Use case assessment
Is AI sentiment analysis worth building?
In the workplace this is prohibited, not merely high-risk. Elsewhere it's allowed and usually not very useful.
Emotion inference in workplace and education contexts is a prohibited practice under the EU AI Act, in force since February 2025. On customer feedback it is permitted and rarely tells you more than reading it would.
What it usually means
Scoring text — reviews, support tickets, survey responses, internal communications — as positive, negative or neutral, or inferring emotional state.
The prohibition, first
This is the rare case where the regulatory answer comes before the product one.
The EU AI Act prohibits AI systems inferring emotions of a natural person in workplace and education contexts, outside narrow medical and safety exceptions. This is a prohibition, not a high-risk classification — there is no compliance path that makes it permitted — and it has been in force since 2 February 2025.
That reaches further than most buyers expect. Engagement scoring in meeting tools. Sentiment on internal survey responses. Emotional tone analysis in performance or productivity software. These features exist, they are often on by default, and switching them on in an EU workplace is not a risk decision.
Where it is permitted
Customer sentiment on reviews, support tickets and public feedback is not covered by that prohibition.
Is it genuinely AI, and is it useful?
It is genuinely AI, and it is weaker than it looks. A three-way positive, negative, neutral score compresses a paragraph into a token and loses the thing you needed — what the person was unhappy about.
Sarcasm, mixed sentiment in one message, and domain-specific phrasing all degrade it. "This is fine" is not reliably classified by anything, including people.
What the simpler version looks like
Topic extraction rather than sentiment. Knowing that 40% of negative tickets concern the same billing flow is actionable. Knowing that sentiment is 62% negative is not.
And at low volume, reading them. A hundred support tickets is an afternoon and produces better insight than any aggregate score.
Where it sits under the EU AI Act
Workplace and education emotion inference: prohibited, in force since February 2025. Customer sentiment: not prohibited, not Annex III high-risk. Text being analysed is usually personal data, so GDPR applies regardless.
When it is worth it
At volume where nobody can read everything, as a triage signal for what to read — not as a metric to report.
Never on employees. And worth auditing the tools you already run, because this feature is switched on by default in more products than most organisations realise.
Common questions
Is sentiment analysis legal under the EU AI Act?
It depends on context. Inferring emotions of a natural person in workplace and education settings is a prohibited practice, in force since February 2025, outside narrow medical and safety exceptions. Customer sentiment on reviews and support tickets is not covered.
What counts as workplace emotion inference?
More than most buyers expect: engagement scoring in meeting tools, sentiment on internal survey responses, emotional tone analysis in performance software. These features are often on by default.
Is sentiment analysis actually useful?
Weaker than it looks. A positive-negative-neutral score compresses a paragraph into a token and loses what the person was unhappy about. Topic extraction is more actionable: 40% of negative tickets concerning one billing flow is something you can fix.
Scoring 40 AI use cases in a regulated enterprise — where a prohibited feature was already switched on in a tool nobody had reviewed.