Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

See the free Radar Brief

Get to Usable AI Output Faster

Workspace

This AI prompt optimizer clarifies the desired result, preserves your facts and constraints, and shows every change before you test the rewritten prompt in ChatGPT, Claude, Gemini, Cursor Agent, or any LLM.

0 / 12,000

Paste an existing prompt. The optimizer rewrites it; it does not run it.

Rewritten prompt

Your rewrite will appear here

The optimizer returns one revised prompt, a readable diff, and the reasons behind each change.

We do not store prompt text. OpenRouter requests require zero data retention and no data collection. Dictation is handled by your browser, which may send audio to its speech service; we do not receive or store that audio. Anonymous usage limits use a salted, one-way hash of your IP address. Result ratings store structured metadata, never the prompt or rewrite.

A Cursor prompt should bound the change

Select Cursor Agent to turn a rough coding request into one reviewable repository job. The Cursor prompt optimizer keeps the requested behavior visible, carries forward known paths and constraints, and makes the completion claim easier to check. It does not inspect the repository or run the task.

Rough Cursor prompt

Add severity filtering to the activity feed.

Reviewable task prompt

Implement severity filtering for the activity feed. Before editing - Inspect [relevant activity-feed paths] and the nearest tests. - Follow the repository instructions already in scope. Behavior - Add [confirmed severity options]. - Preserve the current default and ordering unless the existing tests specify otherwise. Change boundary - Edit only the implementation and tests needed for this behavior. - Do not add dependencies or refactor unrelated feed code. Verification - Run [focused test command] and the repository's required checks. Return - Summarize the behavior, files changed, commands and results, and any remaining review risk.

What the rewrite protects

  • The requested behavior stays the job.
  • Unknown paths and commands remain placeholders.
  • Unrelated cleanup stays outside the change boundary.
  • Tests and command results support the completion claim.
  • The final response gives the reviewer a short receipt.

Keep one task separate from standing rules

A Cursor Agent prompt asks for one change. Project Rules and AGENTS.md carry instructions that should survive across tasks. Mixing those jobs can turn a focused request into permanent repository policy.

The Cursor setting preserves named rules and repository evidence already present in your prompt. It will not invent a file tree, promote a temporary preference into a rule, or claim that a passing command replaces code review.

Instruction jobBest homeWhat belongs there
One feature, fix, or refactorCursor Agent promptThe outcome, relevant context, allowed scope, checks, and return receipt for this task.
Scoped project convention.cursor/rules/*.mdcInstructions selected by path, description, or an explicit rule reference.
Portable repository instructionsAGENTS.mdReadable instructions shared with coding agents that support the file.
Mandatory safety or quality gateTests, hooks, or CIA deterministic check that should fail when the agent violates the requirement.

Sources: Cursor Agent prompting and Cursor Rules. .

A rewrite you can inspect before you run it

A short prompt can hide an unsupported claim. This static example shows how the optimizer makes the instructions explicit without changing the job.

Original prompt

Write a launch email for our API tool. Make it exciting and mention it saves lots of time. Keep it short.

Revised prompt with diff

Removed · Added

Write a launch email for [target audience] about our API tool.

Make it exciting and mention it saves lots of time.

Use an energetic tone. Mention time savings only if [approved evidence] supports the claim; otherwise omit it.

Keep it short.

Keep it concise. If a firm length is required, use [confirmed word limit].

Preserved constraints

  • The deliverable remains a launch email.
  • The subject remains the same API tool.
  • The requested energetic tone remains visible.
  • Brevity is preserved without inventing a word limit.
  • The time-saved claim becomes conditional on approved evidence.

What makes the next result easier to check

A useful prompt enhancer should make the assignment easier for the model to follow and the eventual output easier for you to judge. The rewrite keeps the goal, evidence, constraints, and required format visible.

Clear outcome
States the task, audience, and intended result without changing the underlying goal.
Usable context
Organizes facts, examples, and inputs so the model can separate evidence from instructions.
Explicit constraints
Keeps length, tone, format, variables, code fences, and other requirements visible.
Reviewable output
Acts as an inspectable prompt check: additions, removals, reasons, and preserved constraints stay visible. There is no unexplained quality score.

Use the prompt optimizer in three steps

  1. 1

    Paste the working prompt

    Start with the prompt you already use, including its variables, examples, and format requirements.

  2. 2

    Choose the target

    Use Any LLM for a neutral rewrite or select ChatGPT, Claude, Gemini, or Cursor Agent.

  3. 3

    Review before copying

    Check the clean result, inspect the diff, and confirm that the preserved constraints match your intent.

What stays intact

The rewrite is instructed to preserve facts, quoted text, placeholders, variable names, code, language, examples, safety boundaries, and required output formats. Missing facts should remain visible as placeholders instead of being invented.

The optimizer is a drafting aid, not a guarantee that every model will produce the same result. Test the revised prompt against the model and workflow where you plan to use it.

Prompt optimizer questions

What does an AI prompt optimizer change?

It rewrites an existing prompt to make the goal, context, constraints, and requested output easier for a model to follow. The result includes a diff and specific change notes so you can inspect the rewrite before using it.

Does the prompt optimizer run my prompt?

No. It edits the prompt but does not execute the task inside it. You receive one optimized prompt to review and copy into the model or application you choose.

Is this also a prompt enhancer or prompt improver?

Yes. Prompt enhancer and prompt improver describe the same core job here: rewrite an existing prompt without changing its assignment. The optimizer adds a visible diff, change reasons, and preserved constraints, so you can check each edit instead of accepting an unexplained rewrite.

Does BenchLM store my prompt?

No. BenchLM does not save prompt text in its database, logs, or analytics. The request is sent through OpenRouter with zero-data-retention and no-data-collection routing enabled. A salted, one-way hash of the client IP is stored temporarily to enforce anonymous usage limits.

Can I optimize prompts for ChatGPT, Claude, Gemini, and Cursor?

Yes. Choose Any LLM for a model-neutral rewrite, or select ChatGPT / OpenAI, Claude, Gemini, or Cursor Agent. The Cursor setting structures a one-off repository task around the requested behavior, relevant context, change boundary, checks, and completion receipt before you copy it into the agent.

Does Cursor mode read my repository?

No. The Cursor prompt optimizer rewrites only the text you paste. It cannot inspect your files, infer the correct paths, or run commands. Include known paths, symbols, repository rules, and verification commands when they matter; the rewrite should leave unknown details as visible placeholders.

Should a Cursor prompt go in Project Rules or AGENTS.md?

Use the rewritten prompt for one task in Cursor Agent. Put durable repository conventions in scoped .cursor/rules/*.mdc files or AGENTS.md. The optimizer can preserve references to those files, but it does not create, install, or validate persistent rules for your project.

Is the prompt optimizer free?

The tool is free to use without an account. The generator and optimizer share 10 requests every 24 hours for each IP address, with up to 5 requests every 10 minutes.

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.