Inside the What2Prompt engine

One idea. A playbook for the tool you chose. A prompt with a clear job to do.

Your context. Every new brief.

Tell What2Prompt what you do, who it is for and the tone you want. Your context is saved privately to your account, and you can edit or delete it in Settings or turn it off for a prompt. Tools and goals help describe your workflow without guessing facts about you.

The engine combines that context with your next idea when you choose to use it. Tweak uses your saved previous prompt and revision instruction, and saves each finished revision as a separate version.

Edit your context →

A deliberate pipeline

The engine follows these instructions when writing your prompt. Results still need review; the examples on this site are illustrations, not guarantees.

  1. 1

    Understand your goal

    Identify what you want to make, who it is for and which constraints matter. Missing information should become a question or an explicit placeholder, not a guess.

  2. 2

    Pick the playbook for the AI you chose

    Select a structure suited to the chosen model or tool. “Other — describe it” keeps your exact tool name with your idea.

  3. 3

    Add the five building blocks

    Role sets the perspective. Context supplies the background. Task defines the work. Format shapes the answer. Guardrails set boundaries. Use the blocks that actually help the task.

  4. 4

    Tune for that model’s quirks

    Separate instructions from source material for text models; keep image descriptions direct; give coding tools file scope and observable acceptance criteria.

  5. 5

    Quality check before it prints

    Check that the prompt matches the goal, uses the requested format, avoids contradictory instructions and identifies unsupported facts. This is a prompt check, not verification of the downstream AI’s answer.

Built on the official playbooks

Each AI’s playbook follows the prompting guidance its maker publishes. We translate that guidance into useful prompt structure, rather than claiming one formula works everywhere. Guidance evolves, and examples need testing against the specific model you use.

These are independent reference sources, not partnerships, endorsements or certifications.

Different AIs need different briefs

Claude

Clear sections keep instructions, context and examples distinct. XML tags can help delimit complex inputs; they are not required for every task.

ChatGPT

State the task and desired output explicitly. Give relevant context and examples; ask for uncertainty to be marked rather than hidden.

Gemini

Supply useful context and concrete examples. For multimodal tasks, say what the model should inspect in each input and how to report it.

Midjourney

A concise descriptive line: subject, setting, composition and visual style. Put supported parameters at the end; do not send a long chatbot-style instruction.

Suno

Describe genre, instrumentation, vocals and mood in the style line. Give lyrics or song structure separately, using section labels where appropriate.

Claude Code

Define the goal, relevant files, constraints and acceptance criteria. Ask for a plan before edits, then small changes and checks. Never put credentials in the brief.

What the engine won’t do

The engine is instructed to avoid generic filler, keep unknowns visible and refuse harmful requests. These safeguards are not a guarantee: review every result before using it.

A well-written prompt cannot guarantee a factual answer, working code or safe output from another AI. Review results, check sources and test code before using them.

Write my prompt — free