Your Website Is Invisible to ChatGPT. Ours Was Too.
TL;DR: When a plant manager asks ChatGPT to recommend an AI consultant for energy operations, ChatGPT reads websites to build its answer. It does not run JavaScript — and most modern websites are pure JavaScript. Ours was. On a 100-point AI-findability audit, novanix.ai scored 18. Nine days later it scored in the high 70s. This is exactly what was broken, what we fixed, and a five-minute check you can run on your own site today.
The problem nobody can see
Your buyers changed how they search. They don't type keywords into Google and click through ten links anymore. They ask ChatGPT, Claude, or Perplexity a full question — "find me an AI consultant who understands industrial operations" — and get two or three names back. If your website can't be read by those systems, you are not one of the names. There is no page two to be buried on. You simply don't exist.
Here's the technical trap: AI crawlers read the raw HTML your server sends. They do not execute JavaScript. A modern single-page web app — React, Vue, most sites built in the last five years — sends an empty shell and lets JavaScript paint the content afterward. Humans see a finished site. Crawlers see this:
<div id="root"></div>That was our entire website, as far as AI search was concerned. Every service description, every article, every credential — invisible.
What the audit found
We scored novanix.ai against a 100-point rubric covering crawlability, structured data, citability, and authority signals. Results:
- 18/100. The served HTML contained one sentence — the meta description.
- No structured data of any kind. To an AI system, "Novanix" was an unidentified string, easily confused with similarly named companies.
- Two different machine-readable descriptions of who we serve — an old one and a current one — feeding contradictory facts to any system that looked.
- No sitemap. Unknown URLs returned the homepage instead of an error, so crawlers couldn't tell real pages from dead ends.
The fix, in five moves
1. Put the content in the HTML.
We prerendered every page at build time. The same site, the same design — but now the full text of every page ships in the initial HTML. This was 80% of the problem and the first thing we fixed.
2. Tell one story everywhere.
One positioning sentence — who we are, who we serve — now appears identically in the meta tags, the structured data, and our llms.txt file. AI systems triangulate identity across sources; contradictions cost you trust you never knew you were losing.
3. Welcome the crawlers explicitly.
Our robots.txt now names GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot with explicit allow rules, and points to a clean sitemap.
4. Make identity machine-readable.
Organization, Person, and Article structured data now describe the company, the founders, and every article — including author credentials and publication dates, which AI systems weigh when deciding what to cite.
5. Finish the hygiene.
Unique title and description for every page, one canonical domain, and real 404s for pages that don't exist.
Nine days after the first audit: high 70s out of 100.
What a score doesn't tell you
Honesty matters here. That score measures readiness — whether AI systems can read, understand, and attribute your site. It is not a guarantee of citations. Getting recommended takes readiness plus authority: consistent presence beyond your own site, content worth citing, and freshness. Industry data suggests structured data gets picked up within weeks, while authority compounds over months. Readiness is the entry ticket. We're now in the building-authority phase, and we'll publish what we learn.
The other half is having something worth citing. For industrial operators that usually starts inside the business rather than on the website — making your own technical documentation retrievable is the same problem pointed inward.
The five-minute check for your own site
- The view-source test. Open your website, right-click, "View Page Source." Search for a sentence you can see on the page. If it isn't in the source, AI crawlers can't see it either.
- The robots test. Visit yoursite.com/robots.txt. If it blocks GPTBot or simply never considered AI crawlers, you're leaving visibility on the table.
- The ChatGPT test. Ask ChatGPT: "What does [your company] in [your city] do?" If it doesn't know, or describes a different company with a similar name, your buyers are getting that same answer.
If any of those three fail, you have the same problem we had — and it's very fixable.
Lesley Ward
Founder & CEO, Novanix AILesley spent over 10 years building and deploying AI systems as a Forward Deployed Engineer in energy and industrial operations. She co-founded Novanix AI to help energy, construction, and industrial companies deploy AI that is technically sound, operationally trusted, and governance-ready. Her work spans predictive maintenance, document intelligence, compliance automation, and AI strategy for critical infrastructure. Lesley holds a cybersecurity certification and an AI safety certification. She is an active participant in the AI security/AI safety community.
Novanix AI is building the security and governance infrastructure for AI deployment in industrial environments. To follow the development of the Industrial MCP Security Framework, follow Novanix AI on LinkedIn.
AI in the Field
Practical insights on deploying AI in energy, construction, and industrial operations. No hype, no fluff. Just what works in the field.