Sentiment without drivers is a vanity metric. Classify each response and extract the reason behind it, so you know not just how people feel but why — with one prompt that grades the result.
In short: A sentiment split on its own — 60% positive, 30% neutral, 10% negative — tells you the temperature but not the cause. Useful sentiment analysis classifies each response and extracts the driver behind it, then reports the split alongside the top three drivers per sentiment, each with a quote. One prompt does this and grades the result green, amber, or red.
Point the Sopact Sense Assistant at your dataset so it works from clean records with persistent contact IDs, then have it load your Decision Brief first. Pairing each sentiment label with a driver is what turns a vanity percentage into something you can act on.
You are the Sopact Sense Assistant working over the DEMO-04 · Open-Text Barriers dataset (clean data + persistent contact IDs). Load my Decision Brief (decision, audience, outcomes, indicators, evidence standard) first, then wait for my task.
For the barriers question, classify each response's sentiment (pos/neu/neg) and extract its driver; report split + top 3 drivers per sentiment with quotes. Grade green/amber/red.
Five elements make this prompt work: it runs over your dataset; it pairs sentiment + driver on every response; it makes the split reconcile so the percentages add to 100; it flags driverless responses that have a label but no reason; and it ends with a grade of green, amber, or red.
Run on the Open-Text Barriers dataset (DEMO-04) already loaded in Sopact Sense.
GRADE: green | Driver named | reason clear; amber | Negative logistics | cause vague; red | Driverless | label, no reason
A green response carries a clear, named driver behind its sentiment. An amber negative points at logistics but the cause is too vague to act on. A red response is scored for sentiment with no driver extracted at all — a label with no reason underneath it.
Take the lowest-graded element above and fix it using only what the program could realistically measure. Show the before → after grade and the single indicator/edit that moves it to green.
Create a 'missing & incomplete' report from this analysis in Sopact branding [or paste your website URL / brand guideline to apply your own]. List every element graded amber or red, what is missing, and the one input that fixes each. Lead with the decision this report informs.
Create a shareable link for this report and open it in a new tab.
Always pair sentiment with a driver. The label tells you the mood; the driver tells you what to change. A sentiment split with no drivers is the definition of a vanity metric.
Make the split reconcile. Positive, neutral, and negative should account for every response and add to 100%. If they don't, some responses were dropped or double-counted.
Sharpen vague negatives. 'It was hard' is amber until you know what was hard. Re-prompt the assistant to pin each negative to a specific, fixable cause.
Treat driverless responses as a gap. A sentiment score with no extractable reason is the most common red — re-prompt for a one-line reason rather than guessing it.
List every response scored for sentiment but missing a driver, and suggest the single follow-up prompt that would recover a reason for each.
Classify each response as positive, neutral, or negative, then extract the driver — the specific reason — behind that sentiment. Report the overall split together with the top three drivers per sentiment, each backed by a quote, so you know both how people feel and why.
Because a percentage tells you the mood but not the cause, and you cannot act on a mood. The driver behind each response is what points to a specific, fixable change — which is why sentiment without drivers is a vanity metric.
Flag them as driverless rather than inventing a reason. The one-fix prompt will name the single follow-up that recovers a driver for each, so next cycle every scored response carries a reason.
Open Sopact Sense, paste your program description, and put it to work.
Try in Sopact
