Can AI see your website? How to check
Plenty of good websites make it harder for AI answer systems and search features to understand them. These checks separate measurable recommendation visibility from crawler and markup proxies.
Check 1: is your content in the raw HTML?
Right-click your homepage, choose View Page Source, and search for a sentence you know appears in your main content - a product description, a service line, a paragraph from your about page. If you cannot find it in that raw HTML, the browser is building your content afterwards, once JavaScript runs and hydrates the page. Some crawlers and answer systems may render JavaScript, but many lightweight fetches, previews and extraction workflows work best when the important text is present in the initial response. This is a practical visibility risk, not proof of exclusion from every AI system. The fix is prerendering your public pages at build time, or rendering them server-side per request, so the HTML that arrives at the crawler already contains the words a customer would read.
Check 2: are AI-related crawlers handled deliberately?
Open yourdomain.com/robots.txt in a browser and read through the Disallow lines, checking specifically for the agents relevant to your policy and measurement goal. Be precise: a model-training crawler, a live-search fetcher, Googlebot, Google-Extended and a third-party answer bot do not all mean the same thing. Blocking a crawler is a legitimate, defensible choice when you make it deliberately, but a copied rule sitting untouched from an old template is not a current policy. Record what is blocked, why, and which measured prompt or search outcome you expect it to affect.
Check 3: do you have accurate structured data?
Paste your homepage URL into Google's Rich Results Test or validator.schema.org and see what markup comes back. At minimum, Organization or LocalBusiness schema can state who you are; Product, Service and BreadcrumbList markup can help where they match visible page content. Structured data matters because it reduces ambiguity around page facts, not because JSON-LD guarantees a ranking, rich result, or AI citation. Use supported markup that matches what visitors can actually see, and avoid adding retired or irrelevant types just to make a tool go green.
Check 4: does llms.txt exist, and is it accurate?
Visit yourdomain.com/llms.txt directly. Most sites return a 404 today, which is normal because the format is unofficial and optional. If the file exists, check whether it states current facts, links to canonical pages, and avoids promotional filler. If it does not exist, that is not automatically a defect; Google says no AI text file is required for AI Overviews or AI Mode. The practical question is whether you have a known consumer or workflow for the file, and whether maintaining it improves factual consistency.
Check 5: ask a fixed prompt panel
Use a small, fixed panel of realistic prompts instead of one ad hoc question: what does [yourdomain.com] do, best [category] in [location], alternatives to [competitor], is [business] legitimate, and a buying-intent comparison prompt. Record whether the business appears, how it is described, which competitors appear, and which sources the answer cites. A vague or stale answer can point you back to the previous checks, but the important part is repeatability: the same prompts, same notes, and same schedule so you are measuring recommendation share rather than chasing one lucky response.
What weak visibility can cost you
None of this is abstract. When someone asks an assistant to recommend a supplier, compare options, or explain what a business does before they visit its website, the answer depends on what the system can retrieve, interpret and corroborate. A site that fails these checks may be absent, described vaguely, or beaten by competitors with clearer public facts. The frustrating part is that many of these failures are configuration or clarity problems rather than product problems. That makes them worth fixing once measured, but it does not justify claiming traffic loss or penalty without Search Console and analytics evidence.
Start with a technical baseline
A free Quick Scan checks selected public website signals and returns a surface score, six result categories and selected issue previews in about 60 seconds. It does not run the complete nine-suite Full Audit or measure recommendation share in AI assistants. Use it to find a starting point, then work through the checks above and record the fixed prompt panel. For the deeper diagnostic, a Full Audit is a separate, account-required path.
Frequently asked questions
My site ranks well on Google - surely AI can see it?
Not necessarily, and this is the assumption that catches out the most otherwise well-run websites. Googlebot renders JavaScript through a full headless browser as part of its indexing pipeline, so a page built with client-side rendering can still get crawled, rendered and ranked normally in classic search. Some AI retrieval systems, previews and extraction workflows may not render the page the same way, or may rely more heavily on the initial HTML response. The result is a site that ranks respectably on Google while exposing much less useful text to another system. The raw-HTML check in this article is a fast way to spot that risk, then the prompt panel tells you whether it is affecting real recommendation outcomes.
Which AI crawlers matter in 2026?
The practical set depends on your policy and measurement goal. GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Googlebot, Google-Extended and Bingbot each do different jobs, and some relate to training while others relate to search or retrieval. The distinction matters because blocking one does not necessarily block another even when they come from the same company. Treat crawler access as one input to investigate alongside prompt-panel results, not as proof that a business will or will not appear in live AI answers.
Is AI visibility worth it for a local business?
Yes, and increasingly so, because the kind of question people used to type into Google as a list of blue links - best cafe near me, reliable electrician in [suburb], who does [service] locally - is exactly the kind of question people now ask an assistant directly, expecting one confident recommendation rather than ten options to compare themselves. Local businesses with strong review signals, consistent entity information across directories, and readable, well-structured websites are what these models reach for when they have to name a specific business rather than describe a category. A local business that has never thought about AI visibility isn't neutral in that moment - it is simply absent from the shortlist the assistant offers.
Start with a free Quick Scan
A surface score, six result categories and selected issue previews in about 60 seconds. No signup or card. This technical baseline does not measure AI recommendation share.
Run free Quick Scan