How to Build an AI Lead Capture Form, Field by Field

Short answer: an AI lead capture form is four to six fields, generated in seconds from a description of your business, then edited for about ten minutes to remove what the draft over-collected — and it is not finished until notification and automatic reply are wired. As of August 2026 the generation part is the easy half. This guide covers the half that decides whether the form produces leads you can act on.

Start from the response, not the fields

The usual way people build a lead form is to open a builder and start adding the fields that seem relevant. That produces a form which collects information nobody uses.

Work backwards instead. Write down, in one sentence, what you do when a lead arrives. For a mobile service business it is usually something like: call them within ten minutes, confirm they are in our area, quote a range, and book a slot.

Now every field on the form has to serve that sentence. To call them you need a phone number. To confirm the area you need a location. To quote a range you need to know roughly what the job is. That is three fields plus a name, and you have a form.

The test for any additional field is specific: name the different action you take depending on the answer. If a "how did you hear about us?" dropdown returns "Google" or "a friend" and you do the exact same thing either way, it is not a qualifying question — it is a reporting question, and reporting questions belong in analytics rather than on the critical path of a stranger's first contact with you.

What to hand the AI generator

Describing the business is more productive than describing the form. Generators produce noticeably better drafts from a sentence about who fills this in and why than from a list of fields, because the field list is the thing you want it to work out.

A weak prompt: "contact form with name, email, phone, message."

A better one: "Intake form for a mobile locksmith. Customers are usually stranded and using a phone. I need to know how to reach them, where they are, what vehicle or lock, and whether it is urgent."

The second gives the generator the constraints that matter — mobile context, urgency, the fact that the person is under stress. Drafts from that prompt come back shorter and with better field types, because "usually on a phone and stranded" is the kind of instruction that changes design decisions.

BotForms generates from exactly this kind of description, and can also rebuild an existing paper form from a photo or PDF rather than a prompt, which is the better route if you already have a document your staff trust. That path is covered separately in converting a PDF to an online form.

The ten minutes of editing that always follow

Generated drafts fail in the same three ways, consistently enough to be a checklist. We wrote these up in more detail in what AI can and cannot do with form questions, but the short version:

It over-collects. Company name, job title, a second contact method, preferred contact time. Each is defensible in isolation and collectively they turn a four-field form into nine. Delete anything that fails the different-action test.

Validation is stricter than reality. Phone patterns that reject a legitimate format, postcode rules that reject a rural address, a required dropdown with no option that fits. Every one of these is an invisible lost lead — the visitor sees an error they cannot resolve and leaves, and you never learn it happened.

Too much is marked required. Drafts mark nearly everything required because that produces complete data. It also produces fewer submissions, and a partial lead you can call is worth more than a complete lead you never received. Required should mean I cannot respond without this. In practice that is contact method, and usually nothing else.

The field order that works

Order matters more than field count on a short form, because it determines how far someone gets before the first thing that makes them hesitate.

Position	Field	Required	Why here
1	Name	No	Trivial to answer, builds momentum, useful even if they stop
2	Phone or email	Yes	The one you will actually respond on
3	What they need	No	Free text or a short list; shapes your first sentence back
4	Where they are	No	Service-area check before you promise anything
5	Urgency or timing	No	The qualifying question, after they are invested
6	Anything else	No	Optional catch-all; the useful detail often lands here

Two things about this table are deliberate. Only one field is required, and the qualifying question sits near the end rather than the start.

Both come from the same observation: a form is abandoned at the field the visitor does not want to answer, and everything above that field is still captured only if the form saves progressively or the earlier fields were the easy ones. Put the friction late. The mechanics of where people actually leave are covered in reducing form abandonment.

Wire the two automatic messages before you publish

A lead form that only stores submissions is a database, not a lead system. Two messages have to fire on submit.

The notification to you. Speed is the variable that most reliably predicts whether a lead converts, and it is entirely within your control. Email works when someone is at a desk; for field teams it usually does not, because email is checked between jobs and the enquiry has aged an hour by then. SMS costs more and gets read. The tradeoff, with real numbers, is in speed to lead: SMS versus email notifications.

