Skip to main content

Solutions

Products

Recommended products

Questions? We are here to help.
Contact us

Purchase

Questions? We are here to help.

Resources

Content guidelines for Help and Support

Writing for AI

Generally speaking, we should be writing for people and structuring for Artifical Intelligence (AI).

In-line links/duplicative content

Every time we use an in-line link to another article, the system creates a copy of the original. Content that is repeated is considered by AI to be more important, even if we're linking to reference type information. This can make inconsequential content weighted more heavily in AI responses, giving an answer that isn't as useful or helpful as it could be.
If you have several articles repeating the same content, consider restructuring the articles. You may be able to combine them together or pull the repeated part out into a reference type article and link to it as a related content link.

Tone of voice/language/word choice

AI recognizes patterns, so if we're consistent in our word choices and writing style, when the AI responds it gives consistent responses and sounds more like us.
Using approachable, human language has several benefits:
  • Reduces reading level, which increases ease of comprehension.
  • Makes AI responses sound more "human".
  • Makes AI responses easier to read and understand.
  • Is more accessible to all.
  • Brand compliance.

Images

AI can't give images in responses or use information in images.

DITA/structured documents

DITA (Darwin Information Typing Architecture) helps AI understand context better. Without a "wintitle" tag, for example, the AI doesn't understand that a particular phrase is the name of a screen in the application.
With better context, you get better answers.
The DITA standards have been around for more than 20 years.
This goes beyond the tags on individual words or phrases - the overall structure of the document is important too. How are you using headings? Subheadings? Lists, paragraphs, blocks of code?

Training

A note about training AI: It's really hard to make an AI "forget" things it was trained on, so it's best to start with a really good content dataset. Making the above changes after the AI has been trained is not as effective as making the content great before training.

This article applies to:

Related Content