Chatbot CSAT Calculator

How the CSAT math works

The score itself is the easy part. CSAT = positive responses ÷ total responses, where «positive» means the top two boxes on a 5-point scale (a 4 or a 5) or a thumbs-up on a binary survey. If 310 of your 380 respondents rated the bot positively, your CSAT is 81.6%. The full definition, healthy ranges, and scale variants are covered in our chatbot CSAT glossary entry.

The part dashboards skip is uncertainty. 380 responses is a sample, and a sample carries sampling error. This calculator wraps your score in a Wilson 95% confidence interval: the range where your true satisfaction rate plausibly sits. At 310/380 that range is roughly 77% to 85%: wide enough that a «drop» from 82% to 79% week-over-week means nothing at all. We use the Wilson interval rather than the textbook normal approximation because it stays honest at small sample sizes and at scores near 100% — exactly where chatbot surveys live.

The third number is the response rate: responses received ÷ surveys shown. It is context, not decoration. A 90% CSAT built on 8% of users answering is a different fact than a 90% CSAT from half of them, because the people who bother to answer skew toward the extremes. Under roughly 15% response rate, the calculator flags the score as directional rather than precise — self-selection bias at that level can move the number more than the confidence interval suggests.

What counts as a healthy chatbot CSAT?

Our editorial consensus, kept consistent across the glossary entry and our platform reviews: treat 80% positive (roughly a 4.0/5 average) as a floor to defend, not a number to maximize. A bot can buy CSAT points in ways that hurt the business — answering only the easy questions and silently dumping the hard ones on your team, or making the human handoff so pleasant that people rate the escalation experience rather than the bot. That is why the verdict card above is deliberately phrased as «defensible» rather than «good»: CSAT only means something read next to containment, escalation rate, and response rate at the same time. Our chatbot metrics guide covers how the four fit together.

The bot-vs-human comparison, done honestly

The most common misuse of CSAT we see in operator communities is the panicked comparison: «our human agents get 86%, the bot gets 82%, the bot is failing.» Maybe. But two percentages from two samples of different sizes cannot be compared by eyeballing them. The optional comparison above runs a pooled two-proportion z-test (the standard test for «are these two rates actually different?») and reports a p-value. At 380 bot responses and 540 human responses, an 86%-vs-82% gap is not statistically significant: the honest conclusion is «we cannot tell yet,» not «the bot is worse.»

One structural caveat the test cannot fix: bot and human conversations are not the same conversations. Bots absorb the short, repetitive questions; humans get the angry edge cases the bot escalated. Even a real, significant gap may say more about conversation mix than about quality. Use the test to rule noise out — then read transcripts before assigning blame.

Common CSAT mistakes

1. Reacting to moves inside the margin of error

If your margin of error is ±4pp, a week-over-week move from 81% to 84% is not a trend — it is weather. The sample-size table in the calculator tells you how many responses you need before a ±2pp or ±3pp move is worth a standup discussion. Until you have that volume, aggregate over longer windows.

2. Surveying only resolved conversations

Some platforms trigger the survey only when the bot marks a conversation resolved. That silently deletes your unhappiest users from the denominator — the people who rage-quit mid-flow never see the survey. The score goes up; the truth does not. Survey every conversation at a consistent trigger point, including abandoned and escalated ones, and keep the trigger identical when comparing bot and human segments. Our CSAT survey-design guide covers placement, scale choice, and the follow-up question in detail.

3. Treating CSAT as the loyalty metric

CSAT measures one interaction, right after it happened. It does not tell you whether the customer would recommend you or plans to churn — that is NPS territory, measured on a different cadence with different mechanics (see our guide to running NPS through a chatbot). Keeping the two apart prevents the classic dashboard mistake of celebrating high transactional CSAT while relationship health quietly erodes.

4. Optimizing the score instead of the failure modes

A below-floor CSAT is almost never uniformly distributed. In practice a handful of failing intents drag the aggregate down while the rest of the bot performs fine. Before touching the survey, segment negative responses by intent or topic — the fix is usually rewriting three flows, not redesigning the bot. If deflection is high while CSAT is low, you are measuring the exact trade-off described in our deflection-vs-containment entry: the bot is keeping people away from agents without actually helping them.

Related Chatbotscape tools and resources

• Chatbot ROI calculator — the business-case companion to this quality metric
• How to improve chatbot CSAT — survey design that does not poison the number
• Chatbot metrics guide — CSAT, containment, escalation, and deflection as one system
• Intercom review — how a helpdesk-native platform reports CSAT
• Tidio review — built-in post-chat ratings on the SMB tier
• SendPulse review — survey flows across messenger channels

FAQ

What is a good survey response rate?

For post-chat surveys, 15-40% is the workable range we see across helpdesk platforms; one-tap surveys embedded in the chat window land at the top of it, emailed surveys at the bottom. Below 15% the calculator flags the score because self-selection starts to dominate. If your rate is low, fix placement first: a single thumbs-up/down inside the conversation outperforms any emailed questionnaire.

Why a Wilson interval instead of the usual ± formula?

The textbook Wald interval (p ± z·√(p(1−p)/n)) misbehaves exactly where CSAT lives: small samples and scores near 100%. It can produce intervals extending past 100% and collapses to ±0 when every response is positive — implying impossible certainty from ten happy respondents. The Wilson score interval handles both cases correctly, which is why it is the standard recommendation for proportions like this.

My platform reports CSAT as an average score, not a percentage.

Same data, different summary. Platforms reporting «4.4/5» are averaging the scale values; percentage CSAT counts the share of 4s and 5s. To use this calculator, pull the response distribution and count 4s + 5s as positive. Top-2-box is the more robust summary — one bitter 1-star drags an average far more than it drags a proportion.

Does the calculator send my data anywhere?

No. All math runs in your browser — nothing you type is uploaded, stored, or logged. Refresh the page and the numbers are gone.

Can I embed this calculator on my site?

Yes — free. Copy the iframe snippet from the embed section above. The embed strips Chatbotscape navigation and keeps the calculator plus attribution.