Academy · 2026-07-27 · 6 min read
How to reduce duplicate feature requests
By Feedlark Team
Key takeaways
- • Duplicate requests are the single biggest reason vote counts understate real demand: the same idea gets split across three or four separate posts.
- • Manual duplicate-checking doesn't scale past a few dozen open requests. Teams either stop checking or spend hours a week on it.
- • AI-assisted duplicate detection matches requests on meaning, not exact wording, catching the paraphrased versions a keyword search misses.
- • A clear merge policy (who merges, how voters are told, what happens to their votes) matters as much as the detection technology itself.
Open any long-running feedback board and you'll find the same pattern: 'dark mode', 'add a night theme', 'option to switch to dark UI', and 'please darker colors' sitting as four separate posts with three votes each, when it's really one 12-vote request. Duplicates don't just clutter a board, they actively hide which features customers want most, because demand gets split instead of counted.
Why duplicates happen even on a well-run board
Customers don't search before posting, and they shouldn't have to guess the exact phrasing someone used six months earlier. 'Export to CSV', 'download my data', and 'let me get a spreadsheet of this' are the same request in three different vocabularies. A keyword search for 'export' misses the other two entirely, which is why manual dedup, scanning titles for similar words, catches maybe half of the actual duplicates on a busy board.
What AI-assisted duplicate detection actually does
Rather than matching on exact words, semantic duplicate detection compares the meaning of a new post against existing open requests, so 'download my data' and 'export to CSV' get flagged as likely duplicates even with zero shared vocabulary. Feedlark runs this check automatically when a request is submitted, surfacing the closest existing matches to the team (or the submitter) before a genuine duplicate goes live as a separate post.
| Manual review | AI-assisted detection | |
|---|---|---|
| Catches exact-wording duplicates | Yes | Yes |
| Catches paraphrased duplicates | Inconsistent, depends on reviewer memory | Yes, matches on meaning |
| Scales past 50 open requests | No, becomes a part-time job | Yes, runs on every submission |
| Speed | Minutes to hours per batch | Instant, at submission time |
A merge policy that keeps votes honest
- Decide who can merge: usually a PM or admin, not open to any voter, to avoid accidental or bad-faith merges.
- When two requests merge, combine the vote counts and notify voters on both threads so nobody's vote silently disappears.
- Keep the original wording of the more popular post as the canonical title; it's usually the clearer phrasing, which is part of why it attracted more votes.
- Merge early. A duplicate caught at 2 votes is a quick fix; the same duplicate caught at 40 votes on each side means an awkward retroactive merge and voter confusion.
“The cost of an unmerged duplicate isn't clutter, it's a wrong prioritization call. If your actual top request is split across three posts averaging 8 votes each, it looks like a mid-priority nice-to-have instead of the 24-vote request it really is.”
— Feedlark Team
Getting started without a big cleanup project
You don't need to audit your entire historical backlog in one sitting. Start by turning on duplicate detection for new submissions so the problem stops growing, then merge the handful of obvious existing duplicates you already know about. A full historical sweep can happen gradually, prioritizing your highest-traffic board first.
Frequently asked questions
- How do duplicate feature requests affect prioritization?
- They split demand for the same underlying idea across multiple posts, so each individual post looks lower-priority than the combined demand actually is. This is one of the most common causes of a team missing its own most-requested feature.
- Can AI reliably detect duplicate feature requests?
- Semantic matching, comparing meaning rather than exact wording, catches paraphrased duplicates that keyword search misses, and is a strong first pass. A human should still confirm before merging, since some requests are similar but not actually the same ask.
- What happens to votes when two requests are merged?
- The vote counts should combine, and voters on both original posts should be notified of the merge so they know their vote still counts and can track the combined request going forward.
- How often should a feedback board be checked for duplicates?
- Ideally at submission time, via automated detection, rather than in a periodic cleanup pass. Catching a duplicate early avoids the harder problem of merging two established posts with their own vote histories.
Feedlark Team. The Feedlark team builds and maintains the feedback, roadmap and changelog platform referenced in this guide.