Publishing workflow
Write, review, and publish articles with draft states and scheduled visibility.
Self-service knowledge base
Help customers help themselves Articles, portal, and deflection
Publish a branded knowledge base that powers your customer portal, reduces ticket volume, and feeds AI deflection across chat and email.
Last updated on 2026-07-08
Written by Malik S
Reviewed by Gandral V
Semantic search
Customers and agents find answers by meaning, not just keywords.
Article analytics
See which articles deflect tickets and where content gaps remain.
Overview
A knowledge base earns its place when it answers the questions your agents answer twice a day. Helpefi publishes articles to a branded portal, feeds them to AI deflection on chat and portal search, and shows which ones actually stop tickets from being created. Content you can measure is content you can prune.
Writing articles that deflect tickets
Start from real tickets rather than a taxonomy. If five customers asked the same billing question last month, that question is the article title. Write the answer in the first two sentences, then add steps, screenshots, or limits. Keep one question per article so search matching stays precise and an update does not require rewriting an entire guide. Articles that try to answer six questions rank for none of them and help nobody.
Collections, visibility, and multi-brand content
Collections group articles by product, audience, or client. Visibility settings decide whether a collection appears on the public portal, only in the agent workspace, or only inside a specific brand. Internal runbooks belong in an agent-only collection so AI can cite them for staff without exposing them to customers. Each brand portal reads its own collections, which keeps a client from seeing another client's policy language.
Semantic search versus keyword search
Keyword search fails when a customer describes a problem in their own words. Semantic search matches meaning, so a query about a card being declined can surface an article titled payment authorization failures. Agents get the same matching inside the ticket view, which is often where the real time saving appears. Both modes stay grounded in published content, so a weak match returns a shorter list rather than an invented answer.
Draft, review, publish, archive
Articles move through draft, review, and published states, with a named reviewer before anything customer-facing goes live. Scheduling handles launches that should not appear early. Archiving matters as much as publishing: outdated content is worse than missing content because it produces confidently wrong answers. Review analytics monthly, archive what no longer matches the product, and keep the top twenty articles accurate rather than growing the library indefinitely.
Measuring deflection honestly
Deflection is only meaningful when you can see what happened next. Helpefi tracks article views, searches that ended without a ticket, and searches that ended with one. The second number is the useful one: a search that leads straight to a ticket is a content gap, not a success. Read CSAT on self-service threads and reopen rates on deflected tickets before claiming a percentage. A smaller, accurate library beats a large one that customers abandon.
Use case: SaaS reducing repeat questions
Publish articles for the fifteen most common support intents and let portal deflection absorb the repeat questions.
Use case: Ecommerce peak season
Shipping and returns articles go live before the sale, so holiday volume does not land entirely on agents.
Use case: Agency with per-client content
Separate collections per client, each with its own portal and AI grounding, keep policy language from crossing brands.
Use case: Internal IT self-service
Password reset, VPN setup, and hardware request articles run in an agent-visible collection for employee deflection.
Use case: A startup building from scratch
Begin with twenty articles on the most common questions, then let search analytics decide what to write next.
A smaller, accurate library wins
Publish twenty articles that answer real questions, measure which ones end a search without a ticket, and archive the rest. Accuracy beats volume, and search analytics tell you which article to write next.
Frequently asked questions
- Can we restrict articles by audience?
- Yes. Collections can be public, agent-only, or limited to a brand. AI grounding follows the same visibility rules as human readers.
- Does the knowledge base work with AI?
- Articles are the grounding source for Copilot and portal deflection. Admins choose which collections AI may cite.
- Can the portal use a custom domain?
- Yes, on Professional and Enterprise. Point a CNAME at your workspace subdomain and the portal serves on your own hostname.
- What content formats are supported?
- Rich text with headings, lists, tables, callouts, images, and embedded video links. Attachments can be added to an article or a ticket.
- Can agents write articles?
- Agents can create drafts from inside a ticket, which is the fastest way to capture an answer while the context is fresh. A reviewer publishes it.
- How do we migrate existing articles?
- Export from your current tool as HTML or CSV and import into collections. Internal links and images usually need a pass before publishing; plan a review rather than a straight copy.
- Which plans include the knowledge base?
- The knowledge base is included on every plan. Custom portal domains and higher storage limits depend on tier.
References & sources
-
Helpefi on G2 →
User reviews and ratings for Helpefi Helpdesk on G2.
Launch your knowledge base
Reduce repeat questions and give customers 24/7 self-service without adding headcount.
Start free trial