--- title: "AI Search Visibility: A Seven-Blocker Audit You Can Score in an Hour" description: "Score seven reasons assistants answer your question without naming you, from edge-level bot policies and facts trapped behind tabs to a brand nobody else writes about. Each blocker has a five-minute test and one decision." author: "Team" category: "Search & AI Visibility" date: 2026-10-23T10:10:01.476Z canonical: "https://contentagents.dev/blog/why-you-are-missing-from-ai-answers-a-seven-blocker-audit" --- # AI Search Visibility: A Seven-Blocker Audit You Can Score in an Hour ![common blue, butterfly, insect](https://cdn.pixabay.com/photo/2023/09/30/10/53/common-blue-8285440_1280.jpg?w=1200&q=75) > Score seven reasons assistants answer your question without naming you, from edge-level bot policies and facts trapped behind tabs to a brand nobody else writes about. Each blocker has a five-minute test and one decision. A page of yours holds position three for a question. An assistant answers that exact question and names four other companies. Nothing is broken, nothing is reported, and the gap costs you the part of the funnel that never reaches your analytics. Seven things usually explain it, and five of them are an afternoon's work. For marketers and site owners who want a diagnosis before they commission anything. Allow an hour with your robots.txt, your CDN dashboard and two assistants open. If your site is not indexed in ordinary search at all, stop here and fix that first; nothing below will help. ## Scoring, in three lines Give each blocker a 0 if it is genuinely fine, a 1 if it is partly true of you, and a 2 if it is clearly the problem. Seven blockers, so fourteen is the worst possible total. AirOps makes the same argument in its post on [AEO and GEO blockers](https://www.airops.com/blog/geo-best-practices-blockers), where Josh Spilker contends that invisibility comes from a short list of fixable obstacles rather than from weak writing. That is a vendor's framing; this audit is how you test it against your own domain. ## Blocker one: there is no index entry to draw on **Symptom.** The page does not surface in Google even when you search its exact title in quotation marks. **Five-minute test.** Search the title as a quoted phrase, then run the URL through Search Console's inspection tool. Google's documentation on [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) states that a page has to be indexed and eligible to appear with a snippet before it can show up as a supporting link in those features. **Fix.** Clear the noindex directive, repair the canonical, unblock the path. One decision: make the page eligible or accept that it is out of scope. **Not fixed by this.** Being worth including. Google is explicit that meeting the conditions guarantees no crawling, indexing or serving. ## Blocker two: something between you and the crawler is refusing **Symptom.** Your access rules deny agents you never intended to deny, or your edge network denies them without telling your robots.txt. **Five-minute test.** Read your robots.txt for broad denials, then, separately, open your CDN's bot settings. This second half is new and most audits omit it. Cloudflare's documentation on [blocking AI bots](https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/) now sorts them into three behavioural groups, Search, Agent and Training, with a per-group choice of blocking everywhere, blocking only on pages carrying ads, or allowing. It also sets new defaults for newly onboarded domains from 15 September 2026: Training and Agent blocked on ad-bearing pages, Search permitted. A decision made in a dashboard outranks whatever your robots.txt says. AirOps, citing Cloudflare, notes that only around [14 per cent of the top ten thousand domains publish any AI-specific crawler directive](https://www.airops.com/blog/geo-best-practices-blockers), which means most sites are living with somebody else's default. Our walkthrough on [reading your access logs](/how-to-read-server-logs-to-see-whats-actually-crawling-you-0ae4) shows you what is actually arriving. **Fix.** Decide per group and per surface, deliberately, in both places. **Not fixed by this.** Selection. Access buys you a ballot, not a vote. ## Blocker three: the answer sits nine paragraphs down **Symptom.** Your page answers the question it promises, eventually, after the context and the history and the definitions. **Five-minute test.** Read the opening of the page and nothing else. Does a reader now have the answer? Then do the same under each heading. **Fix.** Open every section with the answer in one or two sentences and expand underneath. One decision: the conclusion goes first. **Not fixed by this.** Having nothing to add. Reordering a page that restates other pages changes the order, not the value. ## Blocker four: the facts are not text **Symptom.** Your prices, specifications or comparisons live inside an image, behind a tab, or in markup a script writes after loading. **Five-minute test.** Open view-source, not the inspector, and search for your key figures. Then click nothing and read what remains. Google's guidance asks for important content to be present in textual form, and two 2026 investigations show why that phrasing is doing heavy lifting. RESONEO's January 2026 [sentinel test](https://think.resoneo.com/sentinel/geo-llm-crawler-report.html) planted markers in fifteen different injection methods and found ChatGPT and Claude reading only server-delivered markup, Bing's assistant allowing roughly fifty milliseconds for scripts, and Google's AI Mode unable to fetch a URL on request at all. Peec's logs of Deep Research sessions, snapshotted in June 2026, add the detail that [the agent's click command was never used once](https://docs.peec.ai/research/how-deep-research-reads-your-site.md), so an accordion or a filtered table is unreachable even when the text is technically in the response. **Fix.** Put the facts in plain markup, visible without interaction, and let images support rather than carry them. Our breakdown of [what an assistant actually reads](/what-chatgpt-actually-reads-on-your-page) goes through the mechanics. **Not fixed by this.** Facts that are wrong. Legibility makes an error easier to quote. ## Blocker five: your name appears only on pages you own **Symptom.** Searching your brand returns your site, your social profiles, and nothing written by anybody else. **Five-minute test.