When a Website Chatbot Helps Sales and When It Hurts
By CodexierPublished 6 min read
A chat widget is not good or bad in itself. It helps the visitor who has one blocking question and would otherwise leave, and it hurts the visitor who was already reading the price list and now has a bouncing bubble in the corner. Whether a chatbot improves sales on your site depends on which of those two visitors you have more of, and on where and when the widget appears. This guide argues from how buyers behave, not from vendor claims.
The short answer
The mechanism is simple. A purchase or booking is a sequence of small decisions, and a visitor leaves at the first one they cannot resolve. A chatbot that resolves that decision in ten seconds keeps the visitor. A chatbot that adds a decision, such as whether to close the bubble, loses some.
Visitors who want answers now
Some questions are cheap to answer and expensive to leave open. They are the ones where the visitor would otherwise phone, email and wait, or simply pick a competitor whose site answered. For these, a well-built bot is faster than any page because the visitor does not have to find the page.
- Do you deliver to my postcode, and what does it cost? The visitor has an address; the bot has the shipping table.
- Is this in stock in size 42, and when can I have it? Data the bot can read from the shop.
- Do you take Klarna, Swish or invoice? A one-line answer that decides the purchase.
- Can I book Thursday after 17:00? If the bot can read the calendar, it can also book the slot.
- Is this service covered by ROT or RUT? A rule-based answer, and a common reason Swedish visitors abandon a quote form.
Visitors who find chat intrusive
The other group is larger on most business sites: people comparing, reading, or already in the checkout. For them a proactive chat window is an interruption, and an unsolicited greeting from a bot pretending to be a person reads as a sales tactic. The damage is rarely dramatic, but it is measurable as slightly shorter visits and slightly lower form completion on the pages where the widget is most aggressive.
| Page type | Visitor state | Chat behaviour that helps | Chat behaviour that hurts |
|---|---|---|---|
| Home and category pages | Orienting | Closed icon, opens on click | Auto-open after a few seconds |
| Product or service page | Evaluating | Small prompt after real dwell time, question-specific | Generic greeting on load |
| Pricing page | Deciding | Offer to explain a specific plan difference | Pushing a demo or a discount |
| Checkout and booking forms | Acting | Nothing, unless an error occurs | Any pop-up at all |
Placement, timing and triggers
Most chatbot tools ship with the most aggressive settings on by default, because the vendor is measured on conversations started, not on your sales. Change three things before launch.
Placement
Show the widget only on the pages where the questions above actually occur: shipping, stock, payment, booking, eligibility. Hide it on the checkout and on legal pages.
Timing
No auto-open on page load. If you use a proactive prompt at all, trigger it on a signal that suggests a stuck visitor, such as a long dwell on the pricing page or a return to the same product twice.
Honesty and handover
Say it is a bot in the first line. Give a visible route to a human, with a real expected reply time, and make sure the bot says I do not know instead of guessing.
Measuring the effect honestly
Conversations started is the vendor's number, not yours. The number that matters is whether more visitors complete the action the page exists for. Set a baseline before launch and compare like with like: same pages, same traffic sources, a period long enough to include a full weekly cycle. Then look at three things.
- Completion rate on the target action, for visitors who used the chat and for those who did not, on the same page.
- The questions actually asked. A recurring question is a page you should write, and once written the bot answers it more reliably too.
- Handover volume and reply time. If most conversations end in a request for a human who replies the next day, the bot has added a slower channel, not a faster one.
Then decide. If completion did not move and the transcripts show mostly greetings and dead ends, remove the widget; a site without a bot is better than a site with a bad one. If it moved on specific pages, keep it there and nowhere else. When you do not need this from us: a shop with fewer than a handful of visitors a day, or a service where every enquiry needs a human anyway, gains nothing from a bot that a good FAQ page does not give. We build bots that read your own data through the AI chatbot setup service, always with a handover and always measured; pricing is on the pricing page, and if you want an honest opinion on whether your site is a candidate, book a call.
Frequently asked questions
Should the chatbot pretend to be a person?
No. Visitors notice within two messages, and the trust lost then carries over to the rest of the site. Introduce it as an assistant, keep the tone plain, and make the route to a real person visible.
Is a chatbot better than a good FAQ page?
For most business sites the FAQ page comes first, because it is cheaper, indexable by search engines and never wrong in a new way. A chatbot adds value when the answer depends on the visitor's own data, such as a postcode, a date or a stock level, which a static page cannot use.
How do we know the bot is not inventing answers?
Restrict it to your own sources, log every answer with the source it used, and read the transcripts weekly for the first month. A bot that is allowed to answer only from your pages and your product data, and says so when it cannot, is far safer than a general model with a friendly prompt.
What about GDPR for chat transcripts?
Transcripts are personal data as soon as someone types a name or an email. Tell visitors what is stored and for how long, keep retention short, and make sure the chat vendor has a data processing agreement and an EU processing option.
Wondering whether a bot would help your site?
Fifteen minutes: you show us the pages where visitors get stuck, we tell you whether a chatbot, a better FAQ page or a form change is the right fix, and what it would cost.
Book a free 15-minute call