The acknowledgement to them. This is the one that gets skipped, and skipping it is expensive. Someone who submits a form and sees a bare "thanks" has no idea whether you received it, whether you are open, or when you might reply — so a meaningful share of them go and fill in the next form on the next site. An automatic reply that names a timeframe ("we call back within an hour during business hours; we are open until 7pm today") stops that.

On BotForms both are configured on the form itself. The submitter's automatic welcome email can carry an attachment, which is how the same mechanism doubles as lead magnet delivery when the form is a download rather than an enquiry.

Keeping the junk out without punishing real visitors

A lead form on a public page will be found by scrapers, usually within weeks of publishing. The cost is not really the nuisance of deleting messages — it is that bot submissions inflate your conversion numerator, so a page that is genuinely underperforming reads as acceptable and nobody investigates it.

The defences that cost real visitors nothing are a hidden honeypot field, a minimum time-to-submit threshold, and server-side content filtering. Those three catch most automated traffic. A visible CAPTCHA is a last resort, because it demonstrably costs genuine submissions. The full ordering is in stopping contact form spam without a CAPTCHA.

Spam protection is included on every BotForms plan, which is the shape you want — filtering that is a line item you can forget to buy tends to be bought after the first bad week rather than before.

What it costs to run

Published plans, as of August 2026:

• Starter, $29/month ($23 billed annually): 1 user, unlimited forms, 1,000 submissions a month, 100 AI-builder runs, email notifications, PDF generation, spam protection, and an email audience of 500 contacts with an automatic welcome email.
• Pro, $79/month ($63 annually, $49 on the founder price): 3 users, 10,000 submissions, 1,000 AI runs, unlimited PDF tracking, five team collectors, a 5,000-contact audience with unlimited campaign sends, and API access.
• Business, $199/month ($159 annually): 10 users, 50,000 submissions, unlimited collectors, a 25,000-contact audience and white-label PDFs.

Every plan runs a 30-day trial and there is no free tier. SMS is a separate add-on — $29 for 1,000 messages a month, $105 for 5,000 with two-way conversations, $420 for 20,000 — because most businesses do not need it and should not be paying for it inside the base price.

For a solo operator the number worth comparing is not $29 against a free tier elsewhere but $29 against the stack it replaces. BotForms' own published comparison on the homepage puts Formspree Gold at $20, Mailchimp Essentials at $13 and Zapier at $20 to connect them — $53 a month, as listed in August 2026, for a setup that still has three places to check when something breaks. If price is the deciding factor, the cheaper-end analysis is in the cheapest AI form builder for a small business.

A one-hour build

Realistic sequence, start to published:

1. Ten minutes. Write the one-sentence description of what you do when a lead arrives. Generate the form from a description of the business, not a field list.
2. Ten minutes. Run the three-part edit: delete over-collected fields, loosen validation, unmark required.
3. Ten minutes. Set the notification channel and write the automatic reply. Name a timeframe in it.
4. Ten minutes. Fill the form in yourself, on a phone, on mobile data. This finds more problems than any other single step — a field that is fine on a laptop and miserable on a phone is the usual discovery.
5. Ten minutes. Embed it and check it renders in the real page rather than the preview. The embed specifics for the common platforms are in embedding a form in WordPress, Webflow or Squarespace.
6. Ten minutes. Leave it. Come back in two weeks and look at where people stopped, which is the only evidence that beats an opinion. What to look at is in the form metrics that matter.

The part people get wrong

Almost every underperforming lead form we have looked at fails on one of two things, and neither is the form builder.

The first is asking for something the visitor is not ready to give at that stage of the relationship — a budget range on a first contact, a company size from a sole trader, a phone number from someone who wanted an email answer. The exchange is not worth what you asked for.

The second is silence after submit. The form worked, the lead arrived, and nothing came back to the person who filled it in, so they kept shopping.

Both are fixable in under an hour and neither requires a better tool. Which is worth knowing before you spend a weekend evaluating platforms: the generation is a commodity now, and the difference between a form that produces bookings and one that produces a folder of unread submissions is almost entirely in the fifteen minutes after the draft appears.

Start a 30-day trial and build one from a description, or look at the feature list and published plans first.