** Search the brand as a quoted phrase and count independent domains in the first twenty results. Fewer than three is a 2. **Fix.** Earn coverage with original data and named expert commentary. Ahrefs examined 75,000 brands and found branded web mentions carrying [the strongest correlation with AI Overview visibility in the study, at 0.664](https://ahrefs.com/blog/ai-overview-brand-correlation/), with the author noting plainly that correlation is not cause. Our [audit of a PR operation](/your-pr-machine-has-five-blockers-score-yours-in-45-minutes) covers the mechanics of earning them. **Not fixed by this.** Speed. This is the blocker measured in quarters. ## Blocker six: your own pages contradict each other **Symptom.** Pricing, documentation and a two-year-old post give three different answers to the same question. **Five-minute test.** Take five facts buyers ask about and check each one everywhere it appears. Any disagreement scores at least a 1. **Fix.** One authoritative location per fact, with everything else updated or removed. AirOps reports that over [70 per cent of cited pages had been updated within the previous year](https://www.airops.com/blog/geo-best-practices-blockers), which is its own unpublished figure, but stale contradictions damage a reader's trust whether or not they affect a citation. **Not fixed by this.** Outdated claims on somebody else's site. Those need an email. ## Blocker seven: nobody has measured any of this **Symptom.** Asked how often assistants name you, the honest answer in the room is a shrug. **Five-minute test.** Put five buyer questions into two assistants and record whether you appear. Then do it properly with our [spreadsheet method](/track-ai-citations-with-a-spreadsheet-a-four-step-playbook), because a single pass is close to meaningless: Gregory Druck and Ethan Smith of Graphite, working from two hundred prompts run four hundred times each in April 2026, report that it takes about ten runs per prompt to get the average error on a visibility estimate down to [roughly 5.6 per cent](https://graphite.io/five-percent/research/demystifying-randomness-in-ai). **Fix.** One fixed prompt list, one date each month, one owner. **Not fixed by this.** Causes. Pew's tracking found clicks on links inside an AI summary happening in [about 1 per cent of visits](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/) in March 2025, so your traffic reports will understate your presence regardless of what you fix. ## Reading your total Out of 14Where to put the hour 0 to 4Foundations are sound. Spend the time on outside mentions and on a monthly measure. 5 to 9Work blockers one to four. They are the cheap ones and they compound. 10 to 14Stop commissioning content. Access, structure and factual consistency come first. These bands are our working heuristic, not a finding from research. ## The order that costs you least - Access first, in both places: robots.txt and the CDN dashboard. Hours of work, and it gates everything else. - Then the opening paragraphs of your five highest-intent pages. An afternoon. - Then move the facts out of images and out from behind interactions. - Start measuring before you begin any larger project, so the project has a baseline to be judged against. ## Access stopped being a robots.txt question The position we will defend: for a site on a managed edge network, auditing robots.txt alone is now the wrong first check, and the advice to start there, which we have given ourselves, is about a year out of date. Open the CDN's bot policy first. It is the layer that decides, and it is the layer nobody on the marketing team can see. The evidence sits in the vendor's own documentation. Cloudflare's current page describes three behavioural categories, Search, Agent and Training, each with its own setting, and sets defaults for newly onboarded domains from [15 September 2026 that block Training and Agent traffic on ad-bearing pages](https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/) while permitting Search. Read that as an operator rather than a publisher and two things follow. A decision with real consequences for your distribution now ships as a default, and the categories do not map onto the bot names your robots.txt lists, so the two files cannot be reconciled by reading one of them. There is a second reason to invert the order, and it is about who holds the lever. A robots.txt line is usually owned by whoever owns the site, which is often the team running this audit. A bot policy is owned by infrastructure or security, who changed it for reasons that had nothing to do with search and will not think to mention it. The blocker you cannot see is the one that stays unfixed for a quarter, so it goes first precisely because it is somebody else's screen. None of this holds for a site serving directly from its own origin with no managed bot filtering in front of it. There, robots.txt genuinely is the access layer and the old order is right. The test for which situation you are in takes a minute: look at who answers your DNS, then open their bot settings and see whether anything is already switched on. ## Questions that follow this audit ## FAQ ### What counts as AI search visibility? How often, and how accurately, assistants and AI search features name or cite your brand for the questions you ought to be able to answer. It is measured as frequency across many prompts and runs, not as a position. ### Do AI Overviews need their own markup? Google says no. Its documentation states there are no additional requirements and no special schema.org type for AI Overviews or AI Mode. A page must be indexed and eligible to appear with a snippet, and that is the condition. ### Is robots.txt enough to control crawler access? Not on a managed edge network. Cloudflare's documentation describes per-group policies for Search, Agent and Training bots configured in its dashboard, with new defaults for newly onboarded domains from 15 September 2026. Check that screen as well as the file. ### How fast should this audit show results? No reliable timeline exists. Unblocking a crawler or moving an answer to the top of a page can register within weeks. Earning independent coverage is measured in quarters. Re-run the same prompts a month later and compare rates rather than single answers. --- Source: https://contentagents.dev/blog/why-you-are-missing-from-ai-answers-a-seven-blocker-audit