Series State of AI Customer Support in E-commerce · edition 2026
Research · AI customer service
State of AI customer support in e-commerce 2026
1.09 million customer messages from up to 95 online stores, February to May 2026. How much AI handles, when shoppers ask and which models stores run.
Key findings
- 01AI agents on Chaterimo answered 1,091,830 customer messages in 452,568 conversations between February and May 2026.
- 02About 42% of conversations started outside business hours, between 18:00 and 8:00 Prague time.
- 03In May 2026, 61.2% of AI replies came from OpenAI models, 21.8% from Google, 13.8% from xAI and 3.2% from Anthropic.
- 04An average store had about 4,300 products in its knowledge base; about half of the catalogue data came from XML product feeds and half from the store's API.
- Data period
- February – May 2026 (monthly platform aggregates)
How much customer support do AI agents in online stores really handle? Between February and May 2026, AI agents on Chaterimo answered 1,091,830 customer messages in 452,568 conversations, in a group of paying stores that grew from 76 to 95. About 42% of the conversations started outside business hours (18:00–8:00).
Methodology
All numbers are aggregates of Chaterimo's production platform: four monthly snapshots from February to May 2026, paying stores only. We counted messages and conversations, the time when conversations start, the content stored in the stores' knowledge bases and the model each chatbot was set to use. The numbers are totals and averages for the whole platform; none describes a single store.
We left out metrics that were not measured reliably in this period, such as the response time of each message and sentiment.
"Outside business hours" means a conversation that started between 18:00 and 8:00 Prague time (CET/CEST), on any day of the week.
Results
Volume
| Month (2026) | Messages | Conversations | Paying stores |
|---|---|---|---|
| February | 219,510 | 97,588 | 76 |
| March | 289,478 | 126,399 | 87 |
| April | 293,231 | 115,244 | 89 |
| May | 289,611 | 113,337 | 95 |
| Total | 1,091,830 | 452,568 | – |
Source: Chaterimo platform aggregates, paying stores, February – May 2026.
From March on, the volume stayed at about 290,000 messages a month while the number of stores grew. 99.6% of conversations took place in the chat window on the store's website; the rest came from messaging platforms such as Facebook Messenger.
When customers ask
About 42% of all conversations started outside business hours, between 18:00 and 8:00 Prague time. A store that measures its support only during office hours does not see four in ten of its customers' questions.
Which models stores use
| # | Model | Share of AI replies |
|---|---|---|
| 1 | GPT-5.4 | 30.4% |
| 2 | GPT-5.4 mini | 28.0% |
| 3 | Gemini 3.5 Flash | 18.3% |
| 4 | Grok 4.3 | 13.8% |
May 2026, 144,860 AI replies with a known model. Source: Chaterimo platform aggregates.
This shows what stores chose, not which model answers best. The most used single model, GPT-5.4, is a flagship; the “mini” and “Flash” models in the table produced 46.3% of AI replies together. How the models score on the same questions is measured in the Chaterimo CX-LLM Benchmark.
What the answers are based on
An average store had about 4,300 products in its knowledge base, and the platform stored about 189 million tokens of product and policy content in seven languages. About half of the catalogue data came from XML product feeds and half from synchronisation over the store's API.
Limitations
- The stores use Chaterimo; they are not a sample of all online stores.
- The group of stores grew during the period (76 to 95), so month-to-month changes mix growth of single stores and new stores.
- Model shares count the model each chatbot was set to use, not the model of every single reply.
Download the data
Messages, conversations and paid stores per month, February – May 2026
data/monthly-volume-2026.csv
License: CC BY 4.0
How to cite
Chaterimo (2026). State of AI customer support in e-commerce 2026. State of AI Customer Support in E-commerce, 2026. https://www.chaterimo.com/en/blog/state-of-ai-customer-support/
Frequently asked questions
How much customer support can AI handle in an online store?
Between February and May 2026, AI agents on Chaterimo answered 1,091,830 customer messages in 452,568 conversations, about 290,000 messages a month from March on.
When do online shoppers contact support?
About 42% of the conversations on Chaterimo started between 18:00 and 8:00 Prague time, outside business hours.
Which AI models do online stores use for customer support?
In May 2026, 61.2% of AI replies on Chaterimo came from OpenAI models, 21.8% from Google, 13.8% from xAI and 3.2% from Anthropic.
Research · LLM models
Best AI for customer support: 12 models rated blind
ChatGPT, Claude, Gemini and Grok on 45 real e-commerce support questions, scored blind by an AI judge model. Edition 2026 Q3.
Article · LLM models
How to choose an LLM for customer support
Accuracy on your data, speed, cost and tone. How to compare ChatGPT, Claude, Gemini and Grok for an online store and test them on your own questions.
Article · AI customer service
AI chatbot for e-commerce: how AI customer service works
What an AI chatbot for an online store does and cannot do, how it differs from live chat, how to set it up, what it costs and when a person takes over.
Changelog
- Rewritten as an edition of the series with methodology, limitations and data download. Removed the claim that 89% of resolved tickets were resolved without a person: the metric counted tickets that came from a chat, and every ticket is resolved by a person.
- Published.
