What is the best free AI prompt generator?
Ask a chatbot to interview you first, then write the prompt. Paste this into Claude, ChatGPT, Gemini or Whizi: "Act as a prompt writer. I want an AI to [task] for [audience]. Ask me up to 5 questions, one at a time. Then write one finished prompt with a role, the context, numbered steps, the output format and 2 examples. Explain each part in one line."

The table lists what each option gives, taken from each tool's own page on 15 November 2026. A cell says "not stated" where the page doesn't say.
| Option | Free use | Sign-up | Paid price | Best at |
|---|---|---|---|---|
| Claude | "Limited usage" every 5 hours and every week | Not stated; the Try Claude link goes to a login page | Pro $20 a month, or $17 a month billed annually | Interview style prompt writing |
| ChatGPT | "Unlimited text chats with GPT-6" under abuse guardrails; other features limited | Not stated | Go $8, Plus $20 a month | Quick rewrites of a prompt you already have |
| Gemini | Free with a Google Account | Google Account | Google AI Pro $19.99 a month | Prompts for pictures, in an app that also makes them |
| Feedough | Unlimited generations | No login | Not stated | Templates for ChatGPT, Claude, image and video models |
| Jotform | "No limit to the number of prompts" | Not stated | Not stated | Choosing between ChatGPT, Claude, Gemini and image tools |
| Promptsera | Unlimited prompts | No login | Not stated | Image and video model targets |
| Prompt Builder | 5 free prompts (5 credits), no card | Not stated | Starter $9, Pro $19, Unlimited $49 a month | A saved library across many models |
| Whizi | 10 messages a day on Whizi AI, Auto and Claude Haiku | Free account, no card | Pro $29.99 a month, first week $1.99 | Writing a prompt on one model and testing it on another |
Whizi loses here in two ways. It has no prompt template library and no generator form, so you write the request in chat. And Feedough and Promptsera ask for no sign-up and ration nothing, which a free Whizi account can't match. Its edge is the next step: the prompt gets tested on a second model in the same thread.
Two more tools show up on page one. Taskade's prompt generator is "included on the Free plan, with no credit card required," and its Pro plan is $10 a month billed annually. RyRob's tool carries a "No Login" title, and the page text states no usage cap. Read each Jotform, Feedough and Promptsera page for your own use, since a free tool's terms change.
What do Anthropic, OpenAI and Google say a good prompt holds?
All three vendors say the same thing in different words: be specific, give context, show examples. Their docs differ on structure, and that's why one model's prompt can underperform on another.
Anthropic compares Claude to a new hire: "Think of Claude as a brilliant but new employee who lacks context on your norms and workflows." Its docs add a test: show your prompt to a colleague with minimal context on the task. In Anthropic's words, "If they'd be confused, Claude will be too." It also recommends "A few well-crafted examples" and XML tags to separate instructions from context. Read the prompting best practices. Anthropic's overview also links a metaprompt recipe in the Claude Cookbook for people who "don't have a first draft prompt."
OpenAI's prompt engineering guide says a developer message usually holds Identity, Instructions, Examples and Context, in that order. It puts context last: "This content is usually best positioned near the end of your prompt, as you may include different context for different generation requests." OpenAI's prompt generation page covers the Generate button in its Playground. Per that page, "A meta-prompt instructs the model to create a good prompt based on your task description or improve an existing one."
Google's prompting strategies page is the most blunt of the three: "We recommend to always include few-shot examples in your prompts." It adds that "Prompts without few-shot examples are likely to be less effective." Its other advice matches the others: clear instructions, the context the model needs, and iteration when the first result falls short.
That gives you a checklist any generator should pass. The prompt names the task, the audience, the format and at least one example. If a tool's output has none of those, it made a longer prompt, not a better one. For a longer walk through the four parts, see prompt engineering for beginners.
Can ChatGPT write prompts for itself?
Yes. Any chatbot can write a prompt for itself or for another model, and prompt 1 above does it in one paste. ChatGPT, Claude and Gemini all take that instruction on their free plans. The same trick covers the search for a "ChatGPT prompt generator": you don't need a separate site, you need a good request.
These four prompts write or fix prompts. Fill the brackets and paste.
- Interview first: "Act as a prompt writer. I want an AI to [task] for [audience]. Ask me up to 5 questions, one at a time. Then write one finished prompt with a role, the context, numbered steps, the output format and 2 examples. Explain each part in one line."
- Score and repair: "Here is a prompt I use: [paste]. Score it from 1 to 5 on role, context, steps, format and examples. Rewrite it to fix the two lowest scores. Show what you changed."
- Make it reusable: "I do [task] every [week or month]. Write one reusable prompt where anything that changes each time is a [placeholder] in square brackets. After it, list each placeholder with one line on how to fill it."
- Fix a bad result: "I asked: [paste your prompt]. The answer was wrong because [what was wrong]. Tell me the one sentence to add to my prompt so this doesn't repeat. Then give me the full new prompt."
Here's an example written for this page, not a tested output. A user types prompt 1 with the task "write a weekly status email for my manager." The model asks who reads it and how long it should be. After a few short answers it returns a prompt that starts "You are an assistant that writes status emails for a busy manager," lists three numbered steps, and ends with a format line: subject, three bullets, one risk. That last part is what you came for.
Which prompts work for email, meetings, study and decisions?
A work prompt works when it names the reader, the length and what to do when information is missing. These four are finished prompts, not generators. Each is specific enough to use on the first try.
- Email: "Write an email to [recipient's role] about [topic]. My goal is that they [action] by [date]. Tone: [tone]. Keep it under 120 words. Give three subject line options."
