AI as a writing tool

Guidelines for intentional, safe, and supportive use

Purpose and philosophy

As writers in web3, we’re aiming for trust. Every interaction carries weight, and our words should help people feel confident, not confused or coerced. AI is a tool to support that work, not replace it.

Used well, it helps us:

  • Explore faster

  • Test more options

  • Scale consistency across complex, high-stakes flows

When to use AI

AI is here to help us do better work, not shortcut the thinking that good content design requires.

✅ Use it for…

  • Exploring tone or messaging

  • Revamping technical copy for comprehension

  • Generating ideas for errors, fallback states, or empty UIs

  • Supporting consistency across patterns and surfaces

  • Unblocking creative thinking

⛔ Don’t use it for…

  • Writing final user-facing copy without content design review

  • Handling compliance, regulatory, or legal messaging

  • Replacing our own judgments about clarity, trust, or safety

  • Sensitive or high-trust flows

💡 Pro tip: If you wouldn’t assign it to a brand-new intern with a slight lying problem, don’t assign it to AI.

But first, some ethical considerations

AI is fast, but not neutral. It reflects the input and the biases of its training and users (ahem, like you, dear reader).

So before you get started, a word to the wise…

  • Always double-check for tone, accuracy, and impact

  • Never outsource judgment about risk or comprehension

  • Review the output like it matters (because it does)

  • AI is a really amazing assistant, but it’s nothing without you

How to write an effective prompt

Context, constraints, and options are the name of the game.

Step 1: Include context

Before you begin, be sure the GPT knows:

  • The audience

  • The goal

  • The risks for the user (and for you, too!)

For example, you might say:

Write 3 permission messages for a crypto staking feature. Your tone should be clear, transparent, and lay out the risks involved in a positive way.

Step 2: Set the parameters

Always tell the AI things to help make the output more accurate, which includes:

  • Space or character limits

  • Tone guidance

  • Functional rules (like if an action is irreversible, for example)

Step 3: Options, please!

Always ask for 3 versions of the same prompt, focusing on different areas of emphasis like:

  1. User control

  2. Safety

  3. Urgency without pressure

✋ A cheat sheet for busy folks ✋

Before you ship anything a GPT has created for you, run through this cheat sheet to ensure the output is up to the task.
Always check for…

Clarity

  • Is the meaning obvious at a glance?

  • Can a new user understand it?

  • Is this an appropriate reading level for your audience? (Not sure? Paste it into Hemingway Editor)

Accuracy

  • Does it reflect actual product behavior?

  • Are there any hallucinations or false claims? (Spoiler: there probably are)

Tone

  • Does it match your product voice?

  • Is the content accessible and inclusive?

Risk

  • Is the user protected from misunderstanding?

  • Would this build or break trust?

  • Is there anything that seems legally questionable?

UX

  • Does this work in the designs or flows?

  • Does it mesh well with existing terms and functionality?

  • Does it make sense for the overall product?