> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rankraft.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# How RankRaft AI measures AI answers

> How answers are collected, how often, and why the numbers are trends rather than single replies.

## Answers are collected the way your buyers see them

RankRaft AI collects answers through a combination of **official APIs** and **browser automation**, choosing the method that most faithfully reproduces what a real person sees on each engine. Answers are collected as a logged-out user would get them: the engine decides for itself which model to use and whether to search the web, as it would for a real buyer.

We store each answer **exactly as the engine returned it**, together with every source it cited, so every figure in the dashboard can be traced back to the text it came from. Open any prompt in [Prompts & Answers](/analytics/prompts/prompts-and-answers) to read the stored answers.

## Engines

| Engine | What it is | Status |
| - | - | - |
| AI Overviews | The answer box on Google results | Live |
| AI Mode | Google's conversational search tab | Live |
| ChatGPT | The most used assistant by a wide margin | Live |
| Gemini | Google's assistant, inside Workspace and Android | Live |
| Claude | Favoured by technical and research buyers | Live |
| Perplexity | Answer engine that always cites its sources | Coming soon |
| Copilot | Microsoft's assistant, in Windows and Edge | Coming soon |

Which engines you get depends on your [plan](/configuration/plans-and-billing).

## Daily, because one answer tells you very little

AI gives different answers to the same question. We ask every active prompt once a day on every engine in your plan and report the trend, because a single answer, however many times you repeat it in one sitting, tells you less than a fortnight of them.

Each site has a fixed daily scan slot, set when the site is created, so your scans run at a consistent time.

## Localised to your market

If you set a main market, every question is asked as a buyer in that market would ask it. In the dashboard a prompt shows a badge like *Asked as “best mixed pickle brand (Pakistan)”*. The difference is large. Without a market, a question about a local category often returns US retailers and publications. With one, it returns the local brands and sites your buyers actually see.

**Global** adds no location.

> [!NOTE]
>
> AI Overviews and AI Mode are always measured as a searcher in a specific place sees them, so they need a country market. If your market is **Global** or empty, those two engines are skipped for your site, and they aren't counted against your results.

## When an engine shows no answer

About 30% of Google result pages show no AI Overview at all. That is a finding about the *question*, not about your brand. Those runs appear in the answer list as **No AI answer shown** and are left out of every rate, rather than being counted as "not mentioned".

## Mentions and citations are found by exact matching

Whether an answer names you is decided by case-insensitive whole-word matching of your brand name and its aliases (**Also known as** in your profile). No AI model is involved in that decision, so the same answer always gives the same result. Add common misspellings or short forms of your name as aliases.

## Partial scans

If some answers can't be collected on a given day, the rest of the scan is kept and the day is marked **P** (*Some answers were unavailable*) on charts. Days where you changed your setup are marked **C** (*config change*).

## Margins and honesty about small samples

Every Visibility Score carries a 95% margin, and low-sample figures are greyed out with **Low sample · N answers**. See [Visibility Score](/analytics/visibility/visibility-score).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.