How Do You Organize Help Center Content for Better User Experience?

Here's the full clean Version A ready to copy-paste:

Great question — and the short answer most teams land on is "not one of those three in isolation, but a blend." Categories, user intent, and product features each solve a different problem, so the trick is layering them rather than picking one.

Lead with user intent at the top level.

People rarely arrive at a help center thinking in terms of your product's feature map — they arrive with a job to do ("get set up," "fix this," "understand billing"). So your top-level navigation works best when it mirrors the customer journey: Getting Started, Using [core workflows], Troubleshooting, Billing & Account, Policies. Once someone is in the right intent bucket, organizing by feature or product area underneath keeps things findable and maintainable.

Write for AI and humans at the same time.

Your help center no longer exists just for the customer who manually searches — it fuels your AI agent's answers too. The same structure that helps a human browse is what helps an AI retrieve the right answer. A few practices that make content work for both, straight from Intercom's knowledge management guide:

• Scannable formatting: clear H1/H2/H3 hierarchy, lists, and tables. Avoid dropdowns or accordions that hide content — they hurt both human scanning and AI retrieval.
• Speak the customer's language: title and tag articles using the words customers actually search for (pull from your search logs), not internal jargon. Restate the question inside the answer, and never leave just a yes/no — expand on what "yes" means so an AI agent has full context.
• One job per article + a FAQ layer for bite-size answers: keep full articles focused, and roll up short, repetitive policy clarifications or edge cases into focused FAQ lists or snippets.
• Add text alongside images and videos: AI can't rely on visuals alone, so always include clear explanatory text. Bonus: it also improves accessibility.
• Audience targeting: if you serve different plans or personas, clearly note who each article is for. Most AI platforms (including Fin) support content targeting so the agent serves the right content to the right segment.

Content before configuration — and maintenance is not optional.

The flywheel only compounds if you treat knowledge management as a continuous function, not a one-off project. Practically that means:

• Dividing your article audit across product owners (at Intercom they divided 700+ articles into product areas and gave each team a week) so it's not one person's mammoth job.
• Building a log system so support reps — who spot gaps every day — can easily submit content requests.
• Making content creation part of every product launch checklist, not an afterthought once tickets start coming in.
• Setting a regular review cadence so nothing goes stale.

On discoverability specifically — structure alone doesn't make content findable. A few patterns that move the needle:

• Mine your failed searches first. Your help center's internal search zero-result queries are the single most honest signal you have: they show exactly what customers looked for and didn't find. Fix those before reorganizing anything else.
• Match article titles to search intent, not product names. "How do I cancel my subscription?" outperforms "Subscription management" as an article title every time. Customers search in questions and plain language, not feature taxonomy.
• Surface related content at the bottom of every article. Intercom's Help Center has a built-in Related Articles feature (enabled under Settings > Help Center > Configure and style, in the Article page section under the Styling tab) that automatically surfaces up to five relevant articles below each article body — no manual curation needed. It uses a relevance algorithm comparing titles, descriptions, and body content, and recalculates whenever an article is updated. One important caveat: only fully public, published articles that have been added to a collection will appear. Articles with audience targeting enabled will not be shown. So depending on your setup — particularly if you use content targeting for different plans or segments — the automated block may show limited or no suggestions for some articles. In that case, manually linking to related articles within the article body itself is a reliable fallback that works regardless of your targeting configuration. Also worth noting: Related Articles works on the web Messenger and Help Center website, but not on Mobile SDK — mobile users can still find content via Search and Browse, just without the recommendations block.
• Let content find the user, not the other way around. The best discoverability pattern is proactive: in-product tooltips, contextual help panels, and triggered messages that surface the right article at the moment of friction — before the customer even opens the help center. Intercom's proactive support features are built around exactly this pattern.
• Use data to validate structure. Search logs, article view counts filtered by "last updated," and — if you're using an AI agent — resolution rate by topic all tell you where content is missing, mislabeled, or too thin to be useful. Let that data drive your reorganization decisions, not intuition.

Short version: intent-based at the top, feature/category-based underneath, modular and consistently tagged throughout — and then treat discoverability as its own layer: search-optimized titles, related article links, proactive surfacing, and regular data review to catch what's falling through the cracks.