Help docs are the post-sale experience layer. Positioning convinces prospects to buy; help docs determine whether they succeed after buying. Treats documentation as a strategic deliverable.
Source synthesis: Tyner Blain (task-based docs), KCS / Consortium for Service Innovation (Knowledge-Centered Service double-loop), Jakob Nielsen (progressive disclosure), help-center teardowns of Intercom, Notion, Stripe, Aircall, HubSpot, Zendesk.
Steps
1. Build product knowledge. Crawl existing docs, read reviews, explore the UI. Output: a verified Product-Knowledge file with features, UI labels, plan tiers, capability descriptions. Every claim in every article must trace back to it.
2. Define audience. Primary personas, technical sophistication, situations driving them to docs (onboarding, troubleshooting, feature discovery, configuration). Litmus test: "Would a sales rep at a 10-person plumbing company understand this?"
3. Audit existing docs. Inventory by type (how-to, conceptual, troubleshooting, reference). Score on accuracy, customer-centricity, structure, completeness. Prioritize rewrites by ticket volume + page views. Output: P0/P1/P2 audit.
4. Map customer journey to content needs. Sequence by first-week progression: setup → core concept → core action → power feature → edge case. Five questions per feature: What is it? How do I do it? How do I undo it? Who can do it? What happens after?
5. Design taxonomy. Hybrid task-first model. Task-oriented top-level categories ("Make and receive calls") with feature-named articles inside ("Call forwarding settings"). 3 levels max (Category > Section > Article). Cap top-level at 5–7. Miller's number for items per level.
6. Apply category naming. Verb-first labels for action categories ("Set up your phone system"). Plain nouns for reference categories ("Account and billing"). No acronyms or product jargon at the category level.
7. Design getting-started experience. Dedicated top-level category, listed first. Role-based quick-starts (admin setup, end-user onboarding). Conceptual overview. 5-minute quick wins. Optionally a narrative onboarding sequence: Motivate → Setup → First Value → Power Use.
8. Article architecture templates. Per type: how-to, onboarding narrative, conceptual explainer, integration doc, troubleshooting, reference. Every template has title, one-line outcome, prerequisite block, body, related articles.
9. Apply transformation rules. Product-to-outcome rewriting ("The filter dropdown lets you select Unread only" → "See which customers haven't replied yet"). Instructions over descriptions. Buyer-as-agent framing.
10. Progressive disclosure. One-line answer / TL;DR first. Step-by-step second. Edge cases and admin-only options in expandable sections.
11. Multi-platform handling. Separate parallel articles when workflows diverge substantially. Inline tabs/callouts when 1–2 steps differ. Identical structure across parallels. "Click" desktop, "Tap" mobile.
12. Cross-linking map. Plan inbound + outbound links per article BEFORE writing. Every article has breadcrumbs, ≥1 inline body link, Related Articles block (3–5 curated), Next Steps prompt for sequenced setups. Concept-action pairs always cross-link.
13. Quality review. Product accuracy, customer centricity, structure, copy craft.
14. Search optimization. Tags + synonyms per article. "Set up your phone menu" tagged "IVR" so technical users find it.
15. Measurement and iteration. Article views, zero-result searches, ticket deflection rate, time-on-page. KCS Solve Loop captures new knowledge from tickets; Evolve Loop refines existing content.
Frameworks
- Hybrid taxonomy: task-oriented top + feature-named articles inside. 3 levels max.
- Progressive disclosure (NNGroup): essentials upfront, complexity behind deliberate user actions.
- KCS double-loop: Solve Loop creates, Evolve Loop refines.
- Narrative + reference layer: guided story for the happy path; modular reference for individual tasks.
- One article, one task: if it needs a second verb, split it.
- Question-as-title: "How do I...?" / "What is...?" / "Why am I...?"
- Five question types per feature: What / How / Undo / Who / After.
Quality gates
- Every noun matches its UI label exactly.
- No unverified claims. Every product claim traces to Product-Knowledge.
- Article intros lead with customer outcome, not feature description. No sentence starts with the product name.
- Buyer is the agent.
- Instructions beat descriptions.
- 3 levels max with 5–7 top-level categories.
- Every article has 3 exit paths (breadcrumbs, inline link, related). No dead ends.
- Progressive disclosure applied.
- Numbered steps with bold UI labels.
- Prerequisites in the first two lines.
- Multi-platform parity. Consistent verbs.
- Plumbing-rep test passes on every category and section name.
Common failure modes
- Feature-first organization ("IVR Configuration") instead of customer-problem categories ("Set up your phone menu"). Persists because easier to write, not because better for users.
- Engineering vocabulary describing how features work internally.
- More than 3 levels of nav. Each extra level cuts findability by ~20%.
- Missing cross-links. Dead-end articles.
- Claiming features not verified. A wrong claim creates a ticket; "may be available" is worse — delete.
- Descriptions instead of instructions.
- Product jargon at the category level.
- One article doing many tasks.
- Buried prerequisites.
- Auto-generated related-articles blocks. Curate.
- Treating taxonomy as one-time. Track zero-result searches and ticket themes.
- One-sentence answers that leave users stuck.
Outputs
1. Taxonomy map (3 levels max).
2. Article architecture templates.
3. Content audit with P0/P1/P2.
4. Cross-linking map.
5. Produced articles per templates.
6. Search tag inventory.
7. Measurement plan.