Every resolved support ticket is a help article waiting to be written. Paste the conversation text into HelpGuides and the AI extracts the problem and resolution, suggests a title and format, and writes a clean article — no mention of the original customer or agent.
To get started, go to Editor → Articles and click Generate from conversation in the prompt at the top of the page.
Step 1 — Paste the conversation
On the first screen, paste the raw conversation into the text box. Any format works:
- Support chat transcript (copied from Intercom, Zendesk, Crisp, etc.)
- Email thread
- Slack message or thread
- A plain-text description of the issue and how it was fixed
The text must be at least 50 characters. The AI reads up to 10,000 characters — longer threads are trimmed automatically.
Tip: Remove customer names, email addresses, and phone numbers before pasting. The AI will warn you on the next screen if it detects personal data, but removing it upfront is faster.
Additional context (optional)
The Additional context field lets you name the product area the conversation covers. This sharpens the AI's understanding of what the article should be about.
Examples: billing settings · invite flow · export feature · two-factor authentication
Once you're ready, click Analyse Conversation.
Step 2 — Review the analysis
The AI reads the conversation and extracts the key information. You'll see a review screen with every field editable — click any field to change it before generating.
| Field | What it is |
|---|---|
| Issue identified | The core problem from the conversation |
| Resolution | The fix or answer provided |
| Suggested title | A user-facing article title |
| Category / product area | The feature area (edit to match your categories) |
| Article format | How the article will be structured (see below) |
Article formats
| Format | Best for |
|---|---|
| Step-by-step guide | Conversations where the resolution involved a sequence of actions |
| FAQ | Short question-and-answer exchanges |
| Conceptual explanation | "How does X work?" conversations |
| Troubleshooting guide | Error reports or unclear root causes — structures as Symptoms → Causes → Solutions → Prevention |
The AI chooses a format automatically, but you can change it here.
Warnings to watch for
Sensitive data detected — shown if the AI finds email addresses, phone numbers, or full names in the text. Go back and scrub the conversation before continuing.
Low confidence — the conversation was ambiguous. Review the extracted fields carefully before generating.
No resolution found — the conversation didn't include a fix. You have two options:
- Draft troubleshooting article — the AI writes around the problem, leaving a placeholder for the solution you can fill in later.
- Mark as known issue — the article notes that your team is investigating and will update the article when a fix is available.
Once you're happy with the fields, click Generate Article.
Step 3 — Article is written and opened in the editor
Generation takes 10–15 seconds. You'll see a live progress indicator as the AI reads the conversation, identifies structure, and writes the content.
When it finishes, the article opens in the editor pre-filled with:
- Title
- Full markdown content, structured by the chosen format
- Description (excerpt)
- Suggested tags
The article is saved as a draft. Review, edit, and publish when you're ready. Every generated article ends with a Still need help? section pointing readers to your support team.
Best practices
Only use resolved conversations. The AI writes better articles when a clear resolution exists. Unresolved tickets produce thin drafts that need significant editing — your time is better spent writing those from scratch.
One issue per generation. Multi-topic threads confuse the extraction. If a conversation covers two separate problems, split it and generate twice.
Strip PII before pasting. Names, email addresses, and account IDs add noise and shouldn't end up in a public article. Delete them from the conversation text before submitting.
Use the context field. Even just a product area name — billing, onboarding, permissions — helps the AI write step descriptions that match your product's language and structure.
Edit the confirmation screen. The extracted title and product area are often a good starting point, not a finished result. A quick review saves editing time after generation.
Change the format if needed. The AI guesses, but it doesn't always get it right. A step-by-step conversation about settings might be better as a conceptual explanation — switch it before generating rather than rewriting the article.
Treat the output as a first draft. Generated articles are clean and structured, but they won't use your exact product terminology or know about plan restrictions and edge cases. Add those before publishing.
Last updated: June 6, 2026