Skip to content
All posts

18 min read

AI Chatbot vs Contact Form: Which Captures Better Leads?

Chatbots and contact forms fail in different places. The real evidence, dated and sourced, plus a 3-question test for which one your site needs.

ConGreeto logo – blue chat bubble with two eyes, author avatar for the ConGreeto blogConGreeto TeamThe team behind ConGreeto, an AI sales assistant and sales pipeline in one, built by Brillnex Systems.
Contact form versus AI chatbot: an abandoned form keeps no fields, an abandoned chat keeps the name and email already typed.
In two 2025 field experiments, a simple rule-based chatbot produced 42 leads to a landing page form's seven. Where it stops working matters as much.

A contact form fails at the submit button. A chatbot fails at the first message.

Most pages comparing chatbots with contact forms ask which one "converts better", then answer with a number nobody can trace. The more useful question is: where does each one lose people, and what do you have left when it does?

A contact form is all-or-nothing. A visitor fills in four of five fields, hesitates at the phone number, and closes the tab. You get nothing, and never learn they existed.

A chatbot fails earlier and more visibly. Some visitors see a chat bubble, decide they will not talk to a bot, and leave without typing a word. But the ones who do talk hand over information a piece at a time, and a well-built assistant keeps each piece as it arrives. A conversation abandoned at message six can still leave you a name, an email and the question that brought them.

Choosing between them is a decision about which failure you can afford.

First, the numbers you will see elsewhere

The same figures come up on every page about this topic. We tried to trace each one.

  • "Contact forms convert at 1–3%; AI chatbots convert at 10–15%." This is the headline number on the page currently ranking first for this search. It cites no source, and we could not find a study behind it.
  • "81% of people abandon contact forms." It is variously credited to Baymard Institute, to The Manifest in 2018, and to a "Salesforce and Clearbit study of 6.8 million contact form sessions". Baymard's published research is about e-commerce checkouts, not contact forms, and we could not find the Salesforce/Clearbit study at all.
  • "Forms complete at 9%, conversations at 44% — the gap hit 4x in 2026." This comes from a vendor's "cross-vendor benchmark" of about 100 SaaS funnels. It gives no sample sizes, no selection criteria and no method you could check. It may be true. As published, it cannot be verified.
  • "73% of consumers say live chat is the most satisfying channel." Usually undated, it describes live chat with humans, not AI chatbots, and we could not find a current primary source for it.

CueFully source-checked form abandonment figures in August 2026 and flagged several more.

Here is what holds up.

What the evidence actually says

How often people finish a form

The best large-scale data on form behaviour comes from Zuko, a form analytics company. Its published benchmarks (2025 data, from forms run by its customers):

  • 45% of people who view a form complete it
  • 66% of people who start a form, by touching at least one field, complete it
  • Desktop view-to-completion is 47%, mobile 42%
  • The email field has a 6.4% abandonment rate and the phone field 6.3%. Among common fields, only the password field is worse

Two caveats: Zuko's customers mostly run longer transactional forms (applications, quotes, checkouts), not contact forms, and the figures ignore visitors who never reach the form at all. One point carries straight across, though: the fields that cost you the most people are the ones you most want.

The closest thing to a head-to-head test

The most rigorous published comparison we found is a 2025 paper in the Journal of Business Research, From static to conversational, by Giuliana Isabella and colleagues at Insper in Brazil. They ran two B2B field experiments with more than 16,000 people. Ad clicks went either to a conventional landing page with a form or to a WhatsApp chatbot.

As reported in September 2026:

Experiment

Traffic

Chatbot

Landing page form

1 — general leads (Facebook / Instagram ads)

9,500+ clicks

42 completed conversions

7

2 — qualified leads: company, contact name, phone (Google ads)

~6,800 clicks

18 leads

5

Three things matter more than the headline. The absolute numbers are small: neither channel converted cold ad traffic well, so anyone promising a 15% chatbot conversion rate is not quoting this. The bot was rule-based, a scripted WhatsApp flow, so the advantage came from the conversational format itself, not from AI. And the authors say where it stops working: the effect was moderated by purchase complexity, the buyer's desire for control and local business culture, and it shrank for large, multi-stakeholder purchases.

