What it does
The incident detail page has a Draft with AI button (Sparkles icon, Business plan only). Click it and Signalog drafts four structured fields on the postmortem form from the incident timeline:
- Summary: 2-3 sentence narrative of what happened
- Root cause: best-guess root cause from the timeline (acknowledges uncertainty)
- Impact: who was affected, severity, duration
- Action items: concrete follow-ups as a markdown bulleted list
The AI never auto-publishes. Every draft lands in an editable form for you to review, refine, and publish. We treat the AI as a writing assistant, not an autonomous author.
Pricing and availability
AI drafting is included on the Business plan with reasonable monthly limits. There’s nothing to configure on your end. Signalog manages the AI infrastructure, billing, and capacity. If you hit the monthly limit, contact us and we’ll work with you on enterprise pricing.
| Plan | AI Drafting |
|---|---|
| Free | — |
| Pro | — |
| Business | ✅ included |
The button is visible but disabled with a “Draft with AI (Business)” tooltip on lower tiers. Making it clear what’s available without hiding the feature.
How to use it
- Open an incident that’s been resolved
- Scroll to the Post-Mortem section
- Click Generate Post-Mortem (creates the structured shell. Timeline, summary stub, components affected)
- Click Draft with AI (Sparkles icon). Drafts the four fields
- Review and edit each field. AI is good at structure but doesn’t know your specific systems
- Toggle Publish to make the postmortem public
The Draft with AI button is disabled when:
- The incident is unresolved (drafts need a complete timeline)
- The postmortem is already published (avoid accidentally overwriting)
- Your team is on Free or Pro (Business only)
What the AI sees
The system prompt instructs the AI to:
- Output strictly valid JSON with four keys
- Be specific, factual, and concise
- Acknowledge gaps in the timeline: if updates are sparse, say so rather than invent details
- Never speculate about causes that aren’t supported by the timeline
The user prompt includes:
- Incident title, severity, status
- Duration (started_at → resolved_at)
- Timeline as a chronological list of updates with their statuses
The AI does not see:
- Logs or metrics. It only knows what’s in the incident timeline
- Other incidents from the same project
- Information about your team or specific people
This means the quality of the draft depends on how well your incident timeline was captured during the incident. Sparse timelines produce honest-but-thin drafts. Detailed timelines produce useful drafts.
When the AI is honest about not knowing
The system prompt explicitly instructs the AI to flag uncertainty. A typical example:
Root cause: Token cache miss in fallback path. Timeline suggests config-related, but logs not attached to confirm. Recommend checking deploy 3a4f2b for recent changes.
The “but logs not attached to confirm” is the AI being honest. If you ignore that hedge and publish the postmortem as fact, you’re making a claim the AI didn’t.
Tips for better drafts
- Post detailed updates during the incident. “Rolled back deploy 3a4f, watching for recovery” is more useful for AI than “trying things.” The AI’s draft quality is bounded by the timeline you wrote.
- Use specific service names. “Auth service throwing 503s” beats “things are broken”. Service names give the AI concrete things to attribute.
- Resolve cleanly. A final update like “Resolved. Issue was X, fix was Y, monitoring for regression” gives the AI the answer to copy.
- Edit aggressively. AI drafts are scaffolding, not gospel. Your team’s cultural style. Tone, jargon, what to emphasize. Should override what the AI produces.
Privacy note
When you click Draft with AI, the incident timeline is sent over HTTPS to the AI provider Signalog uses for your tier. We use enterprise-tier provider terms that prohibit using your data for model training. Drafts are ephemeral on the provider side. We don’t store them outside of your team’s database.
If your incidents contain sensitive data (customer names, internal IPs, PII), consider sanitizing the timeline before drafting, or skip the feature entirely. Internal scrubbing tooling is on the roadmap for enterprise plans.
Next steps
- Understand the incident lifecycle
- Manage incidents end-to-end
- Run a major incident. Bigger incidents benefit most from AI drafting