Takeaways

What to do with all this data. Benchmarks, closing arguments, and why this is an exciting time to be a technical writer.

Make it actionable

Data is useful when you can do something with it.

1. Create benchmarks

   Use the numbers from this dataset as a starting point for your own. AI readership at 35%, one-and-done chat sessions at 55%, agentic PR merge rate at 55%+. Your numbers might be different depending on your goals. These are starting points.

2. Set goals

   Pick one metric from each layer and set a direction. For discovering: do you have llms.txt? For reading: what’s your one-and-done conversation rate, and what would better look like? For writing: if you’re using AI writing tools, what’s your PR merge rate?

3. Don't forget the qualitative data

   The numbers show you what happened. They don’t show you why. User interviews, session replays, and support tickets fill in what the data can’t. Especially for the reading layer — the frustration rate tells you something’s wrong; qualitative data tells you what.

Write the system prompts and context

Everything that powers agents-documentation for them to answer questions, systems prompts, skills—is technical writing.

The scope of the craft is expanding. New specialties of technical writing are here, they might just not be called “technical writing.”

But technical writers are the ones with the best skills to write the prompts and context for high quality agent interactions.