
“Hard to open” could describe the lid, the latch or even the box the product arrived in. An AI model can group reviews by the feature and experience they describe. To turn those groups into a design brief, someone must return to the original words, identify the task that went wrong and decide what a better prototype would let a person do.
From review text to a feature
A 2025 product-design study used natural-language processing and reviews of a target 3D printer and competing products to identify attributes, compare problems and develop redesign options. The case demonstrated a route from comments to a proposed design. It did not make everyone who bought the printer equally likely to appear in the reviews. Study of product redesign using online reviews.

Keep the quote attached to the inference
Keep a short, de-identified excerpt with each candidate theme. Separate observations about a feature from opinions about delivery, price or customer service. A single review can cover two features. Several copied reviews may represent one original complaint. These distinctions change both the counts and the design question.
Suppose repeated comments say users lose their grip while opening a container with wet hands. “Make the handle better” is too loose. A team could compare slips and opening time for two handles in a defined wet-hand task. The size and placement of the grip are design hypotheses, and more than one change may address the same complaint.

What would establish an improvement?
Use the current product as a comparison, vary which design participants try first and record task completion as well as preference. Include people for whom the original task is difficult. A small usability trial can reveal a failure mechanism even if it cannot estimate its frequency among all customers.
AI can help organize the resulting notes. Keep measurements and author-verified quotations alongside its summary so the next designer can see where an inference came from. For a printer-specific example of evaluating the resulting physical object, see How AI Spots an Extrusion Error While a Part Is Printing.

Try it with the evidence
Turn three comments into a fair test
Source notes. Illustrative comments about a container. A says “my wet hand slipped on the latch.” B says “the latch was fine but the packing tape was hard to open.” C says “I can open it with two hands but need one when cooking.” These are three comments, with no purchase denominator.
Prompt. Extract the feature and task for each comment. Write one prototype comparison with a measurable outcome. Mark any conclusion that would require evidence from a broader group of users.
Check your reading
A concerns wet grip on the latch, B concerns shipping packaging, and C concerns one-handed operation. Compare successful openings and slips for two latch shapes in a defined one-handed task, with order varied across participants. The three comments cannot establish the prevalence of any complaint.
Illustrated field notes
People, feedback and design decisions






