Bear Browns Prompt Writing Tips

Most advice on writing prompts treats it like filling out a form: be specific, give context, state your goal. This piece takes a different angle entirely, treating prompt writing as a literary act with its own craft, closer to Hemingway or Woolf than to a technical checklist. The core claim is simple: a computer needs clarity the same way a reader does, and the clearer the prompt, the better the reply.

Think of yourself as a writer

The starting premise is that you already have the skills this requires. You read, you know what makes prose land, you know the difference between telling someone something and showing them. Prompt writing asks you to apply that same instinct to a different kind of reader: the computer on the other end of the exchange. That reader is not human, but it responds to the same qualities good writing has always rewarded, clarity, structure, and intent, dressed up here as "a dance of language between human and AI."

Show, don't tell

The piece leans hard on this classic writing principle. Rather than telling a model what you want in abstract terms, you show it: give an example, model the pattern you're after, and let the model extend it. This connects directly to iteration. You draft a prompt, test what comes back, and refine it, the same way a writer revises a line until it says exactly what they meant.

Three patterns worth knowing

The piece names three concrete prompt patterns, each a different tool for a different job:

The persona pattern. You assign the model a role directly: "you are an expert." This sets expectations for tone, vocabulary, and depth before the actual request even arrives.

The template pattern. You give the model an explicit structure to fill: first X, then Y, conclude with Z. This is useful when you need a predictable shape in the output rather than an open-ended response.

The alternatives pattern. You ask the model to consider multiple angles rather than commit to a single answer immediately, three angles to review, for instance. This is the pattern to reach for when you want breadth before you narrow down to one direction.

Each pattern is a tool, and each prompt built with one is a chance to get a better result than a vague, unstructured request would produce.

Identify the goal first

Before choosing a pattern, the piece insists on a prior step: know what response you're actually trying to get. What are you seeking? That question has to be answered before you can pick the right framework, because the persona, template, and alternatives patterns solve different problems, and choosing blindly wastes the advantage any of them offers.

Test, refine, and treat it as a draft

Nothing here is presented as one-shot. The advice is to test a prompt, then refine it, line by line, the way an editor revises a manuscript. That means debugging your approach the same way you would debug a written piece that isn't landing: figure out where the response missed what you meant, and adjust the prompt rather than assuming the model is simply wrong.

Key takeaways

• Treat prompt writing as a form of writing craft, not just instruction-giving; clarity and structure carry the same weight they do in prose.
• The persona pattern assigns the model a role and expertise level before you state the request.
• The template pattern gives the model an explicit structure: first X, then Y, conclude Z.
• The alternatives pattern asks for multiple angles instead of a single answer, useful when you want breadth before narrowing down.
• Show the model examples rather than only describing what you want; let the example do the teaching.
• Identify your goal before choosing a pattern, and treat every prompt as a draft to test and refine.

Try it yourself

Pick a task you regularly hand to an AI tool, and rewrite your usual prompt using one of the three patterns above: assign a persona, build a template, or ask for alternatives. This piece comes from Studio Sessions with Bear and Humanitarians AI, and it's aimed at anyone who writes prompts often enough that a little more craft would actually pay off.