AI Prompt Checker & Optimizer
Review, score, and rewrite any AI prompt with golden precision for Claude, ChatGPT, and Gemini!
An AI model does what you tell it, not what you meant. That gap — between the instruction you wrote and the one you had in your head — is where almost all disappointing output comes from. Closing it is what prompt engineering is, and it is a skill you can pick up in an afternoon and spend years refining.
What prompt engineering actually is
A prompt is the text you send an AI: a question, an instruction, a task. Prompt engineering is the practice of writing that text so the model returns what you actually want — reliably, without a dozen rounds of “no, not like that”. It is not magic words or secret phrases. It is mostly the unglamorous work of saying clearly what you want, for whom, in what shape, and with what to avoid — the things you would tell a competent stranger doing the task for the first time.
The reason it is a skill at all is that models are literal. They cannot see your audience, your deadline, the tab you have open, or the three things you tried last week that failed. Everything obvious to you is invisible to them unless you write it down. A person fills those gaps with judgement; a model fills them with the most average guess available — which is why so much AI output reads as competent and completely forgettable.
Why a good prompt matters more than people expect
When output is bad, the instinct is to blame the model or reach for a bigger one. Usually the model was capable and the prompt was underspecified. That same model, given a prompt that names the audience and shows one example, produces markedly better work — at no extra cost and with no new subscription. Prompting is the cheapest quality lever you have, and the one most people never pull.
It compounds, too. A better prompt is not just a better answer once; it is a template you reuse, a workflow that stops needing babysitting, an automation you can actually trust. The distance between a vague prompt and a precise one is the distance between a toy and a tool.
What separates a good prompt from a bad one
Good prompts tend to share a short list of traits, and bad ones tend to miss the same few. A good prompt states a concrete task with a verb. It says who the answer is for. It names the format it wants back. It shows an example rather than describing one. It separates your data from your instructions, and it says what to avoid. None of this is advanced — all of it is easy to forget, because the gaps are invisible to the person who left them.
That is exactly what this page checks for. Paste a prompt below and it runs nine checks — the nine things that most reliably break a prompt — in your browser, as you type, with nothing sent anywhere. Each result links to a short guide that shows the failure on a real prompt and fixes it. Prefer to learn by doing? The prompt library has templates that already pass every check, and the builder assembles a clean one from a few questions.
Check your promptFull review — your own chat
One click. We hand the review to Claude, ChatGPT, or Gemini, and the subscription you already pay for does the work. No API key, no cost.
Instant Structural Audit
Start typing a prompt to see instant checks run in real-time!
What the nine checks look for
Every check exists because a specific failure happens without it. They run in your browser as you type — nothing is sent anywhere, and there is no key or sign-up. Each one links to a guide that shows a real broken prompt, and the real checker failing it.
States a concrete task
A topic is not an instruction. Without a verb naming the job, the model guesses whether you wanted a summary, an essay, or a list.
Read the guide →Avoids vague language
“Good”, “engaging”, “appropriate” are judgements, not instructions. A model given nothing to aim at produces the median answer.
Read the guide →Specifies an output format
Left unsaid, models default to prose you then have to parse. One sentence naming the format saves the round trip.
Read the guide →Gives context or audience
The same question answered for a beginner and a specialist are different answers. Unstated audience is why so much AI writing reads as bland.
Read the guide →Avoids dangling references
“Fix it” — fix what? The model cannot see the tab you are looking at, and a pronoun with no antecedent is a coin flip.
Read the guide →Includes an example
One worked example pins down format, tone and edge cases more reliably than four sentences describing them.
Read the guide →Separates data from instructions
Paste a document with no delimiters and a line inside it that looks like a command may get followed.
Read the guide →Avoids over-aggressive phrasing
“YOU MUST” and “CRITICAL:” backfire. Current models follow instructions literally, so shouting makes them overtrigger rather than comply.
Read the guide →Defines what a good answer looks like
Models check their own work — but only against criteria you gave them. With no bar, the review step quietly does nothing.
Read the guide →
Then get the full review, free
The checks above are structural — they can tell you a piece is missing, but they cannot write the missing piece. For that you need a model, and you do not need to pay us for one: hand the review to the Claude, ChatGPT, or Gemini subscription you already have, and it will score the prompt and rewrite it. No API key, no cost.
Not sure what to write in the first place? Build a prompt from a few questions instead, or start from a template that already passes every check.
The techniques behind good prompts
The checks catch what is missing. These guides cover the harder judgement calls — when a technique earns its place in a prompt, and when it is just something people copy because everyone else does.
Few-shot prompting: when examples beat instructions
A worked example pins down structure and edge cases that a paragraph of description cannot. Here is how many to give, and when one is already too many.
Read the guide →Role prompting: when “act as a…” actually helps
Assigning a role is half technique, half superstition. It works when it carries real constraints the model would otherwise miss — and does nothing when it is just flattery.
Read the guide →Reasoning prompts: making the model show its work
“Think step by step” is folklore that sometimes helps and sometimes wastes tokens. The honest version: what it does, when reasoning models make it redundant, and how to ask.
Read the guide →Iterating on a prompt: fixing the part that is actually wrong
When the first output disappoints, the instinct is to rewrite everything. The faster route is to read the failure, change one thing, and know which one.
Read the guide →
Common questions
- Is the prompt checker free?
- Yes, completely. The nine checks run in your browser with no account, no API key, and no usage limit. The optional AI review hands off to a Claude, ChatGPT, or Gemini subscription you already have, so there is nothing to pay us for.
- Does my prompt get sent to a server?
- No. All nine checks run locally in your browser as you type — you can open your network tab and confirm nothing leaves the page. Your prompt is only ever sent somewhere if you explicitly hand it to an AI for the full review.
- What makes a prompt good?
- A good prompt removes the guesswork. It names the task with a verb, says who the answer is for, states the format, and shows an example. Most weak output traces to one of those being missing — which is exactly what the checker looks for.
- Which AI models does this work with?
- The checks are model-agnostic: they catch structural gaps that hurt output on any current model, including Claude, ChatGPT, and Gemini. The advice reflects how today's instruction-following models actually behave — which is why it warns against forceful phrasing like “YOU MUST”.
- Do I need to know prompt engineering to use it?
- No. Paste a prompt and the checker tells you which of the nine things is missing and links to a short guide on fixing it. If you do not have a prompt yet, the builder assembles one from a few plain questions.
- Is this only for developers?
- No. The checks apply to any prompt — writing, analysis, data extraction, or code. You pick the task type, and the checker only applies the rules relevant to it, so a blog-post prompt is not judged by rules meant for a SQL query.