What is a customer feedback taxonomy?

Most taxonomies are hierarchical. A top level (often called L1) holds broad themes like Performance, Payments, or Onboarding. Beneath each theme sit subcategories specific enough to act on: Payments might split into failed transactions, duplicate charges, and refund delays. High-volume areas sometimes add a third level.

Each incoming verbatim is assigned to one or more categories. Historically that meant keyword rules or manual tagging. Modern platforms use AI models, typically large language models, to read each message and classify it, which handles slang, typos, and mixed-topic feedback far better than keyword matching. The best AI taxonomies are validated against large real-world corpora; unitQ’s, for instance, is sharpened against a benchmark of 67.7M+ app reviews.²

The part teams underestimate is maintenance. Products change, so taxonomies drift. A category built for last year’s checkout flow will quietly misfile feedback about the new one, and anything the system cannot place lands in an uncategorized bin. Healthy taxonomies get reviewed on a cadence: merge overlapping categories, retire stale ones, and add categories when a new feature starts generating its own complaint patterns.

Three things become possible once feedback is categorized consistently.

The failure modes are just as instructive. A mega-bin category like “General issues” that absorbs a third of all feedback tells you nothing. Overlapping categories split one real problem across several small counts, hiding its true size. Both are taxonomy-design problems, not analysis problems.

Tagging and taxonomy get used interchangeably, but they behave differently at scale. Tags are ad hoc labels anyone can add, so a busy account accumulates synonyms and near-duplicates (“login bug,” “can’t log in,” “sign-in issue”) that fragment the same problem across several small counts. A taxonomy is a governed system: categories have definitions, owners, and a place in the hierarchy, so the same issue counts the same way every time.

The practical test is whether two people would file the same verbatim the same way. Free-form tags usually fail that test; a maintained taxonomy is built to pass it, which is what keeps the numbers comparable across teams and over time.

A food delivery app receives this review: “My driver never showed up but I was still charged.” A good taxonomy files it under two categories, delivery reliability and refunds, because both teams need to see it. Over the next month, the refunds category climbs sharply while delivery reliability holds flat. That divergence points the investigation at the refund pipeline, not the courier network, and saves days of guessing.

Platforms like unitQ Monitor build and maintain taxonomies with AI across every feedback channel, but the underlying principles are the same whether your taxonomy lives in a quality intelligence platform or a carefully tended spreadsheet.¹