Picture someone with a quick question about your returns window. Thirty days or sixty? They tap the chat bubble and, before they've typed anything, the widget throws up a pre-chat form wanting their name, their email and a pick from a dropdown labelled Department. Which department does returns? No idea. They close it and search instead, and the page that answers them probably isn't yours.
We'd delete that form. Of all the habits in support that nobody puts a number on, it's the priciest one we know. It had a good reason to exist once. For most teams with an AI agent answering first, the reason is gone. Below: the case against it, the honest case for it, and a two-week test so you can stop taking our word for any of this.
Written by the ChatterMate team. We build an open-source AI support agent, and we made a deliberate choice not to put a form in front of it. That bias is on the table from the start.
Why the pre-chat form exists in the first place
The pre-chat form was a sensible answer to a staffing problem. Live chat used to mean a person on the other end. People go to lunch, go home, and get stuck in long conversations. If a visitor opened chat while everyone was busy, the team needed some way to reply later. An email address was that way. Asking for it up front guaranteed you could follow up even if the visitor left.
Routing was the second reason. A team with a billing desk and a technical desk didn't want every chat landing in one pile, so a dropdown let the visitor sort themselves. And a name made the opening line feel less cold: "Hi Priya" beats "Hi there".
Vendors still sell it on those grounds. Chatwoot's pre-chat forms page lists name, email and phone as standard fields, lets you add custom ones such as order numbers, and pitches the form as a way to avoid repeated questions and send customers to the right team. Comm100's pre-chat survey best practices makes a similar argument: collect an order ID early and the chat gets shorter.
None of that is wrong. It's just built on an assumption that has quietly stopped being true for a lot of teams: that the first reply has to wait for a human.
What a pre-chat form costs you
Every field is a small tax on someone who came to you with a problem. Form research is unusually clear on this. Baymard Institute's long-running checkout studies found the average checkout had 11.3 form fields in 2024, while most sites need only about 8. Their point is that people give up on forms that ask for more than the task needs, and that the field count hurts more than the number of steps.
A checkout isn't a chat, and we won't pretend the numbers transfer one for one. But the mechanism does. A shopper at checkout has already decided to buy and still bails on extra fields. A support visitor has decided nothing. They're testing whether you'll help them quickly. That's a weaker commitment, so the form should hurt at least as much.
Then there's the field you didn't think could hurt. Expedia once traced a run of failed bookings to an optional "Company" field. Some customers typed their bank's name there, then the bank's address, and card verification failed. As reported by ZDNet, the company's head of analytics said removing it was worth $12m of profit a year. One optional box. Fields have side effects nobody plans for.
What bugs us most is that you can't see the cost. Chat reports start counting at message one. Someone who saw three boxes and bailed isn't in your CSAT, isn't in first response time, isn't anywhere. So your satisfaction score is really a score for the people who put up with the form.
Three things change when an AI answers first
Here is where our opinion comes from. When an AI agent answers the opening message, each of the old reasons for the form gets weaker.
You no longer need an email to reply
The visitor gets an answer in the same session, at 2pm or 2am. The follow-up problem only exists for the minority of conversations the AI can't finish. Asking every visitor for an email to cover that minority is backwards. Ask the minority, at the moment it matters. That's how we built ticketing in ChatterMate: when the AI decides a conversation needs a ticket, the AI ticketing flow checks for an existing ticket and then asks for the customer's email. Not before.
Routing can come from what they actually wrote
A dropdown asks the customer to understand your org chart. They don't, and they shouldn't have to. Is a failed card payment "Billing" or "Technical"? Depends on the company. The first message usually tells you more than any dropdown. "I was charged twice for order 4471" is a billing question with an order number in it. The agent can read that, and if it needs a person, the human handover can carry the full conversation across so nobody asks again.
Personalisation is overrated in the first line
Has anyone ever stuck around on a support chat because it said hi using their first name? We doubt it. People stay when the first reply actually helps. Need a name later, for a refund, say? Ask then. Nobody minds a question that obviously has a reason behind it.
The case for keeping the form, taken seriously
We'd be writing a sales page if we stopped there. There are real cases where some kind of gate earns its place, and a few where we'd keep one ourselves.
No AI and nobody online. If your chat is human-only and the team is offline, the widget is really an email form wearing a chat costume. Asking for an address up front is honest in that case. Better still, say plainly that nobody is around and when they'll reply. We argued this at length in our post on honest after-hours support.
Account-specific actions. If the chat can change an order, cancel a subscription or reveal account details, you need to know who you're talking to. But a typed email isn't verification. Anyone can type anyone's address. For logged-in customers, pass identity from your app instead of asking. For everyone else, use a proper check at the point of the sensitive action. Comm100 makes this point too: skip the survey when you already know the visitor.
Spam and abuse. Some widgets get hammered. A required email does deter some junk, though a determined bot fills it in as easily as anything else. Rate limits and bot checks on the widget handle this better than a form that every real customer also has to fill in.
