> ## Documentation Index > Fetch the complete documentation index at: https://docs.mithunai.com/llms.txt > Use this file to discover all available pages before exploring further. # See how AI search answers your buyers > Ask AI engines the questions your buyers ask, and see which companies they name, which pages they cite, and where a competitor is named instead of you. **AI search visibility** asks AI engines the questions your buyers ask, and shows you what they answer: which companies they name, in what order, and which web pages they point to. It finds the questions where an engine recommends a competitor and never mentions you. > [!WARNING] > > AI search visibility is in **beta**. It appears only when your MITHUNAI operator has switched it > on and added the AI engines' keys. ## Set it up Open **AI search visibility** in the console. You need to be an owner or admin. Once a check has run, the page opens on the results, and everything below is under **Set up** at the foot of the page. 1. **Your brand.** Add your company or product name. Add any other names people use for it, such as an old name or a short form, and your websites, so pages from your site are credited to you. 2. **Competitors.** Add up to 19, the same way. 3. **Questions.** Add the questions your buyers would type into an AI assistant, for example *"Which help desk tool is best for a 20-person support team?"*. Write them the way a buyer would, not the way your marketing would. Up to 50. 4. **Engines.** Turn on the engines you want to check: OpenAI, Anthropic and Perplexity. An engine your operator has not set up is shown but cannot be turned on. 5. **How often.** Choose **Off**, **Weekly** or **Daily**, and how many times each question is asked per engine (1 to 3). ## Run a check Select **Run a check now**. Each question is asked of each engine you turned on, with the engine's own web search switched on. The page shows how many AI questions a check will ask before you start it, and how many you have left this month. A check asks *questions x engines x times asked* AI questions. For example, 10 questions on 3 engines asked once each is 30. When a check would go over your monthly limit, it does not start, and nothing is used. ## Read the results | What you see | What it means | | - | - | | **Mentioned in** | The share of answers that name the company. The range beside it shows how sure the number is: fewer answers, wider range | | **Share of mentions** | Of every time a tracked company was named, the share that was this one | | **Usually named** | Its average position: 1 means it is named first | | **Cited pages** | The share of pages the engines pointed to that are on this company's websites | | **Over time** | How the share of mentions changed from one day to the next, for each engine | **Gaps** lists the questions where an engine named a competitor and never you. Each gap quotes the engine's answer and lists the pages it cited, so you can see what the engine read. Those pages are often the best place to start: they show what content the engine trusts on that question. ## Propose a fix (beta) **Propose a fix** on a gap drafts a section that answers that question, written only from your own knowledge base. It is drafted for your own question, never from the engine's answer or the pages it cited. When your knowledge base does not say enough, it says so and you can write the section yourself. Edit the title and the section, then choose where it goes: * **A pull request in your docs repository**, once the [Marketer's](https://docs.mithunai.com/team/marketer.md) Docs repository is connected to GitHub. Name a `.md` or `.mdx` file; an existing file gains the section at its end. * **A draft on your WordPress site**, once the Marketer's Website and blog is connected to WordPress. * **Copy or download it**, when neither is connected. Either of the first two becomes an approval card. Nothing is sent until a person approves it: then a new branch and a pull request are opened, never a change to your default branch and never a merge, or a WordPress draft is saved, never published. A draft counts as one answer against your [monthly allowance](https://docs.mithunai.com/analyze/usage-and-limits.md). ## What the numbers are, and are not * The answers come from each engine's official API. **They are not exactly what a person sees** in the engine's own app, which can add memory, location and personal settings. Treat the numbers as a consistent measure you can track over time, not as a copy of one person's screen. * AI answers change from one call to the next. Asking each question more than once gives steadier numbers. * A company counts as mentioned when its name, or one of its other names, appears as a whole word. A description without the name ("the big Redmond company") is not counted. * Google Gemini is not checked yet. ## Safety Answers and cited pages come from the open web, and anyone can publish a page an engine repeats. MITHUNAI shows them as plain text, never opens the links for you, and never acts on anything an answer says. Links are shown only if they are ordinary web addresses. ## Related * [Coverage gaps](https://docs.mithunai.com/analyze/coverage-gaps.md): the questions your own assistant could not answer * [Usage and limits](https://docs.mithunai.com/analyze/usage-and-limits.md) ## For operators Each engine is switched on by a key in the server environment, and is unavailable while its key is empty: nothing is sent to it. | Engine | Key | Notes | | - | - | - | | OpenAI | `ARUKZ_GEO_OPENAI_API_KEY` | OpenAI's own API, for this feature only. It never uses `ARUKZ_OPENAI_API_KEY`, which serves answers and may point at another OpenAI-compatible service. `ARUKZ_GEO_OPENAI_BASE_URL` overrides the host (`https://` only) | | Anthropic | `ARUKZ_ANTHROPIC_API_KEY` | The same key the answer path uses | | Perplexity | `ARUKZ_PERPLEXITY_API_KEY` | | This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.