The AI Visibility Playbook: What AI Search Sees — and How We Fix It

Here's an uncomfortable experiment for any business owner: fetch your website the way an AI crawler does — no JavaScript, no rendering, just the raw HTML your server returns. We ran that experiment on our own site. The result: every single page returned the same 4,137 bytes of empty HTML.Our healthcare page, our case studies, our services — to ChatGPT's crawler, they were all the same blank container with a title tag.

The site looked great to humans. Modern JavaScript sites often do — the browser runs the code and builds the page. But most AI crawlers don't run JavaScript. OAI-SearchBot (ChatGPT), ClaudeBot, and PerplexityBot read the raw HTML and move on. If your content only exists after JavaScript runs, you don't exist to them.

Step 1: Test what the crawlers actually see

Don't guess — fetch. We requested our pages using the exact user-agent strings of six crawlers: OAI-SearchBot, GPTBot, Googlebot, Bingbot, ClaudeBot, and PerplexityBot. Two findings mattered. First, all six got HTTP 200 responses — nothing was blocking them, which is the failure everyone checks for. Second, what they received was empty — which is the failure almost nobody checks for. The doors were open; the rooms had no furniture.

Step 2: Serve real HTML on every page

The core fix was pre-rendering: at build time, every public route gets written out as complete HTML — full body content, correct title, meta description, canonical URL, structured data — and the interactive JavaScript layer loads on top. Visitors notice nothing. Crawlers get everything.

The measurable difference: our healthcare page went from the same 4 KB shell as every other URL to 39 KB of its own crawlable content. We verified by re-fetching as OAI-SearchBot and finding the page's actual headline in the response.

Step 3: Fix the identity signals

AI systems don't just read pages — they resolve entities: who is this business, where is it, who runs it, what does it know? Our audit found the sitewide Organization data pointing at an image on a site-builder's storage bucket, a social-card attribution crediting the site-builder's brand instead of ours, and no address, personnel, or expertise areas anywhere in the structured data. Every one of those is now corrected: our own logo on our own domain, our location, our people, our service areas — stated once, machine-readably, on every page.

Step 4: One page, one topic, one canonical answer

We found two pairs of near-duplicate pages competing with each other and a homepage whose main heading was a single rotating word — visually striking, meaningless to a machine. Duplicates now redirect to a single canonical page, and the homepage's primary heading states, in one crawlable sentence, who we are and which industries we serve. If a machine can't compress your page into a clear answer, it won't use your page as one.

Step 5: Give the machines something worth citing

Crawlability gets you read; it doesn't get you referenced. AI assistants cite pages that explain, define, and evidence claims — which is why the same overhaul included case studies with verifiable outcomes and expert-authored articles with named authors in the structured data. A crawlable brochure is still a brochure.

The checklist

• Fetch your key pages with AI crawler user-agents — read what comes back, not just the status code
• If content only appears after JavaScript runs, pre-render or server-render every public page
• One title, one meta description, one canonical per page — in the HTML, not injected afterward
• Organization structured data with your real logo, location, people, and expertise areas
• Kill duplicate pages; make every main heading state its page's answer in plain language
• Publish content that answers real questions, with named authors and honest claims
• Re-test with the same crawler fetches — and keep re-testing monthly

A note on scope, because it matters when you're deciding what this takes: the technical fix is fast when you know exactly what you're doing — the rendering and structured-data work above was a single focused day of engineering. Becoming a business AI systems consistently recommend is the ongoing part: content worth citing, authority built off your own site, and monitoring that tells you what's working. The fix opens the door; the system walks through it.

None of this is a trick, and that's the point. Everything above makes the site better for human visitors too — faster first paint, clearer messaging, honest metadata. AI visibility done right is just clarity, made machine-readable. If you want the deeper breakdown of how these systems decide what to recommend, we've written up what AI visibility actually is — and the packaged version of this entire process lives in our AI Visibility Package.