Does conversation produce better information?

This is the lead-quality question, and survey researchers have been testing it for years, because a survey is a form by another name.

  • CHI 2019 — Kim, Lee and Gweon compared chatbot and web surveys and found chatbot respondents were less likely to satisfice: less likely to give lazy, undifferentiated answers just to get through. A casual conversational style helped further, but only in the chatbot.
  • ACM TOCHI, 2020 — Xiao and colleagues at IBM Research analysed 5,200+ free-text responses. An AI chatbot that asked open questions and probed vague answers got significantly more informative, relevant, specific and clear responses than a standard survey.
  • Bulletin of Sociological Methodology, 2026 — the counterweight. Çavuşoğlu Deveci, Fuchs and Metzler ran a fresh experiment and found respondents rated the chatbot-style design more original and entertaining, but harder to navigate. They found no significant difference in response time or answer length.

The honest reading: a conversation draws out richer answers when it actually converses, probing and following up. A form restyled as speech bubbles gains novelty and loses usability. If your "chatbot" is a form in a costume, expect the 2026 result.

Where each one fails: a side-by-side

Contact form

AI chatbot

Where it loses people

At the fields, especially email and phone, and at submit

At the first message, from people who will not talk to a bot

What you keep when someone leaves halfway

Nothing (unless you run partial-capture tooling)

Whatever they typed, if the bot saves details as they come up

What you learn

The fields you asked for

Their actual question, timeline, objections, and the words they use

Answers the visitor's question first?

No: it asks before it gives

Yes, if it is grounded in your real content

Response time

Whenever someone reads the inbox

Seconds, at any hour

Disclosure cost

None: nobody distrusts a form for being a form

Real and measurable (see below)

Language

Whatever language you built it in

Can reply in the visitor's language

Accessibility and control

Predictable, scannable, easy to review before sending

Harder to navigate for some people (2026 study above)

Structured data (files, account numbers, long applications)

Strong

Weak

How it fails badly

Silently: you never know who left

Loudly: a wrong answer, said with confidence

The last row is the one people skip. A form fails in private; a chatbot that gives the wrong opening hours fails in front of the customer. Grounding the assistant in your real content is the precondition for every other row.

Lead quality: fields versus intent

A form gives you what you asked for. If you asked for name, email, phone and "message", you get a name, an email, a phone number and a message that often says "please call me".

A conversation gives you what the visitor volunteered while trying to get an answer. That is a different kind of data. Someone asking "do you do emergency call-outs on Sundays, and roughly what does that cost?" has told you their urgency, their timeline and their price sensitivity before you asked for anything. A form would have captured the same person as a phone number.

HubSpot's 2026 State of Marketing report found 40% of marketers name lead quality as their single most important metric, more than any other. If quality is the measure, the comparison is not "how many submitted" but "how much did we know about each one before the first call".

Partial capture: the half-finished lead

This is the gap almost every comparison misses.

A form commits at submit; everything before that disappears when the visitor leaves. A conversation has many small commits. If a visitor gives their name at message two and their email at message four, then leaves at message six because their bus arrived, a well-built assistant still has both, plus the question they asked. The unit of capture is the detail, not the submission.

That only works if the assistant saves details during the chat. Many build the lead record only when the conversation formally ends, which recreates the form's all-or-nothing problem inside a chat window. Ask any vendor: if the visitor leaves mid-conversation, what do I get?

The disclosure problem

The research every chatbot vendor would rather you skipped: in a randomised field experiment on more than 6,200 customers, published in Marketing Science in 2019 (Luo, Tong, Fang & Qu), undisclosed chatbots sold as well as proficient human agents. But disclosing up front that the customer was talking to a bot cut purchase rates by 79.7%.

The general mood has not softened since. SurveyMonkey's December 2025 survey of 2,017 US adults found 79% strongly prefer human interaction and 89% want a human option always available. Pew's June 2026 report (5,119 US adults, surveyed February 2026) found 71% expect AI to make personal information less secure, which matters when the bot's job is to ask for an email address.

