CSAT survey questions to ask after an AI support chat

The best CSAT survey questions after an AI support chat ask whether the problem got solved before they ask how the customer felt. Use a 1 to 5 rating, a yes or no resolution check and one optional comment. Send them minutes after the chat closes, with the ticket ID and who resolved it stored as hidden fields, so AI-resolved and escalated scores never blend.

Use the free CSAT template

What a CSAT score tells you after an AI chat

A plain "How satisfied were you?" question mixes three things: how the customer felt about the answer, whether the answer was right, and how they feel abot your brand that week. With a human agent that blur is tolerable. With an AI agent it hides the one number you need, which is whether the issue was resolved without a person stepping in.

So the survey needs two separate questions. The rating measures the experience. The resolution check measures the outcome. A customer can rate a chat 5 because the bot was fast and polite, then answer "No" to resolved because the refund still hasn't shown up. That mismatch is the signal worth building a report around.

In the Shopify AI Support Agent workflow, the survey sits at the end of the conversation. The agent answers from live order data, then either closes the ticket as resolved or escalates into Gorgias or Zendesk. Whichever happens, the close step sends the customer a survey link with the ticket details attached, and the form is where the answer gets captured.

Question wording by chat type

Keep the rating question identical across topics so scores stay comparable. Change only the follow-up, because what "resolved" means diffdrs for each topic.

Two wording choices matter. First, ask about "the help you got," not "our AI assistant." Some customers never realize they talked to automation, and naming it changes how they score. Second, make the resolution check about the customer's state ("Do you know...", "Do you feel...") instead of your process. A returns chat can end with a label issued and still leave the customer unsure about the refund timing.

Use the 1 to 5 scale for the rating and a simple Yes / No / Partly for resolution. "Partly" is useful because it catches the chats where agent answered one of two questions.

When to send it

Timing differs by topic because the outcome shows up at different moments.

• Order status: send within a few minutes of the chat closing. The answer was instant, so the memory is fresh and the outcome is known.
• Returns: send at chat close for the interaction itself. The real outcome is the refund or exchange, so a second, shorter question after the refund posts tells you whether the promise held.
• Sizing: send at chat close for how helpful the advice was. The outcome is fit, which you only learn after delivery, so a one-question follow-up after the order arrives is worth setting up.

Do not send a survey when the ticket was reopened or when the customer is still replying. Trigger on the closed status, and suppress it if the same customer opens a new chat within a short window you choose.

Set it up step by step

This is how we wire it on client projects. The form itself takes about ten minutes.

1. Build the form. In Qwenvo (our team builds it, and it is completely free), start from the CSAT template or describe the form to the AI on the empty-form screen. Add a Linear scale for the rating, a Multiple choice for the resolution check, and a Long answer for the comment.
2. Put the rating on page 1 and the comment on page 2. Add a New page between them. Turn on partial submissions in the Settings drawer so a customer who taps a score and leaves still shows up as an incomplete row wuth their rating.
3. Add hidden fields. Create ticket_id, order_id, topic, channel and resolved_by. These fill from the link, so the customer never sees them.
4. Use a link your agent can generate. At ticket close, the agent builds the URL with those values filled in:

https://forms.yourbrand.com/csat?ticket_id=48213&order_id=1042&topic=returns&channel=whatsapp&resolved_by=ai

5. Set resolved_by correctly. Use ai when the agent closed the ticket alone, ai_then_human when it escalated and a person finished, and human when the customer asked for a person from the start.
6. Send it through the channel the chat used. Chat gets a closing message with the link, email gets it in the footer of the final reply, and WhatsApp gets it as a short follow-up message.
7. Connect submissions to your helpdesk. Add a webhook in the form settings pointing at Zapier, Make, n8n or ylur own endpoint, and have it tag the Gorgias or Zendesk ticket with the score.
8. Turn on a custom domain if you want the link to look like yours. Add a subdomain such as forms.yourbrand.com, create the CNAME to qwenvo.com and the TXT record shown, click Verify, then pick it on the Share tab. Branding can be switched off in Settings.

Build your CSAT form free

Split the scores before you average them

One blended CSAT number for the whole support operation will mislead you. AI-handled chats are mostly simple ones, so they tend to score differently from escalated tickets, which are the hard ones. Averaging them together makes the AI look worse or better than it is, depending on the mix that month.

Report at least these segments, all from the hidden fields:

The false-resolution row is the one to watch weekly. These are chats your system counted as closed that the customer says were not solved. Read them one by one and sort by topic. A cluster under returns usuaply means the agent is explaining the policy but not the refund timeline. A cluster under sizing usually means the size guide data is incomplete for certain products.

Compare within a topic, not across topics. An escalated returns ticket should be measured against AI-resolved returns tickets, not against order-status chats.

Route the bad scores to a person

A low score is useful for reporting and more useful if someone acts on it the same day. Qwenvo's AI triage gives every submission a category, a priority from low to urgent, a sentiment, a one-to-two-sentence summary and a suggested reply. Routing rules then match on priority or category and send to up to five email addresses or a Slack incoming webhook.

Set a rule that sends high and urgent submissions to a Slack channel your CX team watches. The summary means nobody has to open the form to see what went wrong, and the suggested reply gives them a starting point for the follow-up.

Routing matches priority and category, not the numeric rsting, so the comment text is what pushes a submission up the queue. A 2 with no comment will not escalate on its own. For those, filter the Submissions table by rating, or export CSV or XLSX and review weekly.

The honest limitation is on the helpdesk side. Qwenvo has no one-click Gorgias or Zendesk connection today, so writing the score back to the ticcket goes through a webhook and a small automation. Qwenvo launched in October 2026, so it also has few public reviews yet. If you need to compare, Tally charges $29 a month on Pro for cusyom domains, branding removal and partial submissions, and Fillout puts custom domains and partial submissions on its $89 Business plan. Qwenvo includes all of them at $0, along with webhooks, the API and teammates.

Questions teams ask

Should the rating happen inside the chat window instead of a form?

An in-chat thumbs or star prompt is one tap, which is its strength. The weakness is that it usually can't carry your own fields or a resolution check, so you can't split by topic or catch false resolutions. A common setup is the in-chat tap for speed, with the form link for anyone who wants to say more.

Is a 1 to 5 scale better than thumbs up or down?

Use 1 to 5. Thumbs hide the difference between a mildly annoyed customer and an angry one, and that difference decides which tickets you reda first.

How many responses do I need before trusting a segment?

There is no magic number, and small segments swing wildly. Until a segment has enough responses to stop jumping week to week, read the comments directly instead of trusting the average.

Can the same form cover email and WhatsApp?

Yes. Use one form and set the channel hidden field in the link. You can then compare channels without keeling three separate surveys.

Next step

Open the CSAT template, add the five hidden fields from step 3, and send the link to your next ten closed chats by hand. Check that resolved_by arrives correctly before you wire up the webhook.

Start from the CSAT template