Andrej Karpathy, described in the source report as an OpenAI founding member, has suggested a practical way to improve large language model output: ask the model to write in a tightly defined style, such as an aerospace manual standard, or to respond in a specific format such as diagrams, HTML pages, or custom explainer materials.

The development is not a new product launch or regulatory change. Instead, it reflects a broader shift in how businesses are learning to use generative AI more effectively: by giving models clearer structure, stronger constraints, and more explicit output requirements.

What happened

According to Search Engine Journal, Karpathy suggested that users can get more useful results from LLMs by prompting them to explain topics using highly structured writing conventions, including the style of an aircraft manual. He also suggested asking models to produce outputs in alternative formats, including diagrams, HTML pages, or custom explainer videos.

These suggestions are best understood as usage guidance rather than a confirmed new capability or standard. The underlying idea is that LLMs often perform better when businesses specify how an answer should be delivered, not just what should be answered.

Why it matters for European businesses

For SMEs and digital teams, the main implication is operational. Many businesses are already experimenting with AI for internal documentation, customer support content, marketing drafts, onboarding materials, and knowledge-base creation. A common challenge is inconsistency: AI outputs may be correct in parts but too vague, too conversational, or poorly structured for business use.

Structured prompting can help reduce that problem. Asking an LLM to follow a formal style, a predefined layout, or a machine-friendly format may make outputs easier to review, reuse, and integrate into existing workflows.

This matters in several business contexts:

• Documentation: IT, operations, and product teams may get clearer step-by-step instructions when outputs follow a strict format.
• Marketing production: teams can request landing-page drafts, FAQ blocks, or HTML-ready content in a more usable structure.
• Automation: businesses using AI inside workflows may benefit when outputs are predictable enough to feed into CMS, intranet, support, or training systems.
• Internal training: companies can experiment with AI-generated explainers tailored to employees, provided outputs are reviewed before use.

For European organisations, this is especially relevant where accuracy, traceability, and internal approval processes matter. In regulated or quality-sensitive environments, a more formal output style may support human review, although it does not replace it.

Who may be affected

The most relevant groups are businesses already using or testing generative AI in day-to-day operations:

• SMEs creating internal process documentation without dedicated technical writers
• Marketing teams producing structured web content or campaign assets
• IT and digital teams using AI for support articles, SOPs, and implementation notes
• E-commerce businesses building product guidance, help-centre content, or templated page elements
• Digital agencies looking for repeatable workflows across client content and website production

Companies in regulated sectors may also find the idea useful, but they should be cautious not to confuse a polished format with verified accuracy or compliance.

What companies should consider

European businesses experimenting with LLMs may want to treat this as a workflow design lesson rather than a headline technology shift.

• Define output formats clearly: specify whether the response should be a checklist, SOP, HTML section, summary table, or training note.
• Use style constraints: ask for concise, procedural, unambiguous language when documentation quality matters.
• Build review steps: formal-looking AI output still requires human checking, especially for legal, technical, security, or customer-facing use.
• Test structured prompts in repeatable tasks: start with internal documentation, support content, or draft web components where consistency matters.
• Avoid over-reliance on presentation quality: a well-formatted answer can still contain errors, omissions, or outdated information.

In practice, the value for businesses is not that LLMs can imitate an aircraft manual style. It is that more precise prompting may help companies turn AI from a general writing tool into a more controlled business productivity tool.