How to audit your site for AI visibility in 15 minutes

Most "AI visibility" advice falls into two buckets. The first is hype. The second is "buy our tool." This post is neither. It's a list of five concrete checks you can run against any live site in 15 minutes that will catch most of the issues we see in real audits.

If you'd rather have it done for you, Crawlmind runs all five on autopilot and watches them daily. But the manual version is honest work and worth doing once so you know what the tool is looking at.

#1. Is your llms.txt present, parseable, and current? (3 minutes)

Open https://yourdomain.com/llms.txt in a browser. Three failure modes to watch for:

• 404. You don't have one. Almost half of the sites we audit don't, including some that depend on AI traffic for revenue.
• 200 but malformed. Section headers missing, links broken, no one-line summary at the top. Run it through the parser at llmstxt.org and fix what it flags.
• 200 and well-formed but stale. Last update was six months ago, half the links 404 or redirect through old paths. AI engines that index from llms.txt treat staleness as a quality signal.

#2. What does your robots.txt say to AI crawlers? (2 minutes)

https://yourdomain.com/robots.txt. Search for these user-agents:

GPTBot
PerplexityBot
ClaudeBot
Applebot-Extended
CCBot
Google-Extended
Bytespider

Make a deliberate decision per bot. The default of "allow all" is fine for most SaaS, but agencies and companies that monetize content have a real reason to opt some out. The wrong move is to copy a robots.txt from a blog post without understanding what each bot does (GPTBot does NOT train models, for example, despite what most "block all AI" guides claim).

#3. Do your top pages have author and dateModified schema? (4 minutes)

Pick the three pages you care about most. View source on each. Search for "@type":"Article" or "@type":"BlogPosting". You should see:

• A Person or Organization author with at minimum a name and ideally a url to an author page
• datePublished AND dateModified, both ISO 8601 format
• An image with explicit width and height

If any are missing, the page is competing with one hand tied behind its back when AI engines decide which source to cite.

#4. Does ChatGPT actually cite you for your money queries? (3 minutes)

Open ChatGPT with browsing enabled. Ask it the three queries you most want to rank for, framed as questions a customer would ask. Not "best SEO tools" but "what's the best SEO tool for a 5 person SaaS team?"

For each answer, look at the citation chips at the end. If your domain isn't in the top 5, you have an AI visibility problem regardless of where you rank in Google. The fix isn't to ask ChatGPT nicely. The fix is to be the page that the engines decide is most useful for that question.

#5. Are your pages legible as "atomic answers?" (3 minutes)

AI engines lift sentences. Open one of your highest-traffic pages and ask: can a 25 word excerpt from this page stand alone as the answer to a specific user question?

If the first 100 words of the page are throat-clearing ("In today's fast-paced digital landscape..."), the engines have nothing to lift. Rewrite the lead so the first paragraph is a tight answer to a specific question, then expand.

#The pattern

If you score this honestly, you'll usually find that the gap isn't where you think it is. SaaS sites tend to nail performance and miss on schema. Agency sites nail schema and miss on robots-txt posture. Content sites nail everything except llms.txt, which they don't even know exists.

The good news is that all five issues are cheap to fix. The bad news is that nobody fixes them until they measure them. Do the 15 minutes.