The State of Ecommerce Chat 2026

1. Chat volume on live stores grew 4.3x in three months

Storefront conversations per day went from 19.9 in June to 26.7 in July, 58.5 in August and 85.7 in September. The first 24 days of September alone hold 2,057 conversations, more than a third of the 5,565 in the whole window. The number of assistants with at least one chat in the month rose from 61 in June to 97 in August and 95 in the partial September.

Storefront chats only. With dashboard test chats (no page URL) included, the total would be 6,417 conversations; the report never uses them.

2. 44.7% of storefront chats start when the shop is closed

Of 5,401 conversations with a known visitor country, 55.3% start between 09:00 and 17:59 in the visitor’s local time. The other 44.7% do not: 24.7% start in the evening (18:00 to 22:59) and 20.0% at night (23:00 to 08:59). Chats that start in business hours convert to a lead more often (16.7%) than evening (10.6%) or night chats (9.6%), but the after-hours volume is far too large to leave to a contact form.

164 conversations (2.9%) from countries outside the offset map are excluded from this table and the next.

Weekends are the most after-hours days

48.2% of Sunday chats and 47.7% of Saturday chats start outside 09:00 to 17:59, against 41.1% on Fridays. Weekends carry 26.2% of all storefront conversations (1,417 of 5,401), so a Saturday or Sunday is a normal shopping day for chat even though almost half of it happens outside office hours.

Noon is the busiest hour on the visitor’s clock

The 12:00 to 12:59 local hour holds 6.7% of conversations (364 of 5,401) and 04:00 is the quietest hour (33). Measured in UTC instead, the peak is 14:00 (360 of 5,565), which mostly reflects the mix of US and European visitors.

3. Phones carry the traffic, desktops produce the leads

54.3% of storefront chats happen on a phone, 42.4% on a desktop and 2.0% on a tablet (3,021, 2,357 and 109 of 5,565, classified from the browser user agent). The device changes what happens next: 19.0% of desktop conversations end with the visitor leaving a name, email or phone number (448 of 2,357) versus 10.0% on phones (301 of 3,021), and desktop visitors ask a follow-up question in 53.7% of chats versus 40.0% on phones.

Tablet and script rows are below the size at which lead rates are reported. Script rows (automated clients) are kept in the denominator.

4. 13.6% of conversations end with contact details, but the typical assistant sees 5%

759 of 5,565 storefront conversations end with the visitor leaving a name, email or phone number in the chat, without a form. The pooled figure is conversation-weighted and reflects a few high-volume assistants that capture most of the leads: the median assistant with 20 or more chats converts 5.0% (31 assistants). Both numbers are true; the second is the one to benchmark a single site against.

Service and lead-gen sites capture leads at nearly three times the rate of stores

On sites whose assistant is not selling from a catalogue, 25.6% of conversations produce contact details (403 of 1,576 across 50 assistants). On online stores it is 8.9% (356 of 3,989 across 132 assistants). The per-assistant medians tell the same story at a lower level: 10.1% for service sites versus 4.8% for stores. A shopper asking about a product mostly wants the answer, not a callback; a visitor to a law firm or an agency is often there precisely to be contacted.

By platform

Pooled lead rates are 20.3% on Wix sites (390 of 1,918), 10.3% on Shopify stores (311 of 3,030) and 9.4% on other platforms (58 of 617). The Wix figure is inflated by service sites and by one very large assistant that records leads through a pre-chat form; per assistant, Wix (4.8%) and Shopify (5.6%) are close. WooCommerce had no live storefront conversations in the window and is not reported.

Platform is the store’s app installation or the platform detected on the site. Store assistants are the catalogue and Shopify assistant types; service and lead-gen sites use the standard type.

5. The fourth visitor message is the tipping point

A conversation that reaches four visitor messages turns into a lead 47.7% of the time (488 of 1,023), against 3.8% for single-message chats (117 of 3,040). Most chats never get there: 54.6% of storefront conversations are a single visitor message, 45.2% go beyond one question, 18.4% reach four or more visitor messages and 1.2% run to eleven or more. The median conversation has 1 visitor message and the 90th percentile has 5; the mean is 2.34 visitor messages and 4.68 messages in total.

The 11-or-more bucket is small (65 conversations) and its lead rate should be read as indicative.

6. Product questions are the largest topic; requests for a human are rare

36.7% of conversations are product questions (2,041 of 5,565), followed by support at 14.0%, pricing 10.4%, shipping 8.5%, contact and lead requests 8.1%, order status 2.2%, complaints 0.8% and returns 0.7%. Another 10.6% fall under other topics and 8.0% could not be classified. Topics are automatic per-conversation labels, the same definition used in the June to September study of what shoppers ask.

Unclassified conversations are kept in the denominator. Of classified conversations alone, product questions are 39.9%.

Only 1.6% ask for a human, 2.2% complain

Applying Chatonbo’s production escalation rules (multilingual phrase patterns for “talk to a person”, “refund”, “never arrived” and so on) to every visitor message, 86 of 5,376 conversations with at least one message of four or more characters contain a request for a human (1.6%) and 119 contain a complaint (2.2%). The people who did ask for a human were unusually likely to leave contact details in the same chat: 34.9% did, versus 13.6% overall. A request for a person is a lead in disguise, not a failure.

