AI chatbot statistics 2026: what real store data shows
Fourteen figures from live website assistants and published customer results. Product questions are a third of all chats, 39% happen after hours, 12.8% turn into a lead, and 1% ask for a human. Every number carries its data set and date.
What visitors ask an AI chat assistant
From 5,470 real website conversations across 246 live assistants (June 13 to September 11, 2026), plus a message-level pass over 9,480 visitor messages from live storefronts (July 21 to September 19, 2026).
32.7%
of conversations are product questions.
Support questions are 17.2% and pricing 10.3%. Product questions are the single largest category on every kind of site, not only stores.
Source: 5,470 conversations, 246 assistants, 90 days to September 11, 2026
7.0%
of storefront messages ask for a discount.
Discounts are the most common single intent on stores, ahead of product recommendations (4.7%), price (4.6%) and order status (4.2%). Shipping (2.7%), sizing (2.6%) and stock (0.8%) follow.
Source: 9,480 visitor messages on live storefronts, 60 days to September 19, 2026
1.0%
of conversations ask to be handed to a human.
The fear that visitors reject an AI assistant does not show up in the data. On storefronts the figure is 1.3% of messages, and complaints are 0.4%.
Source: Same two data sets as above
19%
of storefront first messages are written in a non-Latin script.
Hebrew, Arabic, Chinese, Japanese, Korean, Cyrillic and Thai among them. A store that answers only in English leaves one visitor in five without a real answer.
Source: 9,480 visitor messages, 60 days to September 19, 2026
When and where the conversations happen
Timing is measured in the visitor’s local time, taken from the browser at the start of the chat.
39%
of conversations start outside 9:00 to 18:00 local time.
20.7% in the evening and 18.3% at night. Only 50.2% happen during business hours, which is the window a human team usually covers.
Source: 5,470 conversations, 90 days to September 11, 2026
47.7%
of conversations happen on a phone.
On live storefronts specifically the share is higher, 55% of conversations. A widget that covers the mobile add-to-cart button costs sales.
Source: Same 5,470 conversations; storefront share from the September 19, 2026 pass
46.3%
of conversations go beyond a single question.
28% reach a third visitor message. Most of the value of an assistant is in the follow-up: the size, the shipping time, the alternative product.
Source: Same 5,470 conversations
What an assistant changes
Lead capture is measured on the assistant itself. Sales and signup lifts are from published Chatonbo customer case studies, each with the customer’s own numbers.
12.8%
of conversations end with the visitor leaving contact details.
A name, an email or a phone number left in the chat, without a form. On a site with 500 conversations a month that is 64 leads.
Source: 5,470 conversations, 90 days to September 11, 2026
$107,725
in sales attributed to the chat on one Shopify store.
Fujiiryoki, massage chairs. Attribution matches the shopper’s cart to the conversation and counts only orders placed after a chat.
Source: Published case study, chatonbo.com/customers/fujiiryoki
44%
more online hosting purchases at JetServer.
Kamatera reported 34% more billable free-trial signups, OMC 40% more website leads with 30% of support tickets resolved by the assistant, and Agura 25% more leads in the first month.
Source: Published case studies, chatonbo.com/customers
4x
higher retention for stores whose assistant has 5+ conversations in its first 14 days.
Among paid accounts, 22% of the ones that stayed had 5 or more conversations in the first two weeks, against 6% of the ones that left. The first two weeks decide.
Source: Paid accounts 14 to 120 days old, 193 accounts, September 19, 2026
How fast an assistant works
Measured on the Chatonbo platform itself.
6 min
median time from signup to a live assistant on the site.
The assistant is drafted from the website in about two minutes; the rest is installing the script or app.
Source: Chatonbo signups, August 2026
290
characters in an average reply, down from 458.
Since September 15, 2026 replies answer first in two to three sentences and ask at most one question. Long replies are read less and rated lower.
Source: Evaluation of 30 replies across 5 assistants before and after the change
4 to 5 s
from question to a complete answer.
Including the knowledge search and, for stores, the live stock or order lookup.
Source: Production measurement, September 15, 2026
Chatbot statistics FAQ
Where do these chatbot statistics come from?
From the Chatonbo production database (real conversations on live customer websites, with the query dates listed under each figure) and from published Chatonbo customer case studies. No third-party surveys are mixed in, so every number can be traced to a date and a data set.
What do people ask an AI chatbot on a website most often?
Product questions, 32.7% of all conversations. On online stores the single most common intent is asking for a discount (7.0% of messages), followed by product recommendations, price and order status.
What percentage of chatbot conversations become leads?
12.8% of conversations end with the visitor leaving a name, email or phone number in the chat, measured across 5,470 conversations on 246 assistants.
Do website visitors ask to talk to a human instead of the chatbot?
Rarely: 1.0% of conversations ask for a human, and on storefronts 1.3% of messages. Complaints are 0.4% of messages.
Can I cite these statistics?
Yes, with a link to this page. The data will be refreshed periodically and the dates updated.
Full study: What shoppers actually ask an AI chat assistant →
See what your visitors would ask
Enter your website and Chatonbo drafts an assistant from your content in about two minutes. Free plan, $19/mo flat after that, and we install it for you if you want.
Try it on my site