- Meeting notes: "Turn these notes into three lists: decisions, owners with dates, and open questions. Notes: [paste]. If an owner or date is missing, write 'missing' instead of guessing."
- Study: "Teach me [topic]. I already know [background]. Explain it in five steps. Then ask me three questions, one at a time, and wait for my answer before you reveal the right one."
- Decision: "I'm choosing between [option A] and [option B] for [goal]. Name the three criteria that matter most. Score each option from 1 to 5 on each, with one line of reasoning. Then argue the strongest case against your pick."
- Coding brief: "Write a prompt for a coding assistant to [task] in [language]. Include the file names, the input and output I expect, and what must not change. End with how I'll know it worked."
Prompt 6 shows a habit worth copying. Telling the model what to write when something is missing stops it from filling the gap with a guess. For email-specific prompts, prompt to summarize emails has more.
What prompts work for images, photo edits and video?
A picture prompt needs the subject, the setting, the light, the shape of the frame and a list of what to leave out. Models fill every blank you leave with the most average choice. These four work on any current image model. The last one is for a video model.
- Product photo: "A [product] on [surface], shot from [angle], [lighting], plain [color] background, no text, no logos, 4:5 portrait."
- Repeat character: "[Name], [age], [hair], wearing [outfit], standing in [place], in [art style]. Keep this exact face and outfit in every image I ask for next. Confirm you'll do that before you make the first one."
- Photo edit: "Using the photo I attached, change only [one element]. Keep the framing, the people and the lighting the same."
- Video clip: "A [number] second clip: [subject] [action] in [setting]. The camera [movement]. Mood: [mood]. One continuous shot, no cuts, no text on screen."
Gemini's free plan lists "Image generation and editing" on Google's subscription page, so prompts 10 to 12 can run there. The sample prompts are original to this page. For more photo prompt ideas, see Gemini AI photo prompts.
Why write the prompt on one model and run it on another?
A prompt written by one model carries that model's habits, and the vendors' own docs disagree on structure. Anthropic likes XML tags. OpenAI lists four labelled parts. Google wants examples every time. Running the finished prompt on a second model shows you which parts of the prompt do the work and which only fit the model that wrote it.
Whizi, which publishes this page, makes that a single thread. You switch the model per message from the picker, and the next model sees the whole conversation, so Claude can write the prompt and GPT or Gemini can run it without a paste. Switch models mid-conversation covers the details. Pro carries Claude, GPT and Gemini in one app, and its first week is $1.99. Side by side model chats on Pro and Powerhouse run one prompt on two models in two columns, each billed at its own model's credit rate.
The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, no card, which is enough to write and refine a prompt. Cross vendor testing needs Pro. Pictures a month: Free 2, Starter 20, Pro 100, Powerhouse 500. Photo edits with Nano Banana take an uploaded photo and up to 4 reference images, and keep the photo's framing. Music and video generation are Powerhouse only. Word, Excel, PDF and slide files made in chat have no plan gate.
If you only need a prompt and never test it, a free web tool or a plain chatbot is the cheaper answer. This setup earns its place when you write prompts for work and want a second opinion before you reuse one.
What mistakes ruin a generated prompt?
Five mistakes cause most of the bad results. Each one has a fix that takes under a minute.
- The task is vague, so the prompt is generic. Fix it by letting the model interview you first, with prompt 1.
- There are no examples. Google recommends always including them and Anthropic recommends a few well-crafted ones, so paste one real example of a good result.
- The output still holds leftover [brackets] or a role line like "You are a world class expert." Read the prompt once and cut every line you couldn't defend.
- You saved it after one lucky run. Run it on three real inputs first.
- You pasted private details into a free web form. This page didn't read each tool's privacy terms, so use made up names and numbers until you have.
What can an AI prompt generator not do?
It can't supply facts you didn't give it, and it can't promise the answer will be right. A generated prompt makes a request clearer. It doesn't make the model's answer true, so check any number, quote or citation the result contains.
It also can't make one prompt behave the same on every model. Anthropic, OpenAI and Google each document different structure, as the section above shows, so retest after you switch.
And pasting untrusted text into a prompt carries its own risk, because instructions hidden in that text can steer the model. What is prompt injection explains how. For how the standing instructions behind a chat work, see what is a system prompt.
- Start with a free chatbot and the interview prompt, not a new website
- Make sure every prompt names the task, the audience, the format and one example
- Delete leftover brackets and filler role lines
- Run the finished prompt on a second model
- Test on three real inputs before you save a prompt
Frequently asked questions
What is the best free AI prompt generator?
Claude, ChatGPT and Gemini are the best free choices, because each writes prompts on a free plan as of November 2026. Feedough, Jotform and Promptsera are free web tools that say they have no limit, and Prompt Builder gives 5 free prompts (5 credits). ChatGPT vs Claude vs Gemini compares the three chatbots.
Is there an AI prompt generator with no login?
Yes. Feedough says its generator is free with no login or sign-up, and Promptsera says "No Login Required," as of November 2026. RyRob's page carries a "No Login" title. Jotform's page doesn't state whether a login is needed.
Can ChatGPT generate prompts for me?
Yes, ChatGPT can write prompts for you and for other models. Paste prompt 1 from this page, answer its questions, and it returns a finished prompt. OpenAI's Playground also has a Generate button for API users that builds a prompt from a task description.
Does a prompt written by one AI work on another AI?
Usually yes, but not equally well. Anthropic, OpenAI and Google each document a different prompt structure, so a prompt may need a small edit on a new model. Run it on both and compare, then keep the version that needs the least cleanup.