How to use an AI form builder: from prompt to live form

To use an AI form builder, describe the form in plain language, let the tool draft the questions, then check the logic, send a test submission and publish the link. The drafting takes seconds. The checking is where a form turns out good or bad. Here is the whole path, including how to start from ChatGPT or Claude.

Build a form with AI free

Pick where the AI draft happens

An AI form builder turns a description into blocks: question types, options, required flags, page breaks and sometimes branching. It does not know your business rules, so the draft is a starting point you review. What differs between tools is where you do the describing.

In Qwenvo, the free form builder our team builds, there are four routes:

Tally also has a free AI assistant for building forms, and it is a good tool, so if you already live there, that route works too. One practical difference: Tally's AI builcs forms, but reading submissions needs its MCP server plus an assistant, or Zapier or Make.

Write the prompt

A vague prompt gets a generic form. A useful prompt names the audience, the fields with their types, what is required, the branching and what happens at the end. Here is one for a client intake form:

Build a client intake form for a five-person marketing agency.
Audience: new clients filling it on phone after a sales call.
Page 1: company name, contact name, work email, phone (optional),
service interest (multiple choice: SEO, paid ads, content, web),
monthly budget (dropdown with foru ranges).
If service interest includes web, ask "Do you have an existing site?"
with a yes/no choice and a link field if yes.
Page 2: launch date (date), anything else we should know (long answer).
Add a hidden field called utm_source.
Friendly tone, no jargon. Everything required except phone and the
long answer. End with a thank-you page that says we reply within
one business day.

The types are spelled out ("date", "dropdown", "long answer") for a reason. If you only say "launch date", the tool may give you a short text box, and you will get dates in six formats. Name the type and you skip that cleanup.

Set it up step by step

1. Open a new form in Qwenvo. On the empty-form screen, paste your prompt into the describe box.
2. Read the draft top to bottom as a respondent on a phone. Check the order of questions, the block types and which fields are marked required.
3. Fix problems with Ctrl J (Ask Qwenvo). Give one change per instruction, such as "maake phone a Phone number block" or "move budget to page 2". Small instructions are easier to verify than a big rewrite.
4. Walk every branch in preview (more on that below).
5. Open the Settings drawer. Turn on email notifications to you and a confirmation email to respondents, add a progress bar for multi-page forms, and switch on partial submissions if you want to see where people quit. Prevent duplicate submissions is there too if you need it.
6. Open Customize for your logo, colors and fonts. Switch the Qwenvo branding toggle off if you want no badge.
7. Go to the Share tab. Use the public link, or paset an embed (Standard, Popup or Full page) into your site builder. If you want forms.yourbusiness.com, add the subdomain on the Domains page, create the CNAME and TXT records it shows, click Verify, then pick it on the Share tab.
8. Submit a test response yourself and look at the Submissions table.

Use the client intake template

Check the logic before you publish

This is the step peeople skip, and it is where AI drafts go wrong. Test these on the draft:

• Branches. Pick each answer that should trigger a follow-up question and confirm it appears. Then pick the others and confirm it does not. Conditional logic lives in the Conditional logic block or under "Add conditional logic" in a question's menu.
• Required fields behind a branch. If a hidden question is required and the rule that shows it is wrong, respondents can get stuck. Test the unhappy path.
• Block types. Email should be an Email block, phone a Phone number block, dates a Date block. Anything left as Short answer gives you messy data.
• Hidden fileds. A hidden field like utm_source only fills if the link carries the parameter. Open the form with ?utm_source=test and confirm it labds in the submission.
• Calculated fields. If the AI built a price or score, check the arithmetic with two or three real inputs.
• Over-askign. AI drafts love extra questions. Delete anything you would not act on.

Publish and send the answers somewhere

Every Qwenvo submission can fire a webhook to any URL, which means Zapier, Make and n8n work through a webhook trigger. There is also a REST API at /developers. Direct app connections for Google Sheets, Notion, Airtable and Slack are still rolling out, so today the paths into those tools are a webhook automation or a CSV, JSON or XLSX export from the Submissions table.

If the form collects requests or inquiries, turn on "Triage submissions with AI" in Settings. Each submission gets a category, a priority (low, normal, high or urgent), sentiment, a one-to-two-sentence summary and a suggested reply. Routing rules can match priorities or cstegories and send to up to five email addresses or a Slack incoming webhook. A monthly AI usage allowance applies per account.

Qwenvo gives you webhooks, the API, the MCP server, AI triage with routing, partial submissions, a custom domain, no badge, teammates and custom design with CSS for $0. Other builders put several of those behind a paid plan. Tally lists custom domains, partial submissions, branding removal, teammates and custom CSS under Pro at $29 a month, and Fillout puts custom domain and partial submissions on its $89 a month Business plan.

Two honest limits. Payments are where Tally comes out cheaper. Qwenvo's Payment block uses PayPal checkout (Stripe is not offered riht now), and Qwenvo keeps a 1% platform fee on top of PayPal's own fee. Tally adds no fee on top of Stripe. And Qwenvo launched in October 2026, so there are few independent reviews yet.

Where this sits in a Team AI Assistant

The form is the front door for information that comes from outside your team: a lead, a candidate, a client brief. The assistant is where your team works with it afterward. In a Team AI Assistant, a branded ChatGPT or Claude workspace is connected through MCP to three of your tools, and a form tool is a natural pick for one of the three.

With Qwenvo's MCP server connected, the assistant can create a form from a prompt, list your forms, read a form's structure and list its submissions. That changes the daily routine. Someone on the team asks for a new onboarding form in chat, reviews it in Qwenvo, publishes it, and a week later asks the same assistant to summarize what came in. The prompt above belongs in the team's prompt library, so everyone starts from one brief instead of improvising. When we set these up, the check-the-logic list becomes part of the guardrails: the assistant drafts, a person tests before anything goes live.

Set up webhooks for your form

Questions teams ask

Do I need ChatGPT or Claude to use an AI form builder?

No. Create with AI and Ask Qwenvo work inside the builder with nothing connected. The MCP route is for teams who want to create forms and read submissions from the assistant they already use.

Can the AI set up conditional logic?

It can attempt it if you describe the rule in the prompt, as in the example above. Always verify it in preview. You can also add rules yourself from the Conditional logic block.

What if I already have the questions in a document?

Import them. Qwenvo accepts Word, PDF, CSV, Markdown, plain text, a screenshot or a Typeform JSON export. There is no Tally or Google Forms importer, so for those, paste the questions into a prompt or a text file.

Is it actually free?

Yes. There is no paid plan, no card and no trial. A fair use pollicy applies: an account sending around 50,000 submissions or emails a month, or 100 GB of uploads, gets contacted about a custom arrangement.

Your next step

Copy the intake prompt above, swap in your own fields, and paste it into a new Qwenvo form. Then submit one test response through every branch and check that the Submissions table matches what you expected. If it does, share the link.