Have you ever wondered who is reading your documentation? For decades, technical writers have written primarily for people. Today, documentation has another increasingly important reader: the AI agent.

The shift is already visible in Internet traffic. Cloudflare reports that, as of June 2026, 52% of crawler requests were associated with AI training, while mixed-use crawlers combining search, agent use, and training represented more than 36% of crawler activity (Cloudflare Agentic Internet Bot Report). Akamai recorded a 300% increase in AI bot activity during 2025 in its analysis of global bot traffic trends (Akamai report on AI bot activity).

Not all of this traffic comes from agents acting directly on behalf of users, but the direction is clear: more documentation is being discovered, retrieved, processed, summarized, and acted upon by AI systems. This creates a new requirement for technical writers: communicate the right information while consuming as little unnecessary context as possible.

Not all of this traffic comes from agents acting directly on behalf of users, but the direction is clear: more documentation is being discovered, retrieved, processed, summarized, and acted upon by AI systems. This creates a new requirement for technical writers: communicate the right information while consuming as little unnecessary context as possible.

Here are five ways to do it.

1. Make topics self-contained

AI agents rarely need your entire documentation set. They retrieve specific pages, sections, or chunks relevant to the task they are trying to perform. That means a section should contain enough context to make sense on its own.

Avoid references such as:

"As explained previously, configure the setting accordingly."

Instead, identify the setting, requirement, or condition explicitly. The additional few words can actually save tokens overall if they prevent the agent from having to retrieve another page to understand the instruction. Think in independent information units, not just pages.

2. Put the important information first

Don't make an agent read three introductory paragraphs before discovering what a feature does. Start sections with the information most likely to answer the question (What is it? What does it do? When should I use it?) then provide the details.

This is good technical writing for humans too, but it becomes especially important when documentation is retrieved as context. A well-written opening makes it easier for an AI system to determine whether a section is relevant before consuming additional content.

3. Use Markdown for clear, predictable structure

Make content easier to interpret by using  headings, lists, tables, parameter definitions, code blocks, etc.. The industry is already moving in this direction. In 2026, Cloudflare introduced Markdown for Agents, which lets AI systems request a Markdown representation of a web page instead of its full HTML. Cloudflare says Markdown's simpler structure reduces unnecessary token overhead and makes content easier for AI systems to process.

In one example in Cloudflare's documentation, an HTML response representing 12,345 tokens was reduced to 725 tokens when delivered as Markdown. That is a specific example rather than a universal benchmark, but it illustrates just how much irrelevant page structure an agent may otherwise have to process.

‍

4. Be ruthless about redundancy

Today, every unnecessary word may consume not only your reader's attention, but also an AI agent's tokens. However, token optimization does not mean making documentation as short as possible. Instead, look for token waste:

• Repeating the same explanation across multiple sections

• Long introductions that add little information

• Restating what a heading already says

• Using three terms for the same object

• Adding marketing language inside procedural documentation

• Repeating obvious UI information without explaining what it means

5. Use consistent terminology

Humans are surprisingly good at understanding that "integration," "connection," and "connector" might refer to the same thing. AI agents can usually infer this too, but every ambiguity increases the amount of context required to interpret the documentation correctly.

Choose one term for each product concept and use it consistently. Make relationships explicit. Prefer:

"A Workflow processes one or more Insights."

over:

"It processes the relevant items."

Clear nouns, explicit relationships, and consistent terminology make content easier to retrieve, combine, and act upon.

‍