7. Going live takes minutes, and catalogue size does not predict chat volume

Among the 344 workspaces created in 2026 that put an assistant live on their site, half did so within 12 minutes of creating the workspace (median 0.2 hours). 86.9% went live within 24 hours and 94.8% within 7 days. The median is pulled down by Shopify and Wix app installs, which create the workspace and embed the widget in a single flow; for direct website signups that install the script themselves, the median is 24 minutes.

The median connected Shopify store has 63 products (interquartile range 20 to 267, 90th percentile 653, largest 14,379, across 130 stores with a synced catalogue). Splitting the stores at the median, live stores below it averaged 14.5 conversations each and stores above it 14.1. Larger catalogues were somewhat more likely to get any chat at all (69.8% versus 59.6% of live stores), but a big catalogue does not bring more conversations per store. Chat volume follows site traffic, not product count.

Stores with no synced product count (138) are excluded. Finer size buckets fell below the reporting threshold and are withheld.

Methodology

Source

Chatonbo’s production PostgreSQL database, queried read-only on September 24, 2026. The server clock is UTC and the last conversation counted starts before 00:00 UTC on September 25. Only aggregate counts were extracted. No customer name, visitor identity, message text or site URL was read for this report; the single per-row text operation (the escalation phrase match in section 6) ran inside the database and returned counts only.

Window

March 13, 2026, the date of the first customer conversation in the database, to September 24, 2026 inclusive. Signup and go-live figures use workspace and assistant creation dates in the same window. September is a partial month of 24 days, so per-day rates are given alongside monthly totals.

Exclusions

Chatonbo’s own workspaces and internal test accounts are removed: 10 workspaces, 12 assistants and 3,510 conversations, mostly internal Hebrew traffic. Dashboard test chats, identified by a missing page URL (852 conversations), are excluded from every conversation figure. Per-assistant medians use only assistants with at least 5 conversations, or 20 where stated. No cell smaller than 20 is reported; where a cut fell below that (WooCommerce, finer catalogue size buckets, the per-assistant median for other platforms, tablet lead rates) the number is withheld and the reason given.

Definitions

• Storefront conversation: a chat session opened from the widget on a customer’s live website, identified by a recorded page URL.
• Live assistant: an assistant whose widget has loaded on the customer’s site at least once.
• Lead: a conversation in which the visitor leaves a name, email or phone number and the assistant acknowledges it. Same definition as the June to September study (12.8% there, 13.6% here on the wider window). The separate leads table agrees to within one conversation (813 versus 812).
• Topic: the automatic per-conversation label Chatonbo assigns, the same field the earlier study used. Unclassified conversations stay in the denominator.
• Asked for a human, complaint: the exact multilingual phrase patterns Chatonbo uses in production to escalate a chat (English, Hebrew, Spanish, French, German and Portuguese), applied to every visitor message of 4 to 4,000 characters. A conversation counts once, and “human” takes precedence over “complaint”. The stored escalation flag only started on September 20, 2026, so it is not used.
• Device: from the browser user agent stored on the conversation. Phone = iPhone, iPod, Android with “Mobile”, Windows Phone; tablet = iPad, “Tablet”, or Android without “Mobile”; script = curl, node, python, bot, crawler or headless clients (51 conversations, kept in the denominator).
• Local time: the conversation start time in UTC shifted by one fixed offset per visitor country (44 countries mapped, summer offsets since the window is mostly March to September, the United States treated as UTC-5). Conversations from unmapped countries (164, 2.9%) are excluded from time-of-day and weekday figures.
• Signup to live: the gap between a workspace’s creation and the first time any of its assistants loaded on a site. Shopify and Wix app installs create the workspace and embed the widget in one flow; “direct” means every other workspace.
• Catalogue size: the synced product count of a Shopify store linked to an assistant, where that count is above zero (138 stores with no synced count are excluded).

Caveats

• Time zones. Local-time figures use one offset per country and ignore daylight-saving changes and multi-zone countries (the US, Canada, Australia, Brazil and Chile changed offsets on September 7). Treat the after-hours split as accurate to a few percentage points.
• Concentration. Ten assistants generate 73.5% of conversations and the single busiest one 20.7%. Pooled percentages are conversation-weighted and reflect those large sites; per-assistant medians are given wherever the cell size allows.
• Free versus paid. 52.9% of conversations come from free-plan assistants and 47.1% from paid plans, but three Business-plan assistants account for 24.4% of all chats. Plan is measured on September 24, not at the time of the chat.
• Lead detection. It depends on the visitor typing contact details into the chat. Assistants that use a pre-chat form record leads differently, and one such Wix assistant (266 leads in 397 chats) lifts the Wix and service-site pooled rates.
• Topic labels. They are model-generated and 8.0% of conversations are unclassified.
• Go-live timing. The 12-minute median is driven by app-store installs, which embed the widget as part of installation. The direct-signup median of 24 minutes is based on 67 workspaces.

Frequently asked questions