Tools · The What2Prompt engine

Prompt Optimizer: Improve Any AI Prompt (Free)

Paste a prompt, choose the AI or tool and get a score, three strengths, three concrete fixes and an improved version. This prompt optimizer is the What2Prompt rating tool, not a separate scoring shortcut. Visitors share two free tries between Generate and Rate; new accounts get five credits, and a successful signed-in rating costs one credit.

1 credit
94

A clearer brief. A stronger prompt.

Illustrative example · 62 → 94 / 100

Three small fixes can make your instructions much clearer.

  • 01 Give the AI a specific role.
  • 02 Name your audience and goal.
  • 03 Define the format and what to avoid.

Your real score and improved prompt appear here after rating.

Example prompts

Example

Check a vague writing request

Role

Act as a prompt checker, not the writer executing the request.

Context

The original prompt is: write me a YouTube script about paper bridges. The intended audience is curious beginners and the video must use original demonstrations. I have not yet supplied research notes or an exact runtime.

Task

Assess whether the prompt states a role, audience, deliverable, output format, evidence boundaries and missing inputs. Suggest three specific improvements that preserve the paper-bridge topic. Keep runtime and source details as placeholders rather than inventing them.

Format

A score out of 100, three strengths, three fixes and one revised prompt with Role, Context, Task, Format and Guardrails.

Guardrails

Do not write the finished script, invent facts about bridge strength or claim the score predicts video performance. Evaluate the instructions, not the subject’s popularity.
Example

Enhance a photo editing brief

Role

Act as a prompt improver for a bounded image edit.

Context

The original request is: make my bowl photo look professional. The source is my own unbranded ceramic bowl, and the allowed change is a cleaner background. The bowl’s shape and glaze must stay unchanged.

Task

Identify missing edit boundaries and ambiguity in the word professional. Preserve the product and specify one backdrop change, matching light and a contact shadow. Name the source image as the factual reference.

Format

Give a concise critique and a ready-to-paste image edit prompt, followed by checks for outline, glaze, camera angle and unwanted additions.

Guardrails

No invented product features, logos, named artists or real-person likenesses. Do not claim that clearer wording guarantees the model will preserve every detail.
Example

Improve a coding request with acceptance cases

Role

Act as a prompt enhancer for a coding agent.

Context

The original request is: fix my task list. The observed issue is oldest tasks appearing first after reload. Expected behavior is newest first while owner access stays unchanged. I will provide source files and two known timestamps.

Task

Preserve the ordering problem and turn it into a scoped investigation, smallest fix and regression test. Require inspection of current code before edits. Keep unknown paths as questions or placeholders and ask the agent to report tests actually run.

Format

Output the improved coding brief with explicit expected order, protected permissions and a completion checklist.

Guardrails

Do not propose a new framework, broad refactor or unverified test result. A prompt is not a permission system or proof that the resulting code is correct.

Use the rating tool as a diagnosis, not a trophy

A score is useful when it points to an actionable change. If the prompt has no output format, add the columns or sections the answer needs. If the task is vague, define the deliverable and a check for success. If context is missing, supply the actual audience or source material. Do not optimize only for a higher number. A prompt can satisfy a tidy rubric while still asking an unsuitable model to perform a task it cannot reliably complete.

Choose the target AI before submitting. An image edit needs protected visual details; a coding request needs scope and tests; a music brief needs sound and structure. The optimizer should preserve your subject and intent while adapting the instruction shape. It returns three strengths so you can retain useful parts, and three fixes so the revision stays manageable. Compare the improved prompt with the original rather than copying it just because the score looks more reassuring.

Understand the six scoring criteria

Role specificity contributes up to 15 points: the perspective should suit the work, not merely announce a famous expert persona. Context completeness contributes 20: audience, resources and relevant source material should be present when they affect the task. Task clarity contributes 25: the deliverable and useful steps should be observable. A simple image line should not become an unnecessary planning essay just to imitate a coding brief. Good scoring respects the target and the job.

Output format contributes 20 points because a correct answer still needs a usable shape. Guardrails contribute 10 for relevant boundaries such as preserved facts, safe scope or rights-sensitive content. Placeholders contribute 10 when they correctly expose missing inputs rather than conceal them with invented details. The total is 100, but it remains the engine’s judgment, not a reproducible benchmark across every model run. Check the specific critique and the rewritten text, not only the numeric result.

