Sentiment analysis is the process of sorting what people say about a brand, product or subject into positive, negative or neutral, usually with software, so you can see how opinion moves over time. It is applied to social posts, comments, reviews, forum threads and news coverage, and it is normally one layer of a wider social listening setup.
How sentiment analysis works
First, a tool gathers mentions: posts containing your brand name, product names, common misspellings and campaign hashtags, plus reviews and comments on your own channels. Each piece of text is then scored by a language model. Older tools matched words against lists (“great” is positive, “refund” is negative). Newer ones, including large language models, read the whole sentence and weigh context, which makes them noticeably better, though still far from perfect.
The scores are rolled up into a dashboard: the share of positive, negative and neutral mentions, a net sentiment figure, and trends by day or week. Better tools offer aspect-based sentiment, which separates opinions about different things in the same post. “Love the coat, the delivery was a shambles” is positive about the product and negative about fulfilment, and you need to know both.
British English is hard for these systems. “Not bad at all” is usually praise. “Brilliant, another parcel left in the rain” is a complaint. Understatement, sarcasm, regional slang and emoji all trip up the scoring, so the error rate on UK conversation can be higher than a tool’s sales demo suggests.
Why it matters for a UK business
Used well, sentiment analysis gives early warning. A rise in negative mentions about late deliveries tells you a courier problem exists before it shows up in refund figures. A spike after a campaign launch tells you whether the message landed or offended. Tracked against competitors, it shows where rivals are letting customers down, which is useful for positioning and for brand mention monitoring more broadly.
There is a legal side. Public posts written by identifiable people are still personal data under UK GDPR. Analysing them in aggregate to understand opinion of your brand is usually defensible under legitimate interests, provided you record that assessment, mention the activity in your privacy notice, keep only what you need and do not use the data to build profiles of individual people.
Common mistakes
- Reporting an overall sentiment score without reading any of the posts behind it.
- Drawing conclusions from tiny volumes. Twelve mentions in a month is a handful of opinions, not a trend.
- Treating neutral as good news. Often it means the tool could not decide.
- Mixing sources with different baselines, such as review sites (where unhappy customers are over-represented) and your own Instagram comments (where fans are).
- Watching sentiment fall without telling anyone who can fix the cause.
How to act on it
Start by deciding which questions you want answered: how customers feel about delivery, whether a price change has hurt goodwill, how a new product is landing. Set up the search terms to match, including misspellings and product nicknames.
Before trusting the dashboard, take a random sample of around a hundred scored mentions and mark them yourself. If you disagree with the tool on a large share, adjust its rules, retrain it where the tool allows, or treat its figures as rough direction only. Repeat the check every few months.
Then build a routine. Review sentiment by topic each week or month, pull out three or four real quotes that explain the numbers, and pass them to whoever owns the issue, whether that is operations, customer service or product. Sentiment analysis is only worth paying for if someone acts on it. Deciding which listening and reporting tools a business actually needs is part of my digital marketing strategy and consulting work.