A form pays none of this cost. The wrong conclusion is to hide the bot: the trust cost of being caught is total. The right one is that a disclosed assistant has to earn the details it asks for, answering first and asking second.

Adam and Benlian, in the Information Systems Journal (2024), tested chatbots against web forms in a randomised field experiment (n=386) and a follow-up online experiment (n=182). They also tested "reciprocation triggers", where the bot gives useful information before asking for any. Dialogue design changed disclosure through two levers: how present the other side feels, and how worried people are about privacy. A form pulls neither lever. A chatbot pulls both, in one direction or the other.

Language: the form nobody mentions

A contact form in English is a closed door to a visitor who does not read English comfortably.

CSA Research surveyed 8,709 consumers in 29 countries between November 2019 and January 2020. 76% preferred to buy products with information in their own language, and 40% said they would never buy from websites in other languages. The study is from 2020, and we label it as such, but it is still the most-cited primary dataset we could find.

Translating a form is a project. An assistant that replies in the visitor's language opens the door.

Why this matters more in 2026

Pew's 2026 survey found 49% of US adults now use AI chatbots, up from about a third in 2024, and 60% read AI search summaries. Seer Interactive's April 2026 update, which covers 53 brands, 5.47 million queries and 2.43 billion impressions, found organic click-through on queries showing a Google AI Overview fell to 1.3% in December 2025 before recovering to 2.4% in February 2026. Seer warns against reading a trend into two months.

Fewer people click through, and those who do have already had the generic answer. They arrive with the follow-up the summary could not answer: does this apply to me, what would it cost me. Ahrefs saw it in June 2025: AI-search visitors were 0.5% of its traffic but 12.1% of signups.

A form asks that visitor to write the question down and wait, and in RevenueHero's March 2024 test 63.5% of 1,000 B2B companies never replied to a demo request at all. We covered that in Speed to Lead in 2026: The Real Benchmarks.

Keep your form if…

We build an AI chatbot. Here is when we would tell you not to rely on one.

  • Your visitors bring paperwork, not questions. File uploads, insurance details, account numbers, multi-page applications. A conversation is a bad way to collect forty structured fields.
  • Your purchase is complex and committee-driven. The Journal of Business Research study found the chatbot advantage shrank for complex, high-investment B2B purchases. Your form, or better a direct booking link, is fine there.
  • Your audience wants control. Some visitors want to see every field, review what they wrote and press send. The 2026 survey study found chat designs harder to navigate. Forcing those people into a chat costs you their goodwill.
  • Nobody arrives with questions. If people know exactly what they want ("table for four, Friday"), a short form or booking widget beats any conversation.
  • You cannot ground the bot. If you are not willing to give it your real prices, hours and policies, it will make them up. A form that captures nothing is better than an assistant that misinforms.
  • Accessibility is non-negotiable for your users. A well-built form with proper labels is a known, testable quantity. Check any chat widget with a screen reader before you remove the alternative.

Even then the answer is usually "form available", not "form only".

The Fail-Point Test: three questions to decide

1. What does a visitor arrive holding: a question or paperwork? Questions ("can you do it this week?", "what would it cost for my house?") belong in a conversation, because the answer is what moves them forward. Paperwork belongs in a form.

2. What is half a lead worth to you? If a name and email without a phone number is still worth a follow-up, as it is for most service businesses, you need a channel that keeps partial details. If an incomplete enquiry is useless (a regulated application, say), the form's all-or-nothing model costs you little.

3. Who can your current form not serve? Visitors at 11pm who want an answer now, visitors who do not read your language, visitors on a phone who will not type a long message. That is the gap a chat could close.

Read the answers together:

If most visitors bring…

…and half a lead is…

Put first

Keep as fallback

Questions

Valuable

Grounded AI chat

Short form

Questions

Worthless

Chat that answers, then hands over to the form

Form

Paperwork

Valuable

Form, with a chat for questions about the form

—

Paperwork

Worthless

Form

Chat optional

One sentence to take away: put the conversation where visitors have questions, and the form where they have paperwork. On most small business sites, that means both.

