LLM Cost Estimator
Estimate multi-model API cost from tokens or pasted prompts.
OpenCheck a prompt against five essentials and get a restructured version with the gaps filled.
Drop in the prompt you already use, however short.
See which of the five essentials were found.
Missing sections are added around your original goal.
Replace the generic lines with specifics, then copy.
The prompt your team has used for months is a single sentence, and the answers vary every time someone runs it. Before anyone starts tuning wording, it is worth checking whether the basics are even there. Prompt Optimizer runs that check and fills the obvious gaps.
Paste a prompt and the checklist updates as you type. It looks for five things: a role or persona, a task longer than a fragment, constraints or rules, examples, and an output format. Passing items turn green; missing ones are marked as added. The optimized version keeps your original text under a Goal heading and adds a Role section, a Constraints section, an Examples reminder or an Output format section only where the checklist found nothing. The status line reports how many of the five criteria your original met.
Detection is keyword based. Words like 'you are', 'avoid', 'e.g.' or 'JSON' count as signals, so a prompt can pass a check on a stray word. Treat the rewritten output as a scaffold: replace the generic Role and Examples lines with ones specific to your task. For a finer comparison between drafts, score them in Prompt Eval Rubric. If the prompt is long rather than thin, a filler-word trimmer shortens it, and Prompt Generator can rebuild it from scratch.
The analysis is plain pattern matching in your browser. Nothing is sent to a language model, so it is safe to paste internal prompts.
Keyboard shortcut
Answers for this tool. For site-wide help, open the FAQ hub.
Barely. Your text is kept as the Goal section, with any leading Markdown heading markers removed. The tool adds sections around it rather than rewriting your sentences.
Detection relies on keywords such as limit, avoid or rules. If one of those words appears for another reason, the check passes. Read the checklist as a hint, not a verdict.
One short, real input and the exact output you want for it. A single concrete pair usually steers a model more than several extra sentences of instruction.