L'Analyseur de Résultats d'Enquête : Transformez des Réponses Brutes en Thèmes, Sentiment et Actions
Pourquoi ce prompt est important
Manually reading and coding 340 open-ended responses takes a trained analyst 8-12 hours, and most teams skip that step entirely, reporting only the multiple-choice scores while ignoring the free-text answers that usually contain the most actionable insight. Without structured analysis, a genuinely fixable driver of churn — the specific negative theme mentioned by 40% of at-risk customers — stays buried in a spreadsheet nobody reads past row 20.
À quoi nous l'utilisons
You just closed a customer satisfaction survey with 340 open-text responses to "What would make you more likely to renew?" and a board meeting in two days where you need to present findings, not raw data.
Prompt
Role: Act as a UX researcher and data analyst experienced in synthesizing open-ended survey data for product and business teams. Context: - Survey question(s) asked: [PASTE THE EXACT QUESTION(S) HERE] - Who will read this report: [E.G. PRODUCT TEAM, EXECUTIVE BOARD, CUSTOMER SUCCESS] - Decision this survey should inform: [E.G. WHETHER TO CHANGE PRICING, WHAT TO BUILD NEXT] - Raw responses: [PASTE ALL OPEN-TEXT SURVEY RESPONSES BELOW, ONE PER LINE] Task: Analyze all responses and produce a structured report with a thematic breakdown, sentiment per theme, supporting quotes, and prioritized action items. Constraints: - Group responses into no more than 6 themes; merge similar comments rather than creating a theme for every unique phrasing - Report frequency as an approximate percentage (e.g. "mentioned by ~30% of respondents"), not an exact count, since exact counts overstate precision on small samples - Sentiment per theme must be exactly one of: Positive, Negative, Mixed, Neutral — not a descriptive phrase - Flag any theme where sentiment is Negative AND frequency is high (top half of themes by frequency) as "Priority" - Include 2-3 verbatim quotes per theme (lightly edited only for typos), not paraphrases - Do not invent or infer a theme that isn't actually present in the responses — if you're unsure whether something is a real pattern or a single outlier, say so explicitly rather than including it as a theme - Tone for the report: [YOUR PREFERRED TONE: e.g. clinical / conversational] Output Format: 1. Executive summary (3-4 sentences) 2. Theme table: Theme | Frequency | Sentiment | Priority flag 3. Quotes section, grouped by theme 4. Prioritized action items, ranked, each tied to a specific theme
Résultat
Executive Summary: Across the sampled responses to "What would make you more likely to renew?", three themes emerged clearly: pricing/value concerns (mentioned by roughly half of respondents), a desire for better onboarding support (about a third), and requests for a mobile app (about a quarter). Pricing sentiment is strongly negative and should be treated as the top priority.
| Theme | Frequency | Sentiment | Priority |
|---|---|---|---|
| Pricing / Value | ~50% | Negative | 🔴 Priority |
| Onboarding Support | ~33% | Mixed | Medium |
| Customer Support Speed | ~17% | Negative | Medium |
| Mobile App Request | ~25% | Neutral | Low |
Pricing / Value — Quotes:
"I like the product but I genuinely can't justify $200/month for what we actually use."
"Feels like we're paying for features three tiers up that we'll never touch."
Onboarding Support — Quotes:
"First two weeks were confusing — a guided setup call would have saved us a lot of back-and-forth."
"Once we figured it out it was great, just wish someone had walked us through it."
Prioritized Action Items:
- Pricing / Value (Negative, ~50%): Pilot a tiered or usage-based pricing option for the segment citing feature bloat — the single highest-leverage fix given both frequency and sentiment severity.
- Customer Support Speed (Negative, ~17%): Audit response-time SLAs; a modest improvement here directly addresses a named renewal blocker.
- Onboarding Support (Mixed, ~33%): Pilot a guided onboarding call for new accounts next quarter — lower urgency than pricing but affects a larger share of respondents.
- Mobile App Request (Neutral, ~25%): Log as a roadmap feature request — neutral sentiment suggests a nice-to-have, not a renewal blocker.
L'échec le plus courant dans l'analyse d'enquêtes générée par IA est la prolifération de thèmes : demandez à un modèle de classer 300 réponses en texte libre sans limite, et il produira souvent 15 ou 20 micro-thèmes qui ne sont en réalité que des paraphrases les uns des autres. Ce prompt corrige cela avec un plafond strict de six thèmes.
Le sentiment est limité à quatre étiquettes fixes — Positif, Négatif, Mixte, Neutre — plutôt qu'un texte descriptif ouvert, car une colonne de tableur indiquant "généralement frustré mais reconnaissant des correctifs récents" ne peut être ni triée ni filtrée. La règle du drapeau Priorité (sentiment Négatif plus fréquence élevée) effectue le vrai travail de priorisation.
L'exigence de 2-3 citations textuelles par thème existe car les résultats résumés d'enquête sont régulièrement remis en question en réunion — quelqu'un demande toujours "puis-je voir un exemple réel ?" — et disposer du langage réel du client, non paraphrasé, règle cette question immédiatement.
La contrainte contre l'invention de thèmes non réellement présents dans les réponses compte plus qu'il n'y paraît : les modèles de langage invités à trouver des motifs dans des données ambiguës font parfois émerger un thème plausible qui est en réalité une extrapolation, pas quelque chose que les répondants ont réellement dit.