# How to fix a hidden AI visibility gap in 10 minutes > Run a free report to see which buyer questions leave you out. Pick the one closest to a sale. Publish a clear, sourced answer to that exact question on your own domain, in a form AI crawlers can read. Share the reporting behind it, then test the same question again in a month. The finding takes two minutes, the choosing one, and reviewing a drafted answer about five. Updated 2026-10-12 An AI visibility gap is a question your buyers ask an assistant where the answer names your competitors and leaves you out. You cannot see it in your analytics, because the visit never happens. Here is how to find one and close it today. - Your website address - The names of 2 to 6 competitors - An email address - About 10 minutes ## What is an AI visibility gap? When someone asks an assistant “who should I talk to about cyber insurance for a 40-person firm?”, the answer names three or four businesses. If yours is not one of them, that is a gap: a real buying question, a real recommendation, and you are absent from it. It is hidden for three reasons: - **No visit is recorded.** The buyer never reached your site, so nothing shows in analytics. - **Your search rankings look fine.** Assistants compose an answer from several sources; ranking for a keyword does not put you in it. - **The questions are unbranded.** They do not include your name, so brand monitoring misses them. > **Why a single question matters** Assistants favor pages that answer the question asked, directly and with sources. One good page on one specific question does more than a general page about your services. That is why this guide fixes one gap at a time. ## Step 1: Run the free AI visibility report **Enter your website at [endata.com](https://endata.com/#check), confirm your audience, competitors and topics, and open the link Endata emails you.** Endata reads your site and proposes who you serve, who you compete with and what you should be known for. Correct anything it got wrong: the competitors matter most, because the report is about who gets named in your place. It then writes six questions a buyer might ask, none of which names a business, and puts each one to ChatGPT, Claude, Perplexity and Gemini with web search on. That is up to 24 answers, read for every mention of you and of your competitors. The run takes a couple of minutes and an email tells you when it is ready. *A sample report. The bars show how many answers named each business; the list below shows the questions that left the firm out.* > **No account needed** The report needs an email address so the link reaches you. There is no password, account or card. ## Step 2: Pick the gap closest to a sale **From the questions you were left out of, choose the one a buyer asks just before choosing a supplier.** The report lists each missed question with the competitor an assistant named instead. Its action plan is already ordered for you, by two things it counts and does not guess: how many assistants named a competitor, and how close the question is to a purchase. *The action plan: each step answers one missed question, in priority order.* If you want to choose by hand, use this order: 1. **Buying questions** (“who should I hire for…”, “best … for a company like mine”). Fix these first. 2. **Comparing questions** (“X or Y for…”, “what does … cost”). 3. **Learning questions** (“how does … work”). Useful, but further from revenue. Prefer a question where several assistants named the same competitor. That is a consistent gap, not a one-off. ## Step 3: Turn the gap into an assignment **Click “Start fixing these gaps” in the report to open a workspace with your missed questions already in it.** Claiming the report creates your workspace from what you already confirmed: audience, topics, competitors and language. Each step of the plan becomes a question in the workspace. The top three are accepted for you and the rest wait as suggestions. Drafting starts once you choose a plan; Solo is $49 a month and includes 20 researched answers. Until then the workspace keeps your report and questions, and nothing is generated. > **Doing it by hand?** You can follow the rest of this guide without Endata. The work is the same: research the question, write a sourced answer, publish it where crawlers can read it, and retest. It simply takes hours where this takes minutes. ## Step 4: Review the researched draft **Read the draft answer, check its sources, add your own expert view, and approve it.** For each accepted question Endata searches recent news and web sources, keeps up to eight pieces of evidence, and writes an answer that cites that evidence claim by claim. You get a draft to edit, not a page that is already live. *A drafted answer waiting for approval: the question, the answer, and the sources behind it.* This is where your ten minutes are best spent. Three things make an answer worth citing: - **Answer the question in the first two sentences.** Assistants lift the opening. - **Keep the sources.** A claim with a named source is more likely to be repeated than an opinion. - **Add what only you know.** A sentence on how you actually handle this for clients is the part no competitor can copy. Endata has a field for exactly this, attributed to you. > **Don't write an advert** An answer that says “the best choice is us” will not be quoted. Answer the question fairly; your name is on the page, and that is what gets you mentioned. ## Step 5: Publish it where AI can read it **Approve the answer to publish it on your newsroom, at its own address, under your domain.