Sales handoff in B2B. If a chat is mostly a lead-capture surface for a sales team, the sales team wants contact details. Fair. But you'll get more of them by answering the question first and asking once there's a reason, which is the approach behind our conversational lead capture: the contact fields are collected by asking in the chat, not by a form.
Notice the pattern. Every good reason is about a specific moment in a specific kind of conversation. None of them justifies a form in front of every conversation.
A field-by-field rule: ask at the moment of need
Here's the framework we use. For each field on your current pre-chat form, find the moment you actually need it, and move the question there.
| Field | Do you need it to answer the first question? | When you actually need it | Where to ask instead |
|---|---|---|---|
| Name | No | Booking, refund, or a human picking up | In the chat, at handover or when a record is created |
| No, if the AI replies in session | When a ticket is opened or a follow-up is promised | In the chat, with the reason stated ("so we can reply when this is fixed") | |
| Phone | Almost never | Callbacks, urgent field service | Only when a callback is offered and accepted |
| Department dropdown | No | Never, from the customer's side | Infer from the message; let the handover rules route it |
| Order number | Sometimes | Order status or returns questions | Ask only when the question is about an order |
| Account ID | Sometimes | Account changes | Pass from your app for signed-in users; never ask them to type it |
| Consent checkbox | Depends on your rules | Before you store contact details for marketing | Next to the contact question, not in front of the chat |
Run your own form through that table and see how many fields survive at the front door. Our guess is that for most teams, the answer is none, or one at most.
There's a privacy upside too. Under UK GDPR, personal data has to be "adequate, relevant and limited to what is necessary" for the purpose. Collecting a phone number from someone who wanted to know your opening hours is hard to square with that. Asking at the moment of need is simply easier to defend. If you want the wider picture on what a chat tool should keep and for how long, our piece on chatbot data retention goes further.
What the numbers might look like: a worked example
We don't have a cross-industry stat for pre-chat form abandonment, and we'd be suspicious of anyone who claims one. So here's an illustrative model, not data. Plug in your own numbers.
Say your widget is opened 1,000 times a month. With a three-field form, suppose 60% of openers complete it and start a chat. That's 600 conversations. Your AI resolves, say, 55% of them, and 270 need a human or a ticket. You have 600 email addresses, most of which belong to people who got their answer and will never need a follow-up.
Now remove the form. Suppose 85% of openers send a first message. That's 850 conversations. Same resolution rate gives about 470 answered in session and roughly 380 needing more. You ask those 380 for an email at the point of the ticket. If 80% give it, that's around 300 addresses.
So in this sketch you trade 600 mostly unused emails for 300 useful ones, and you start 250 more conversations, about 140 more of them answered on the spot. Whether that trade is good depends on your business. A B2B team that lives on leads might weigh the 300 lost addresses heavily. A store whose real goal is fewer abandoned carts probably won't. The point of the model is that the trade exists and has two sides. Most teams never see the first side because the people who left never appear in a report.
How to test it in two weeks
Opinions are cheap, including ours. Here's how to find out for your own site.
Week 0: measure the gate. Before changing anything, get one number most tools don't show by default: widget opens versus first messages sent. If your chat tool can't show opens, a tag in your analytics on the widget-open event will do. The gap between those two numbers is what your form costs you today.
Weeks 1 and 2: split traffic. Comm100's guidance recommends exactly this kind of A/B test, comparing a form-gated chat against a direct one. Show half your visitors the current form and half a widget that opens straight to the message box. Keep everything else the same: same AI, same knowledge base, same hours.
Track four things per arm. First messages per widget open. Share of conversations resolved without a human. Emails collected for tickets that needed one. And CSAT from the post-chat survey. If you want a refresher on why the CSAT number alone can mislead you, we wrote about chatbot CSAT separately.
Decide on what you'd regret. If the no-form arm helps more people and you still get an email from nearly everyone who needs a follow-up, the form was only collecting addresses you never used. If your sales team loses leads they genuinely worked, keep a light version, ideally a single optional email field shown after the first answer rather than before it.
Two weeks is short. It's enough to see a gap at the top of the funnel, which is where the pre-chat form does its damage. It's probably not enough to judge lead quality, so give the sales side a month if that's what matters to you.
Where we landed, and why
We built ChatterMate around the idea that the first reply should be an answer, grounded in your own docs, not a request for paperwork. Contact details are collected by asking, in the conversation, at the moment there's a reason. The AI asks for an email when it opens a ticket. When a person takes over, they pick up the same conversation instead of starting from zero. If you've already decided a form is right for your business, that's your call, and the table above will at least help you trim it.
But try the test first. We suspect most teams that kept a pre-chat form never tested it against no form at all. They inherited it from a time when a human had to answer every chat, and nobody went back to ask whether the reason still held.
If you'd like to try the no-form approach, ChatterMate is open source and free to start: the Free plan includes 50 messages a month, new accounts get 14 days of the Growth plan with no card, and you can self-host the whole thing if you'd rather keep the data on your own servers. Point it at your docs, open the widget, and see how many people talk to you when you stop asking them who they are first.



