OpenAI Founder Shows LLMs Write Like Aircraft Manuals – Impact for Biz

Summary: Andrej Karpathy, a founding member of OpenAI, proved that large language models can follow the strict ASD‑STE100 style used in aircraft manuals. For local businesses, this breakthrough points to a new way to create AI‑friendly content that ranks higher in AI‑driven search tools.

What Happened

Andrej Karpathy, one of the original architects of OpenAI’s early models, posted a striking experiment on X. He asked a leading LLM to generate text in ASD‑STE100, the controlled English format used for aircraft maintenance manuals. The result was a clean, highly structured explanation that read like a real cockpit handbook. Karpathy highlighted four progressive formats for model output – plain controlled English, diagram‑rich pages, full HTML layouts, and even explainer videos – each step improving readability and AI‑compatibility.

Key Details

• Who: Andrej Karpathy – founding member of OpenAI, now part of Anthropic’s pre‑training team.
• What: LLMs were prompted to write in ASD‑STE100, a stringent, rule‑based English style designed for safety‑critical manuals.
• Why It Matters: The constrained language forces the model to produce clear, unambiguous, and highly structured text – exactly the kind of output AI search engines love to cite.
• Four Output Levels:

1. Controlled English – plain text that follows the ASD‑STE100 rules.

2. Diagram‑Enhanced Docs – adding schematics and flowcharts.

3. HTML Pages – web‑ready, SEO‑friendly markup.

4. Explainer Videos – narrated, visual walkthroughs.

• Quote from Karpathy: “As models do more of the legwork, a lot more of our work will rise up the abstractions into oversight and understanding.”

What It Means For Your Business

1. Better AI Citations

AI‑driven search tools like ChatGPT, Perplexity, Gemini, and Google AI Overviews pull information from sources that are easy for the model to parse. Content written in a controlled, predictable style reduces the risk of misinterpretation, making it more likely to be quoted verbatim.

2. Faster Trust Building

When an LLM can instantly recognize your product description as a well‑structured manual, it treats the information as authoritative. That can translate into higher placement in AI answer boxes, voice assistants, and even emerging “AI‑first” SERPs.

3. Scalable Personalization

Instead of churning out generic blog posts, you can generate template‑based manuals for each service you offer (e.g., “How to Service Your Home HVAC Unit”). The controlled format keeps each version consistent while still allowing local nuances.

4. Lower Invalid Traffic

Recent data from AI Max showed a spike in invalid traffic before Black Friday. Structured, verifiable content helps AI filters weed out low‑quality clicks, protecting your ad spend and analytics.

5. Future‑Proofing Your SEO

As AI models become the primary gateway to information, the language you speak matters. Adopting a controlled‑English approach now positions your brand to stay visible when search evolves from keyword matching to semantic understanding.

Why It Matters

For businesses that depend on being found by AI assistants, the way information is written is becoming as important as the keywords themselves. AI models prioritize content that is clear, consistent, and structurally simple—the exact qualities enforced by ASD‑STE100. By aligning your copy with these standards, you increase the chance that ChatGPT, Gemini, or Google’s AI Overviews will quote your site directly, driving high‑intent traffic without paying for ads.

Moreover, as AI search tools replace traditional SERPs, the signal‑to‑noise ratio grows tighter. Content that looks like a safety‑critical manual passes AI quality filters more easily, reducing the risk of being labeled as spam or low‑value. This translates into better rankings, fewer invalid clicks, and a stronger brand reputation in the emerging AI‑first ecosystem.

Finally, the controlled‑English approach scales. You can produce a library of localized, service‑specific manuals that retain a uniform voice while still addressing regional nuances. This creates a personalized yet standardized content architecture that satisfies both human readers and machine learners, giving small businesses a competitive edge in the AI‑driven future.

Practical Steps to Leverage This Trend

• Create a Style Guide: Adopt key ASD‑STE100 rules – short sentences, active voice, consistent terminology.
• Use Prompt Engineering: When generating content with GPT‑4 or Gemini, include a system prompt like “Write the answer in ASD‑STE100 style.”
• Add Structured Data: Pair the clean text with schema.org markup (e.g., FAQPage, HowTo) to reinforce the hierarchy.
• Visual Supplements: Convert the text into diagrams or simple HTML tables; AI models can read both text and visual cues.
• Test with AI Tools: Run your pages through ChatGPT’s “cite source” feature to see if the model pulls your content.

Bottom Line

Karpathy’s experiment proves that quality beats quantity in the AI age. By shaping your content to match the disciplined tone of aircraft manuals, you give search‑grade AI a clear, trustworthy source to cite – a powerful advantage for any local or small business looking to be discovered online.

VisibilityAI helps local businesses turn technical breakthroughs into real‑world traffic. Want a custom guide on writing AI‑friendly content? Contact us today!

FAQs

• Q: Can I use ASD‑STE100 without being an aviation expert?

A: Yes. The core rules focus on short sentences, active voice, and consistent terminology. You can adopt a simplified version for any product or service.

• Q: Will structured English hurt my brand’s tone?

A: Not if you blend it with a friendly voice in marketing copy. Use the controlled style for core explanations and keep your brand voice in promotional sections.

• Q: How do I test if AI is actually citing my content?

A: Ask a generative AI tool a question that your page answers and look for the citation link. You can also use tools like the OpenAI “search” endpoint to see if your URL appears in the results.