A chatbot does not increase conversions by being on the page. It increases them by removing a specific delay — the one between a visitor forming a question and getting an answer good enough to act on. If it does not remove that delay, it is a widget that covers your content on mobile.
Why visitors leave without contacting you
People who leave a B2B website without making contact usually are not uninterested. They have a question the site did not answer, and the cost of finding out felt higher than the cost of closing the tab.
- The question is specific and the site is general: does this work with our existing system, do you cover our region, can you handle our volume.
- The only way to ask is a form, and a form means waiting an unknown length of time for an answer that might be a sales call they did not want yet.
- They are not ready to identify themselves. Handing over a work email to ask one question feels disproportionate.
- They are comparing three vendors in one session and will not wait on any of them.
This is where chat changes the economics. Not because it is a chat window, but because it lets someone ask a specific question anonymously and get an answer in the same session. Everything that follows depends on the answer actually being useful.
Where a chatbot affects the funnel
There are three distinct effects, and they are worth separating because they need different things to work.
-
1
Response speed
An answer inside the session, while the visitor still has the page open and the question in mind. This is the effect that does not depend on sophistication — even a well-built answer to a common question beats a form confirmation.
-
2
Qualification
By the time a conversation reaches a person it can already carry the context: what they are trying to do, roughly what scale, what timeline. The salesperson starts from an informed position rather than from scratch.
-
3
Routing
A support question goes to support, a pricing question to sales, a partnership enquiry somewhere else entirely. Getting this right removes the delay where a message sits in the wrong inbox.
Notice that only the first is about the bot talking. The other two are about what happens to the conversation afterwards — which is why a chatbot bolted onto a site with no routing behind it rarely moves the numbers.
What to automate and what to hand to a person
The line is not about difficulty. It is about whether the correct answer is already written down somewhere.
- Automate: questions with one right answer that does not change per customer — coverage, compatibility, delivery times, how a process works, what a product does and does not do.
- Automate: the qualifying questions you would ask anyway on a first call, when the visitor has already signalled buying intent.
- Automate: booking. Once a conversation is qualified, putting time in a calendar is a mechanical step that should not wait for a human.
- Hand over: anything where the answer depends on judgement — pricing for a non-standard scope, whether an edge case is feasible, negotiation of any kind.
- Hand over: a visitor who is already unhappy. Automation handles new questions well and existing frustration badly.
- Hand over immediately on request. A visitor who types that they want a person should not have to type it twice.
What to capture without creating friction
The instinct is to treat chat as a form with a conversational skin — ask for name, email, company and phone before answering anything. This reliably reduces conversations, because the visitor came to ask a question, not to be processed.
The better sequence is to answer first and ask second. Once you have given something useful, asking for a contact detail is an exchange rather than a toll.
- Capture passively from the start: the page the conversation began on, the referrer, the campaign, the device. None of this requires asking.
- Ask for one contact detail, not four. Whichever one your follow-up process actually uses — usually email or phone, rarely both.
- Ask qualifying questions only where the answer changes what happens next. If it does not change routing or preparation, it is friction with no return.
- Let the visitor stay anonymous if they want. An answered question builds enough goodwill to bring them back; a refused answer does not.
- Pass the full transcript into the CRM. The exact wording of the question is often more useful to a salesperson than the fields around it.
Measuring whether it works
Most chatbot reporting measures the chatbot: sessions, messages, containment rate. None of those tell you whether the website performs better. Measure the funnel instead, and measure it against a period without the bot.
- Assisted conversions: leads whose journey included a chat conversation, compared with those that did not. This is the number that answers the business question.
- Median time from first message to a useful answer. Not average — one long conversation distorts it.
- Qualified conversation rate: of the conversations that reached a person, how many were a real fit. A high volume of unqualified handoffs means the qualification step is not doing its job.
- Handoff success: how often a request for a human was actually met, and how long it took. A failed handoff is worse than no chat.
- Deflection on repeat questions: how many of the top ten questions are now answered without a person. This is where the operational saving is, if there is one.
- Mobile completion rate specifically. Chat widgets behave very differently on small screens and often cover the content people came for.
If you cannot yet measure assisted conversions, that is the first thing to fix — before adding a bot, not after. Attribution added afterwards tends to produce arguments rather than answers.
Adding chat to an existing website
-
1
Collect the real questions first
Pull the last few months of sales and support enquiries and count what actually gets asked. Build for the top ten, not for what you imagine visitors want.
-
2
Write the answers as content, not as bot scripts
The same answers belong on the website itself. A question that has to be asked in chat because the site does not answer it is a content gap first and a chatbot use case second.
-
3
Define the routing before switching it on
Who receives a sales conversation, who receives support, and what happens outside working hours.
-
4
Decide the out-of-hours behaviour explicitly
Either the bot handles it end to end, or it sets an expectation and captures a contact detail. What it must not do is promise an immediate reply that will not come.
-
5
Place it so it does not obstruct the page
Check mobile specifically. A widget that covers a call-to-action costs more conversions than it creates.
-
6
Review transcripts weekly for the first month
The questions it answers badly are the fastest route to improving both the bot and the site.
Steps two and five are website work rather than bot work, which is the usual finding — see website development and AI integrations for how we normally split that.
When a dedicated platform makes sense
A simple bot answering ten questions from a fixed script is a contained piece of work and can reasonably be built into an existing site. The calculation changes once any of the following is true.
- The same conversations need to continue across channels — website, WhatsApp, Instagram — with the customer recognised as one person rather than three.
- Non-technical staff need to update the answers, which means a knowledge base with a real editing interface rather than a file in a repository.
- Conversations need to reach a live person with the context attached, including outside working hours.
- You want the same knowledge to serve both chat and phone, so a customer gets a consistent answer whichever way they ask.
At that point you are describing a communication platform, not a website feature, and building it is a multi-year commitment to a problem that is not your business. Vexvon covers this category — an AI chatbot working from one knowledge base across web and messaging channels, with handoff to a person built in. The work that remains, and the part we usually take on, is fitting it to the site and the CRM so the conversation, the lead record and the follow-up are the same thread rather than three disconnected ones.
If the follow-up side is what you are unsure about, what happens after a lead form is submitted covers the workflow a chat conversation should feed into.
Frequently asked questions
-
1
Can AI chatbots really increase website conversions?
They can, when they remove a real delay — a specific question answered in the session instead of after a form and a wait. They do not, when they are added to a site whose actual problem is unclear positioning or a broken follow-up process. The honest test is whether visitors are currently leaving with unanswered questions. If they are not, a chatbot has nothing to fix.
-
2
What should a website chatbot ask a lead?
As little as possible, as late as possible. Answer the visitor's question first, then ask for the one contact detail your follow-up actually uses, plus any qualifying question whose answer changes routing. Everything else can be captured passively or asked by a person later.
-
3
Should a chatbot replace live chat?
No — it should sit in front of it. The bot handles the repetitive and out-of-hours volume; the person handles what needs judgement. What matters is that the path from one to the other is fast and carries the conversation with it.
-
4
How do you measure chatbot ROI?
Compare assisted conversions against a comparable period without chat, and set that against the platform cost plus the time spent maintaining answers. Bot-side metrics like session counts and containment rate describe activity, not return.
-
5
Will a chatbot hurt our SEO?
Not by existing, but a heavy widget can affect loading and layout stability on mobile, and both are part of how pages are assessed. Load it after the main content, check it against your Core Web Vitals, and make sure it does not cover the page on small screens.