** Publishing puts the answer on its own page, titled with the question. Every Endata newsroom is built to be read by machines as well as people: - **llms.txt:** A plain index of your answers that AI crawlers can fetch in one request. - **Markdown:** Every answer is also served as clean Markdown, with no layout to parse. - **A public MCP server:** Agents can search and read your answers directly, with no key. - **Your domain:** Use the hosted address, or connect your own domain on any paid plan. Prefer your existing site? Endata can publish to WordPress, Webflow, HubSpot, Ghost, Contentful and other CMSs. Whichever you use, the page carries a note that it was prepared with AI assistance and reviewed by your team. ## Step 6: Share the reporting, then retest in a month **Post a related article to your social channels, and compare next month's report against this one.** Assistants draw on what is being talked about. Endata also writes news articles on your topics, and those can be posted from the same place to LinkedIn, Bluesky, X and Pinterest, with short videos for TikTok and YouTube Shorts on the Business plan. Post one that covers the same ground as your answer and link the two. Then wait. On every paid plan Endata reruns the report each month with the **same six questions**, so the comparison is fair, and adds any new gaps as suggested questions. *Share of answers naming the firm, month by month. It moves both ways: AI answers vary from run to run.* ## What should you expect after a month? Be realistic. One published answer will not change every assistant's reply, and nobody can guarantee a mention: what an assistant recommends is its own decision. What you control is whether a clear, sourced, current answer to that question exists under your name. - **Straight away:** the page is live, announced to search engines, and available to assistants that search the web. - **At the monthly recheck:** you see whether the answers to that question now name you. Some will; expect movement in both directions. - **Over a quarter:** the pattern matters more than any single run. Businesses that answer one gap a week build a body of work assistants keep finding. ## Five mistakes that keep a gap open 1. **Answering a different question.** Title the page with the buyer's question, word for word. 2. **Burying the answer.** Put it first; give the background after. 3. **Publishing without sources.** Unsupported claims are the first thing an assistant drops. 4. **Hiding the page.** A PDF, a gated page or a page blocked to crawlers cannot be cited. 5. **Fixing it once.** News moves. An answer from last year loses to one from last month, which is why Endata proposes an update when the evidence changes. ## Frequently asked questions ### What is an AI visibility gap? A question buyers ask an AI assistant where the answer recommends your competitors and does not mention you. It is invisible in analytics because the buyer never visits your site. ### How do I check whether ChatGPT recommends my business? Ask it the questions your buyers ask, without your name in them, and see who it names. Endata's free report does this for six questions across ChatGPT, Claude, Perplexity and Gemini and lists the ones that leave you out. [Run it here](https://endata.com/#check). ### Can I really fix a gap in 10 minutes? Your part takes about ten minutes: two to start the report, one to choose a gap, one to claim it, five to review the draft and one to publish. The research and drafting happen in between. Whether assistants then name you is their decision, and you check it at the next monthly report. ### What is answer engine optimization (AEO)? The practice of making your business the source AI assistants draw on when they answer a question. In practice it means publishing direct, sourced answers to the specific questions buyers ask, in a form crawlers can read. It is also called generative engine optimization (GEO). ### Does this replace SEO? No. Assistants with web search read the same web that search engines index, so a well-built, well-sourced page helps both. AEO adds a focus on whole questions over keywords and on being quotable. ### How many questions should I fix? Start with one, then work through the action plan in order. The report's plan has up to five steps; the monthly recheck adds new gaps as it finds them. ### Which assistants does the report test? ChatGPT, Claude, Perplexity and Gemini, each through its maker's API with web search on. Those answers are close to, but not identical to, what each company's own app shows a given person. ### Is anything published without my approval? No. Every answer is a draft until someone on your team approves it, and an update to a live answer is a new draft too. ## Keep reading - [Endata vs. Semrush AI Visibility Toolkit](https://endata.com/compare/semrush-ai-toolkit): Tracking the numbers, or publishing the fix: a fair comparison. - [Automate it with the MCP server](https://endata.com/guides/mcp-content-pipeline): Run the same workflow from Claude Code or your own agent. - [Endata for agencies](https://endata.com/agencies): Do this for every client, under your own brand.