Improve missing information without inventing it

A prompt improver should not manufacture your audience, budget, product properties or application version. If those details matter, use a clear placeholder or a question. “For [audience]” is less polished than an invented demographic, but it is honest and easier to correct. Ask the improved prompt to distinguish source facts from style samples and assumptions. This is especially important when the original request is short and the tool has many plausible ways to fill the gaps.

After rating, replace placeholders before using the prompt downstream. Check that the improved version has not changed a photo edit into a redesign, a code repair into a refactor or a research checklist into a claim of verified research. Delete extra steps that do not serve your task. If the critique misunderstands the purpose, revise the context and rate or generate again using the shared credit rules. A longer answer is not inherently a better instruction.

Test the revised prompt on a small real input

Use an input you can evaluate manually: one source paragraph, a small fictional table or a reproducible bug. Run the revised prompt in the selected tool and compare the answer with concrete expectations. A summary should preserve the original claim; a table should use the right units; a code repair should pass the supplied regression case. For image and video, inspect protected details and continuity. The optimizer prepares the instructions, while those checks evaluate the actual result.

Keep a simple record of the original prompt, the useful fix and the observed outcome. What2Prompt saves signed-in prompt results and supports Tweak versions, but a version history is not evidence that a downstream task succeeded. Do not paste secrets or unnecessary personal information into the rating form. Visitors can use the shared free tries; ongoing signed-in ratings use one credit each after a finished result. Failed requests leave your input available for another attempt.

What makes a good prompt optimizer brief

  1. State the actual job so the rating evaluates instructions for the intended use, not an unrelated task.

  2. Pick the real target AI or name a custom tool; image, music and coding briefs need different task shapes.

  3. Give enough context to improve the prompt without inventing audience, source facts, version or product details.

  4. Require a specific output format and concrete acceptance cases when the result can be tested.

  5. Treat the score as a model-assisted diagnosis, not a benchmark, safety certificate or guarantee of the downstream answer.

Bad prompt vs What2Prompt prompt

Bad prompt

“Improve my prompt and make it perfect”

What2Prompt prompt

Role

Act as a prompt editor for a coding agent.

Context

My task list shows oldest entries first after reload; newest first is expected, with owner access unchanged.

Task

Preserve the problem, add source inspection and a narrow regression case using two timestamps, and identify missing inputs as placeholders.

Format

A scored critique with three concrete fixes and a complete revised prompt.

Guardrails

Do not execute the coding task, invent file paths or turn a score into a guarantee of safe code.

The vague request does not define what the prompt is meant to achieve. The stronger version states the failure, expected behavior and protected boundary. A prompt enhancer should preserve that intent while making missing information visible. More words are useful only when they remove ambiguity; a long persona is not automatically an improvement.

Paste only the prompt you need to assess, without credentials or private customer records. The optimizer scores the instructions and produces a revised brief; it does not execute the underlying task. Read the three fixes before copying the improved version, replace any placeholders and review protected facts. Copy your rating result before leaving: ratings do not appear in My prompts.

Reference: What2Prompt engine and five building blocks ↗

Questions about prompt optimizer

Is the prompt optimizer free?

You can copy written examples without an account. Visitors share two free engine tries between Generate and Rate, subject to the existing visitor allowance. New accounts get five credits. After signing in, one successful rating costs one credit; failed requests do not deduct credits. The Free title describes those free entry options, not unlimited rating.

How does the prompt checker score a request?

The engine uses six criteria totaling 100 points: role specificity 15, context completeness 20, task clarity with checkable steps 25, exact output format 20, relevant guardrails 10 and appropriate placeholders 10. It returns three strengths, three fixes and a complete improved prompt. Scores are model-assisted judgments, not audited performance measurements.

Will an improved prompt guarantee a better answer?

No. A clearer brief can reduce ambiguity, but the selected AI still has capability limits and may make mistakes. Inspect the rewritten prompt for changed intent or invented facts, then evaluate the downstream answer against your real goal. A rating does not verify citations, test code, clear rights or certify compliance.

What2Prompt prompt engine, with the W2P mark on cobalt.

Give your next idea a better brief.