Feature requests are sitting in TikTok comments. Here's how to mine them

Every feature-request tool you can buy collects feedback from people who already bought your product. The comment section under a popular TikTok video collects it from people who are still deciding, and they are much blunter. "Does it do X?" "I wish it..." "Why can't I..." Those are feature requests, filed by people who are not your customers yet.

Here is the method, run end to end on one category. We took the three most-commented smart ring videos on TikTok, pulled 50 comments from each, sorted all 147 into buckets, and counted them. The counts are below, along with the ten comments a product team would act on and the twenty-six we threw away.

How to mine feature requests from TikTok comments

1. Find the top videos for the category, not the brand. Search the category phrase, sort by most liked. Brand searches give you one company's fans; the category gives you people comparing.
2. Prefer comparison and review videos over brand announcements. Comment count matters more than views, but a launch post collects thousands of comments that all say the same thing. A comparison video with 180 comments beats it.
3. Pull 50 comments from each. Fifty is enough to see a theme repeat and small enough to read properly.
4. Sort every comment into one bucket. Request, complaint, question, praise, competitor mention, noise. One bucket each, no hedging.
5. Count, then look at what repeats. A theme raised twice or more by different people, especially in different comment sections, is a theme. Once is an anecdote.

We used the smart ring category on 19 September 2026. Three videos: a brand's launch announcement, a creator's product promo, and a creator comparing several rings she owns.

What 147 comments actually contained

Comments pulled from TikTok via TokConnect, 19 September 2026, and sorted by hand. Open the counts and the raw comments.

The launch announcement is almost useless. It had 5,654 comments, thirteen times the other two videos combined, and 27 of the 49 we read were the same sentence: I just bought last year's model. That is a pricing and upgrade-policy problem, which matters, but it is one signal repeated 27 times, not 27 signals.

The comparison video is where the product intelligence is. Every competitor mention in the whole set came from it, along with the most specific requests. People who are choosing between four products tell you exactly what would make them pick one.

Questions outnumber requests five to one, and most people never phrase a request as a request. Of the 49 questions, 21 asked whether the product does something ("does it have NFC", "does it track ovulation") and 28 asked where to buy it or which size to get. The first 21 are the feature requests. Somebody asking "does it do X" has told you they want X and cannot tell whether they can have it.

Ten comments a product team would act on

Quoted as written, without usernames. Open the snapshot. The full method for pulling comments in bulk is in the comment export guide.

Turning counts into a roadmap signal

A single comment is an anecdote. The same theme from unrelated people in unrelated comment sections is a signal. Group the requests and complaints by theme and count the people, not the likes.

Comment counts from our classification; search figures from TikTok search data pulled 19 September 2026.

Sizing is the clearest instruction in the set. Ten people raised it, and four and a half million weekly searches go to a sizing topic with no videos answering it. That is a product change and a content brief in the same finding: widen the size range, then make the video that explains how to pick one. The content gap guide covers how TikTok flags those empty topics.

The sleep-timing theme is the one worth pausing on. Two people in the same comment section said the same thing, one of them with two likes. No search data, no support ticket and no feedback portal would have surfaced it, because shift workers do not file tickets against products they decided not to buy.

What to throw away

Twenty-six of 147 comments were noise, and they clustered on the brand launch video, where fifteen of the forty-nine were riffs on the same cartoon joke. A meme does not become a signal because it got 793 likes.

• Jokes and memes. If the comment is funnier than it is specific, drop it.
• Parasocial replies to the creator. "You were my favourite creator as a kid" tells you about the creator.
• Praise with no object. "so cool", "nice apps". Eighteen comments landed here, and none of them changes a decision.
• Where-do-I-buy questions. Real intent, wrong document. Twenty-eight of them belong in a distribution note, not a roadmap.
• Anything about price alone, unless people name a competitor's price. Then it is positioning, and it counts.

Keep the raw file either way. A theme that looks like noise in one category turns out to be the whole story in another, and re-reading the original text costs nothing.

What Canny, Productboard and UserVoice cannot see

These tools are good at what they do. They just start after the sale.

Use both. UserVoice and Canny tell you which of your existing features to fix. The comment sections tell you which feature you never shipped is costing you the sale, and they do it before you have a customer base to survey.

If you are earlier than that and still choosing what to build, run the same category pass before you write code: it is check four of validating a startup idea, and it pairs with startup ideas, SaaS ideas, app ideas and the founder's guide to TikTok research.