How RankX AI measures visibility
RankX AI asks each tracked prompt on ChatGPT and any assistant you switch on, and records whether your brand was named, where, and which sources were cited.
RankX AI measures AI visibility by asking. RankX AI sends each tracked prompt to each AI assistant you have switched on (ChatGPT by default, with Gemini, Claude and Perplexity one switch away), reads the answer, and records whether your brand was named, where it appeared and which sources the answer cited. One prompt asked on one assistant on one run is one check, and every figure in the product is built from checks with a known verdict.
This page is the method behind the numbers. Knowing it is what lets you tell a real movement from noise. The public summary is on the methodology page.
What is one check?
One check is one prompt, asked on one assistant, on one run. Nothing is averaged before that point and nothing is inferred after it.
Three consequences follow:
- A Website with 25 prompts on 4 assistants performs 100 checks per run, not 25. Your credit spend follows the same multiplication. See credits and metering.
- Per-assistant figures are the real figures. An overall rate summarises four measurements taken on four systems that often disagree with each other.
- A single check is not a measurement. AI assistants are non-deterministic, so asking the same question twice can return a different list of brands. The reliable unit is a rate across a panel of prompts over several runs.
What does RankX AI ask?
A prompt is a natural-language question, written the way a buyer would ask it, attached to a topic cluster and tagged with a buyer stage: Researching, Comparing or Ready to buy.
RankX AI suggests a first set of prompts during setup, from your brand profile, competitors and topic clusters, and you edit it. After that the panel is yours to manage. Every active prompt is checked on its assistants on every run, so the prompt list is both your measurement panel and your recurring cost.
A prompt has three states:
| State | What it does |
|---|---|
| Active | Runs in the scheduled check and is charged when checked |
| Inactive | Paused: costs nothing, keeps its history, and can be switched back on |
| Suggested | A proposal you have not accepted yet |
See prompts and topic clusters.
Which AI assistants does RankX AI ask?
Every Website is checked on ChatGPT, with Gemini, Claude and Perplexity one switch away, on every plan, with no add-ons. You choose which of them each Website is asked on, and each prompt can narrow that list further. A store Website starts on ChatGPT in the same way.
- ChatGPT is the most widely used AI assistant and covers everyday consumer and business questions.
- Gemini is Google's AI, which matters most to brands whose buyers live in Google.
- Claude is Anthropic's AI, often used for research and document-heavy work.
- Perplexity is an AI-native search engine that cites its sources, so it shows clearly which sources appear beside you.
Google AI Overviews is measured separately, because it is a different surface collected a different way: by the rank tracker, on your tracked keywords. See citations, two kinds and tracked AI assistants.
When does RankX AI ask?
| Run | When | What it covers |
|---|---|---|
| Setup baseline | Straight after you complete setup | Every accepted prompt, on every assistant switched on |
| Scheduled run | Every night at 22:00 UTC, for the prompts that are due | Every active prompt whose own cadence has elapsed, on the assistants it targets |
| Manual run | Whenever you ask, as often as you like | One prompt, or all of them, on the assistants they target |
The nightly run is when RankX AI looks, not how often your prompts run. Each prompt runs on the cadence you chose for it, or the Website's if you chose none, raised to a floor. Two floors apply and the slower one wins:
- Your plan's floor, which is one day for AI visibility on every plan. See plans and limits.
- A seven-day floor while you are on trial, on every plan and every check family. Seven days buys a first measurement and a re-check, which is what shows a real movement. See the free trial.
A scheduled run survives interruption. It works through prompts in bounded batches, and if it runs out of time it records what it completed, marks the run partial and dispatches the remainder.
If the credit balance runs out mid-run, the run stops and says so, and it never records the prompts it did not reach as absences. See check cadence.
What does RankX AI record from an answer?
For each check, RankX AI stores the answer and records:
- Whether your brand was named: mentioned, not mentioned or unknown. Unknown means the check could not produce a verdict, and it is never counted as "not mentioned".
- Where your brand appeared, when the answer contains a ranked list.
- Which competitors were named in the same answer.
- Which sources the answer cited, which feed AI Answer Citations.
- How the answer spoke about your brand: positive, negative, mixed or undetermined.
- The answer text itself, which the AI Chat Feed shows you.
How does RankX AI decide that a brand was mentioned?
An AI model reads the answer and proposes which brands were named, and RankX AI's own code checks every proposal against the answer text before anything is stored. The model proposes, the product verifies, and the verified answer is the one recorded.
Three rules make that check strict:
- A brand whose name does not appear in the answer cannot be recorded as mentioned.
- Every tracked competitor whose name does appear is captured, whatever the model proposed.
- Words echoed from the question itself do not count. A term the prompt already contained, or the name of your own city, is not evidence of a brand mention.
What will RankX AI never infer?
These rules are enforced in the product:
- An absence of measurement is never reported as zero. A rate divides by checks with a known verdict. If none has one, the rate is unknown, and unknown is what you see. Null is not zero covers this in full.
- "We did not check" is never merged with "we checked and found nothing." In Google results, "Not in the top 20" and "Not checked" are separate states, and an AI Overview detected but not captured is unknown, not "you were not mentioned".
- A score is never shown without its coverage. AI Readiness shows its score beside the share of checks that were assessed, and below a coverage floor it shows no score rather than a misleading one.
- A failed check is refunded and reported as failed. RankX AI says which assistant failed and returns the credits; it never counts the failure as your brand being absent.
How do I read a movement?
- Compare rates, not counts. A count moves when your prompt list moves.
- Compare per assistant. A drop on one assistant and a rise on another are two facts, and an average hides both.
- Carry the denominator. "Named in 12 of 40 checks" survives being quoted; "30%" does not.
- Compare like with like. Cadence and assistants are set per prompt, so the prompts behind a figure can change between two periods. Compare periods with the same prompt set before reading a trend.
- Give it several runs. Movement on one prompt in one run is noise, and treating it as a signal is the most common mistake with this kind of data.
- Read the AI Chat Feed when a number surprises you. The stored answers usually explain the number in one read, because they show which brands were named instead.
Where to go next
- Metrics defined, for each figure the product shows.
- Null is not zero, for the rule behind every rate.
- Citations, two kinds, for AI Answer Citations against AI Overview Citations.
- The AI visibility tracker, for what RankX AI tracks across the assistants you switch on.
Questions
Why does my visibility change when nothing on my site changed?
AI assistants are non-deterministic, so the same prompt can name different brands from one run to the next. Read a rate across a panel of prompts over several runs, per assistant, before treating a movement as real.
Does RankX AI track Google AI Overviews in the same run?
No. Google AI Overviews are captured by the rank tracker on your tracked keywords, alongside the position check, and reported in the Google Results section. See AI Overviews.
Can I choose which assistants each prompt is asked on?
Yes. Each Website has a list of assistants switched on, and each prompt can narrow that list. A prompt can never be asked on an assistant the Website has switched off.
Where can I read the answers behind a number?
In the AI Chat Feed, which shows the stored answer for every check, the brands it named and the sources it cited.
Last updated