch in the trend chart means the question set changed between those two measurement days. The values to the left and right of it rest on different questions. The number in the tooltip gives the questions per day.
• The notice above the metrics, "The question set changed during this period", tells you how many questions were added or reworded and offers "Switch to the same basis". After that, only questions count that have been active unchanged since your first measurement.

For a comparison of two points in time that you want to present, use the same basis. For the question of how visible you are overall today, use all questions.

Grey bands: partial measurement

A grey-shaded period in the trend chart means: on that measuring day a run checked only part of your questions — fewer than 80 % of the questions that were active at the time. Until 10.10.2026 the plan limited the number of questions per run; on such days, for example, only every fourth question ran.

The value of such a day is therefore based on a smaller and different selection of questions and is only partly comparable with a complete measuring day. Below the chart, each grey period shows how many questions were measured, for example "29.09.2026: 50 of 203 prompts measured". The current day is only evaluated once it is over.

Via the API you get the same information with ?partial_measurements=1 on the visibility-trend endpoint.

Reading the trend, the legend and the pause notice

• The vertical axis follows your values. In the »Visibility trend per AI engine« and in the history under »How is the gap to your competitors developing?«, the percent axis no longer runs to a fixed 100 %. It has four equal steps from 0 and ends just above your highest value (with a little headroom, never above 100 %). If your visibility is 25 %, the axis runs to 30 % instead of 100 %, and the lines use the full height. In the competitor history this applies to the measures »Mentions« and »Citations«; the »Average rank« stays inverted, lower is better. Tooltips and the table still show the same numbers.
• Mentions are counted by the same rule for every brand. In the history under »How is the gap to your competitors developing?« and in the card »Amazon Rufus: mention and recommendation per brand«, an answer counts as a mention if the brand appears in the answer text or one of its pages is among the sources. This applies to your brand and to every active competitor you have added, and a brand counts at most once per answer. The API returns the same count under GET /v1/tracking-projects/{id}/competitor-trend. One small difference remains: competitors are also found by their domain name without the ending, your brand is not. In the one project where we measured this, it made up 0 to 4 answers per day. On Amazon Rufus, a match in a product title of the recommendation list also counts as a mention, for your brand and for your competitors alike. For competitors, short derived search terms under four characters are not searched in titles. Older measurement days are only re-evaluated later; until then, competitors' mentions on those days can still come out too low.
• The legend gives a reason instead of a dash. If a platform has no value in the selected period, the legend says »not in plan« or »no measurement«. Not in plan: the project does not currently query this platform, because it is not selected in the project settings or exceeds the platform limit of your plan. No measurement: the platform is scheduled, but no answer from it arrived in the period. Hover over the text to see the explanation.
• »Tracking paused: insufficient credit balance«. If your team's balance does not even cover one complete run, Rankion skips the scheduled measurements. A notice band on the project page shows this; the project overview shows a band with the number of affected projects and »Paused: credits missing« on each of them. »Top up credits« takes you to the top-up. Your existing data stays; measurements resume as soon as the balance covers a run.

Choosing platforms per prompt

In the window »Add keywords & prompts«, you choose under »Platforms« where the new prompts go. You can pick the platforms that are enabled for your team; next to each one you see what it costs in credits per run. Amazon Rufus is not preselected. In the next step you can switch the platforms of each individual prompt; at least one always stays selected.

General questions only run on the AI assistants. For Amazon Rufus the wizard creates its own product searches in the style of an Amazon search, such as »hand cream dry skin«, because Rufus answers about products.

Analyze citations

The exciting part is the cited sources. Models often link to specific URLs in their answers. Rankion collects them, aggregates per domain and shows you:

• Which URLs are cited most often?
• Which competitors are at the top?
• Which of your own pieces of content aren't picked up at all?

That's your content roadmap: topics where you're missing are your next articles — see Your first article with the AI Content Editor.

Read on

In-depth knowledge of setup, limits, the data model and every endpoint lives in AI Visibility Tracking and the API reference. Questions on credits & quotas are answered in Credits & Rate Limits.