Short answer
Ask about that conversation with three questions at most: satisfaction, stated resolution and effort. Show it when the chat closes if the customer is still in the window, and by email only for asynchronous chats, once per conversation. Separate bot-only chats from chats with an agent, exclude chats abandoned before any reply and publish each CSAT with its response rate and number of responses.
Key verification: Close one test chat with the bot and one with an agent, and check that each receives a single linked survey.
Sources and limitationsA post-chat survey should ask about that specific conversation, with few questions and a written rule for when it appears and to whom. Show it when the chat closes if the customer is still in the window, and by email only when the conversation is asynchronous. Always separate the results of chats handled only by a bot from those handled by a person, and publish each percentage with its response rate and number of responses.
What to ask: a three-question template
Each question should measure something different. Satisfaction (CSAT), effort (CES) and stated resolution may not coincide: a customer can rate the service well and say their query is still open. The template below is an original CallsIQ proposal for you to adapt.
| Question | Answer | What it measures | How it is counted |
|---|---|---|---|
| How would you rate the help you received in this chat? | Scale from 1 to 5, from “very dissatisfied” to “very satisfied” | Conversation CSAT | Satisfied: answers 4 and 5 |
| Has your query been resolved in this chat? | Yes / Partly / No | Resolution as stated by the customer | Only “Yes” counts as resolved |
| “It was easy to handle my query through this chat” | Scale from 1 to 5, from “strongly disagree” to “strongly agree” | Perceived effort (CES) | Easy: answers 4 and 5 |
Make only the first question mandatory; the other two, optional. Do not add a text field asking for a name, email address or order number: the conversation already identifies the customer. If a tool uses a different scale, do not mix it with this one. HubSpot, for example, documents unhappy, neutral and happy responses, and Intercom a five-emoji scale.
When to show it: at chat close or by email
In a live chat, the best moment is the close, while the customer is still in the window. Tools handle this differently, and you should know which event triggers the survey in yours. Documentation consulted on 22 October 2026:
- Zendesk explains that in messaging the CSAT survey is sent when the ticket is set to solved, and that agents cannot launch it manually (CSAT experience for email and messaging).
- Intercom sends the rating request a few seconds after a teammate closes the conversation, and does not send it, among other cases, if there was no teammate reply or if the conversation is shorter than 250 characters (conversation ratings).
- HubSpot lets you show a satisfaction survey in the widget after a chat is closed, with Service Hub Professional or Enterprise (customer satisfaction surveys).
Email makes sense when the chat is asynchronous and the customer is no longer on the website. Send it promptly, only once per conversation, and do not combine it with the widget survey: the same chat would be counted twice. If the query is reopened, decide whether to ask again and apply the rule the same way to every team.
Bot, agent and transfers: who gets the rating
A chat resolved only by a bot and one handled by a person are different experiences. If you mix them, the result shifts with the share of automated conversations even though nobody has improved. Intercom documents a specific survey for its Fin AI agent and a rated agent type (teammate, Fin or chatbot); it also lets you exclude AI-only conversations with the condition that a teammate has replied (Fin CSAT).
With transfers, set a rule: the response is attributed to the last person who handled the chat, or to a “bot transferred to agent” segment. The second option shows whether the handover works. Do not attribute to an agent the rating of a chat that started badly because of a long wait in the bot without reviewing the conversation.
Abandoned chats, asynchronous chats and inactivity closures
A chat abandoned before anyone replied is not an experience that can be rated: count it separately, as abandonment. Automatic closures for inactivity are trickier, because they can close a resolved query or leave a customer without an answer. Zendesk states that if the agent ends the messaging session, the default CSAT trigger is not met, and that sending the survey in that case requires a custom CSAT trigger, available from Suite Growth upwards (ending messaging sessions).
Check in your tool what happens with each type of closure and write it down. If inactivity closures receive no survey, the results only represent those who reached the end of the conversation.
Response rate and sample size: a fictitious example
A fictitious retailer reviews one month of chats. It excludes 300 chats abandoned before any reply and separates the segments.
| Segment | Eligible chats | Responses | Satisfied (4 or 5) | Response rate | CSAT |
|---|---|---|---|---|---|
| Bot only | 1,000 | 120 | 78 | 12.0% | 65.0% |
| With agent (includes transfers) | 800 | 200 | 164 | 25.0% | 82.0% |
| Total | 1,800 | 320 | 242 | 17.8% | 75.6% |
Calculations: response rate = responses ÷ eligible chats (120 ÷ 1,000 = 12.0%; 200 ÷ 800 = 25.0%; 320 ÷ 1,800 = 17.8%). CSAT = satisfied ÷ responses (78 ÷ 120 = 65.0%; 164 ÷ 200 = 82.0%; 242 ÷ 320 = 75.6%).
The total is misleading: chats with an agent are 44.4% of eligible chats (800 ÷ 1,800), but they provide 62.5% of responses (200 ÷ 320), because those customers answer more. If you weight each segment by its chat volume, the result is (0.65 × 1,000 + 0.82 × 800) ÷ 1,800 = 1,306 ÷ 1,800 = 72.6%. Publish the segments and explain which total you use.
On sample size: with 200 responses and 82%, the approximate 95% margin is 1.96 × √(0.82 × 0.18 ÷ 200) ≈ ±5.3 points; with 120 responses and 65%, ≈ ±8.5 points. That margin assumes that those who respond resemble those who do not, which a 12% rate does not guarantee. The guide on NPS and sample size covers the problem of small segments, and the one on first contact resolution explains why stated resolution does not replace the operational data.
Linking the response to the conversation without asking for unnecessary data
The cleanest approach is for the survey to carry the conversation identifier, as native surveys do, so that the response is associated with the chat, channel, segment and agent without asking anything else. If you use an external form, pass an opaque identifier and not the email address or name in the URL.
The AEPD, the Spanish data protection authority, points out on its page on processing principles that only the data needed for each purpose should be processed and kept for the necessary time. Applied here: explain what the response is used for, do not ask for data you already hold, limit who sees the comments and set a retention period. Free-text comments may contain personal data; consider whether to hide them in reports.
Check before publishing the indicator
- Open a test chat with the bot and another with an agent, close them and verify that each receives a single survey.
- Check what happens when a chat closes for inactivity and when the agent ends the session.
- Answer the test surveys and confirm that each response is linked to its conversation and segment.
- Exclude those tests from the report.
- Publish CSAT by segment alongside eligible chats, responses and response rate.
For the phone channel, the invitation rule and the scale change; see the guide on CSAT surveys after phone support. CallsIQ has not configured these tools for this guide; confirm conditions and plans in your account.
Sources and limitations
Documentary review: . Content type: Documentation-based guide with original resource.
Linked sources describe the terms and capabilities each provider publishes. Examples that use fictional data are labelled in the text.
- CSAT experience for email and messagingsupport.zendesk.com
- conversation ratingswww.intercom.com
- customer satisfaction surveysknowledge.hubspot.com
- Fin CSATwww.intercom.com
- ending messaging sessionssupport.zendesk.com
- processing principleswww.aepd.es
Official sources, explained calculations and clearly labelled examples. Read about our methodology and use of AI in writing.