# AI visibility > How to measure whether ChatGPT, Gemini, Claude, Perplexity and Google's AI answers mention your business, and how to read every tab. ## What AI visibility measures More and more people ask an AI assistant instead of searching: "best dentist near the old town", "which accounting software for a small shop". **AI Visibility** asks those questions — your **prompts** — on several AI platforms and records: - whether the answer **mentions** your business, - at what **position** in a list you appear, - whether it **cites** (links to) your website, - the **tone** (sentiment) about you, - which **competitors** it names instead, and - which **sources** (web pages) it relies on. Platforms: **ChatGPT**, **Gemini**, **Claude**, **Perplexity**, **Google AI Overviews** and **Google AI Mode**. ## Setting up: the Settings tab Open **AI Visibility › Settings**: 1. **Brand names & aliases.** The names that count as "you": the business name, short forms, spellings, the domain. Add aliases (for example "Northside Bakery", "Northside", "northside-bakery.example") and **Not you** names that look similar but are someone else. Transliterations are offered as suggestions; you choose. After changing the list, **Recheck stored answers** applies it to answers you already have, for free. 2. **Platforms.** Tick the platforms to ask. The price per prompt depends on the platform and is shown next to it. 3. **Measure via.** How the answer is collected. ChatGPT is measured by **Live browser capture**; there is nothing to choose. Only if live capture is not available on this installation is ChatGPT asked through the API instead, at the API price, and the Settings row then shows "API" with its price. For Google AI Overviews you can choose between two methods. The options are explained below the list. 4. **AI budget.** Limits for one run and per day, so a large run cannot spend more than you expect. ### Measure via: the options - **API** — answers from the platform's model (Gemini, Claude, Perplexity). - **Live capture** — ChatGPT's answer as a customer would see it. 5 credits per prompt. It uses no model key, so the Key / model choice does not apply to it. - **Search results capture** — Google AI Overviews and AI Mode for each prompt. 5 credits per prompt for AI Overviews and 10 for AI Mode. An empty or failed answer is never charged. - **From rank tracking** — Google AI Overviews read from the results pages your [rank tracking](/docs/rank-tracking) already stored. Free. It measures your keywords rather than your prompts, and an AI Overview that loads late on the page can be missing its text — use Search results capture when you need every prompt measured. The methods can give different answers. The numbers never mix them: each platform counts only the method in use (the one you chose, or for ChatGPT live capture), and the screen says how many answers from the other method are hidden. ### Claude answers: Standard or Fast For Claude you choose the model that answers the prompts. **Standard** is the default. **Fast** is Claude Haiku 5.5: a smaller, cheaper model that answers the same prompts. An answer on Fast costs **13 credits**, and the Settings tab shows the price of each choice next to the other one, with what Fast saves. Each answer records which model gave it, so the chart can tell the two apart. Changing the choice does not run anything; the next run is priced when it starts. ## Prompts Open **AI Visibility › Prompts**: - Suggested prompts are generated for your business type, category and location. Review them; disable the ones that do not fit. - **Add** your own. Write them as a real customer would ask. - **Import CSV** with columns `prompt,intent,tags` (tags separated by `|`), and optionally `category` and `enabled`. A preview shows duplicates, invalid lines and lines over your plan limit before anything is saved. A CSV downloaded from Rankfinity can be imported back. - Each prompt has a **category** (for example brand, comparison, how-to) and an **intent**. A notice "N prompts need a category" offers to classify them. - Tags appear as chips and can be used as filters. Click a prompt to see its answers per platform, the history, the competitors named and the sources, and **Run this prompt now** for a single check with its own price. ## Running Click **Run now**. A window shows the bill: platforms × prompts, any "from rank tracking" line, and the total. **Ask every enabled prompt** runs all enabled prompts. While a run is in progress a bar stays at the top: "Run in progress · N of M answers · started HH:MM", with **Cancel run**. The reports are locked until the run finishes and open automatically when it does ("Report ready"). If a run fails or is cancelled, the bar says why and you pay only for answers received. ## The period picker All report tabs read the period chosen at the top: **Latest run**, **Last 2 runs**, **Last 7 days**, **Last 30 days**, **This month**, **Last month**, **All runs**, or a custom range in a two-month calendar where days with a finished run are marked. Changes are compared with the previous run that used the same platforms. ## Overview The five KPI cards come first: | KPI | Meaning | |---|---| | AI Visibility | Share of answers that mention you | | Share of Voice | Your mentions out of all brand mentions in the answers | | Ranking | Your average position when answers list several businesses | | Brand Strength | A combined score of mentions, position and citations | | Sentiment | Average tone of the answers about you | Under each number: "vs previous run" with the change, or "No comparable earlier run". Click a KPI card to show only the charts for that measure; click again to return to the full overview. Below the cards: **Brand check** (per platform and method: mentioned / not mentioned / not measured, with the quoted part of the answer) and **Google AI Overviews** from your rank tracking. ## The other tabs - **AI Visibility, Share of Voice, Ranking, Brand Strength, Sentiment** — each measure in detail: by platform, by category, by intent and over time, with the competitors. - **Source Tracker** — the domains and pages the AI answers cite, the type of page, and which brands appear with each. **Top cited pages** shows the page, its domain and the brands mentioned in answers citing it. - **Domain Comparisons** — your domain against competitor domains as sources. - **Gaps & Wins** — prompts where competitors are named and you are not ("Named instead of you"), and prompts you win. - **Query Fan-out** — the web searches the AI ran to build its answer, which tells you what content it looked for. - **Reports** — PDF reports and CSV downloads. Every row leads to its evidence: the prompt, the platform and the exact words in the answer. ## Reports and PDF In **Reports**, **Full PDF** builds a detailed report in the background (cover, contents, summary with changes, method, results per platform, category and intent, share of voice, sentiment with quotes, cited sources, gaps and wins, trend, an action plan and an appendix with every answer). You see the progress; when done, **Download PDF** is free. Building the same run again in the same language returns the same file without a new charge. **Download CSV** exports prompts, answers, mentions and sources. ## Common questions **Why does ChatGPT show different results here than in my own chat?** AI answers vary with wording, history and time. We ask without history from the project's market. Use **Live browser capture** to measure the real chat interface. **My brand is mentioned but not counted.** The name used in the answer is probably not in **Brand names & aliases**. Add it, then **Recheck stored answers**. **What does "Site cited" mean?** The answer links to your website but does not say your name. It is shown separately from mentions. **Why can't I run only Google AI Overviews "from rank tracking"?** That data comes from your rank checks; on its own it is already visible on the Overview card. Choose at least one other platform to make a run. **Why "not measured" instead of 0%?** The platform was not asked or the answer could not be read. Zero would mean you were asked and not mentioned.