How ConGreeto handles this

ConGreeto is an AI assistant built for the top row of that table: visitors who arrive with questions, on sites where half a lead is still worth calling.

It answers before it asks. Answers come from your indexed website and documents, so it quotes your real prices and hours. Something useful first, then the request.

It captures details as they come up. Name, email and phone are saved to the lead during the chat, as the visitor types them. It asks at a natural pace (a name early, contact details a few messages in) and only fills blanks, so it never overwrites something your team corrected. If the visitor leaves at message six, you keep what they shared.

It scores what it learned. Conversations, including ones the visitor walked away from, are summarised and scored from what was actually said.

It replies in the visitor's language. You choose which languages your assistant supports, and it answers in the one the visitor writes in.

It is honest about what it is: an assistant, not an imitation of your salesperson. Given the Marketing Science finding, that is a deliberate trade.

It does not replace your form. Keep it. ConGreeto handles the visitors your form was never going to catch.

If you want the product reasoning behind it, it is in AI Chatbot for Lead Generation: Why We Built ConGreeto. Installation is one script tag, and plans are on the pricing page.

Try ConGreeto on your site →

Frequently asked questions

Do chatbots convert better than contact forms? Sometimes, and not by the margins usually quoted. The most rigorous recent comparison, two B2B field experiments with 16,000+ people published in the Journal of Business Research in 2025, found a simple rule-based chatbot produced 42 completed conversions against 7 for a landing page form, and 18 qualified leads against 5. The absolute rates were low for both, and the advantage shrank for complex, high-investment purchases.

What is the average contact form conversion rate? There is no credible published benchmark for contact forms specifically. The commonly quoted "1–3%" has no traceable source. The best form data available is Zuko's (2025): 45% of people who view a form complete it, and 66% of those who start one finish. Those figures come mostly from longer transactional forms, and they exclude visitors who never reach the form at all.

Should I replace my contact form with a chatbot? Usually not replace, but add. Put the chat where visitors have questions and keep the form for people who prefer it, or whose enquiry is structured paperwork such as uploads, account numbers or long applications. Remove the form only if you are confident nobody needs it, and test the chat with a screen reader first.

Do chatbots produce better quality leads than forms? They can, when the bot genuinely converses. Survey research (CHI 2019, ACM TOCHI 2020) found chatbots that probe and follow up get more specific, informative answers than static forms. A 2026 study found no quality difference when the chat was essentially a form presented as speech bubbles. The quality comes from follow-up questions, not from the chat window.

What happens if a visitor abandons a chatbot conversation halfway? It depends entirely on the product. Some chatbots only create a lead when the conversation formally ends, so an abandoned chat leaves nothing, just like an abandoned form. Others save contact details as they come up. Ask any vendor directly what you get when a visitor leaves mid-conversation.

What is the form abandonment rate? Zuko's 2025 benchmarks put view-to-completion at 45%, which means 55% of people who see a form leave without submitting. The email and phone fields alone have abandonment rates of 6.4% and 6.3%. Be wary of the widely quoted "81%" figure: we could not trace it to a primary study of contact forms.

Does telling visitors they are talking to a bot hurt conversions? Yes. A 2019 Marketing Science field experiment with 6,200+ customers found up-front disclosure cut purchase rates by 79.7%. You should disclose anyway. The fix is not to hide the bot, but to make it useful enough that it earns the details it asks for.

What is a conversational form? A form that asks one question at a time in a chat-like layout. It is different from an AI chatbot: a conversational form asks fixed questions in a fixed order and cannot answer anything. Research from 2026 suggests this format alone feels more engaging but harder to navigate, without improving answer quality.

Can a chatbot capture leads in other languages? Yes, if it supports them. That matters more than it seems: CSA Research's 2020 survey of 8,709 consumers in 29 countries found 40% would never buy from a website in another language. An English-only form is invisible to those visitors.

Sources & further reading

From us

ConGreeto is built by Brillnex Systems, a software and AI engineering firm building production AI systems for startups and growing businesses.