How to tag customer reviews by topic and mood

Why tag reviews at all?

Your average rating dropped from 4.6 to 4.3 this month. Was it the new supplier, the courier, or the week you were away and nobody answered email? The stars cannot tell you, and rereading two hundred reviews to find out is the job nobody gets round to.

A tag is a short label on each review that answers the question for you. Once every review says what it was about, you can count: twelve complaints about delivery this month, two about quality. That number tells you whom to call, and next month it tells you whether the call worked.

Which review tags should you use?

Use two kinds, and keep both short. Topics say what the review talks about. A list of five to eight works for most shops, as long as each topic points at someone who can fix it: the product itself, its quality, delivery, customer service, price, support. If a topic has no owner, nobody will act on its count.

Mood says how the customer felt, and it is worth keeping apart from the stars. People give four stars and still complain about the packaging, or give three stars to a product they like but find overpriced. A mood of “complaint” on a four-star review is exactly the signal an average hides.

Hannah's review is the kind that matters most: a good rating with a real problem inside. Filed under stars alone, it would never reach the person who chooses your packaging.

Should you tag reviews by hand, by keyword or with AI?

By hand works when you get a few reviews a week, and it makes you read them, which has its own value. Yotpo's article on reviews tagging, last updated on 7 April 2024, describes this: an Add Tag button on each review in its moderation screen, a set of ready tags and your own custom ones.

Keyword rules save time but miss how people actually write. Trustindex's guide to review tagging, published on 6 March 2024, tags a review automatically when it contains words you list, and skips it when it contains words you exclude. A rule for “damaged” will not catch “the box was smashed”, and a rule for “late” will catch “I'd buy it again later”. Keep keywords for names that do not vary, such as a staff member or a product line.

A language model reads meaning rather than words, so it handles both of those sentences, but it is still a guess you should check. When we chose the model for our own mood tags, we wrote 40 test reviews in two languages and labelled them by hand first; the model matched 39 of them, all 20 of the English ones. Those were reviews we wrote ourselves, so take the number as a reason to keep checking, not as a promise. Whatever does the tagging, look at a sample each month and correct the wrong ones.

What do you do with review tags once you have them?

Read the counts once a month and act on the biggest one. Forward every delivery complaint to whoever handles shipping, and every quality complaint to whoever buys stock. Reply to each complaint in public too, because the next shopper reads your answer; our guide to responding to negative reviews covers what to say.

Then use the same counts to check your fix. If you changed couriers on the first of the month, the delivery complaints should fall in the weeks after. If they do not, the courier was not the problem.

Shoppers do a rough version of this themselves. In BrightLocal's 2026 survey of 1,002 US adults, the factor most often named as making a review trustworthy, by 56%, was that other reviews backed it up with similar sentiment. One complaint about delivery is noise to them; five is a pattern, and they will find it whether or not you do.

Should tags change which reviews shoppers see?

No. Tags are a tool for you, and the safest place for them is your dashboard. Trustindex's guide shows the other use: in its widget editor you can pick tags, and only reviews with those tags appear in the widget. That is handy for a widget about one product line. It is also a quiet way to leave the complaints out. The US FTC rule on reviews forbids presenting the reviews you show as most or all of them while you hold some back for their stars or their negative tone. Our article on the FTC fake reviews rule explains what counts.

If you want shoppers to see what reviews are about, show a summary of all of them instead, written from every approved review, as in our article on AI review summaries.

How does Formgong tag reviews?

On the Pro plan, Formgong gives every review a mood, praise, neutral or complaint, and up to six topics: product, quality, delivery, service, price and support. A rating with no text gets its mood from the stars; written text goes to a model after the review is saved, so the reviewer never waits for it. The instruction treats a review as a complaint when the customer is unhappy about anything, even if the rest is positive, which is how Hannah's review above ends up there. You can change any mood by hand, and nothing overwrites your choice. The dashboard can list complaints first and filter by topic, with a count next to each topic. Tags never leave the dashboard: on our demo shop Glyna, read on 3 November 2026, each review in the public data carries 8 fields and none of them is a mood or a topic, so the widget cannot be filtered by them. That is the difference from the Trustindex tag filter above. The trade-off is that you cannot add your own tags, which Yotpo's tagging does allow.

Our own numbers show how new this is. The 153 reviews still held in our main database from before review sites moved to their own stores all have a mood, 146 from the model and 7 from stars, and no owner has corrected one yet, so we have no accuracy figure on real reviews. When we checked on 3 November, none of the 153 had a topic: topics were added on 26 October, and reviews that already had a mood by then were not revisited. We fixed that the same day, so older reviews now get topics too. The AI features docs describe each feature.

How do you sort your reviews?

If you tag reviews today, by hand or otherwise, tell us which tags you actually use at support@formgong.com. It will shape whether we add custom ones.