PromptSmith

You type one sentence. PromptSmith builds a complete, ready-to-paste prompt for every AI model you tick, all at once, without sending a word of it anywhere.

• Reads your sentence and detects the task (explain, research, write, summarize, analyze or compare, extract data, code, plan, or general) and shows what it detected, so the role and rules fit the job. Each task type adds its own rules: for code, never invent libraries, functions or API endpoints; for summaries, add nothing that is not in the original text; for data extraction, output null instead of guessing. A word count and rough token count update as you type.
• Writes the prompt in parts: a role (for example “expert teacher who explains accurately”), your task, the accuracy rules, a step-by-step process, and the output format. Nothing is left for the AI to guess.
• Sets how hard it pushes against made-up answers with a three-way switch. Standard: say “I'm not sure” or “I don't know” instead of guessing, never invent facts, names, numbers, dates, quotes, links, citations or statistics, and state assumptions openly when the request is unclear. Strict (the default) adds: keep facts separate from inference, cite only sources it can name, flag anything outdated or after its knowledge cutoff, and re-read the answer and remove every claim it cannot support. Maximum adds: mark each key claim high, medium or low confidence and leave out anything below medium, prefer a shorter correct answer over a longer uncertain one, and point out false or unverifiable premises before answering.
• Shapes the answer with dropdowns for task type, answer format (table, bullet points, step-by-step and more), length, aspect ratio for image and video models, audience (for example “busy managers”) and tone, plus checkboxes to ask up to three clarifying questions first, label claims with a confidence level, require a source (title and date) for every factual claim or the words “no source”, and add a self-check pass.
• Adds eight optional guardrails: never invent links or web addresses; write today's date and warn about the knowledge cutoff; show calculations step by step and re-check each number; say plainly when a question cannot be answered; include the strongest counterargument; quote exact words instead of paraphrasing the source; use plain language and define jargon; stay on topic and skip unrequested extras.
• Lets the AI use only your material. Paste an article, notes or data into Source material and the prompt wraps it in tags and tells the AI to answer only from it, to quote short supporting passages, and to reply “Not found in the provided material” when the answer is not there.
• Formats the same request differently for each model you tick, side by side. Claude gets tagged sections (<role>, <task>, <source_material>, <accuracy_rules>, <answer>); Gemini gets context first, then the task, then constraints, with the question restated at the end; ChatGPT and similar chatbots get clear Markdown headings; image and video generators get scene-style prompts with aspect ratio. Pick from 130 models in six groups: chatbots and language models, search and research assistants, coding assistants and app builders, open-source and small models, 16 image generators and 9 video generators.
• Scores every prompt out of 100. It checks for a role, a clear task, permission to say “I don't know,” an evidence rule, an output format and a self-check, then lists in plain words what is missing and how to fix it.
• Improves prompts you already have. Switch to Improve my prompt, paste your old prompt, optionally say what to change, and get rebuilt versions that show the old score and the new score (for example 0/100 to 100/100).
• Starts from 14 templates with {blanks}. Pick one, and a Fill in your blanks box appears with one field per blank.
• Remembers your standing instructions (for example “I am a nurse. Use US spelling.”) and adds them to every prompt.
• Saves everything. The last 30 prompts are kept automatically; star or save the ones you want to keep (up to 500); search them; export the whole library to a file and import it on another computer. Save your favorite combinations of models as named sets.
• Copies in one click, or copies and opens Claude, ChatGPT, Gemini or Perplexity in your browser. Keyboard shortcuts: Ctrl/Cmd+Enter to generate, Ctrl/Cmd+S to save, Ctrl/Cmd+Shift+C to copy. Light, dark or automatic theme.
• Works on your computer, not in the cloud. Windows 10/11, Mac (Apple Silicon and Intel) and Android. No account, no analytics. Your text never leaves your machine; only your license key is checked online.

Competitor details come from each company's public website and published reviews, checked October 9, 2026. “Not advertised” or “Not listed” means we did not find it on their pricing or feature pages; it does not prove a tool lacks it. Features and prices change, so please confirm on their sites. PromptSmith writes prompts; it does not run them or guarantee any AI's answers. PromptPerfect, another well-known tool, is not shown because its status is unclear.