Docs / How it works
How ticket analysis works
What AutoKnowledgeBase does with each solved ticket: personal data removal, the worth-documenting check, matching against your help center, and the create, update or skip decision.
Every time a ticket is solved, AutoKnowledgeBase runs it through the same steps. Each step can stop the process, so most tickets never turn into a draft. That's on purpose: a help center full of one-off answers is worse than one that's missing a few.
1. The ticket arrives
When a ticket's status changes to Solved, the trigger AutoKnowledgeBase created in your Zendesk account sends the ticket ID to AutoKnowledgeBase. Tickets that already have the akb_analyzed tag are skipped.
2. Your help center is synced
AutoKnowledgeBase pulls any articles and sections that changed since the last sync. The comparison in step 6 always uses your current help center, including articles you edited a minute ago.
3. Personal data is removed
Before any AI step, AutoKnowledgeBase replaces personal details in the ticket with placeholders such as [EMAIL] or [PHONE]. It looks for:
- Email addresses and phone numbers
- Card numbers, IBANs, bank account, routing and sort code numbers
- Social Security, passport, driving licence, National Insurance and NHS numbers
- Street addresses, ZIP codes and UK postcodes
- IP addresses
- Names in context, such as "Hi John Smith"
Only the scrubbed version is used from here on. The text is processed through the OpenAI API, which doesn't use API data to train its models. See the privacy policy for details.
4. One credit is used
The ticket uses one credit at this point, whatever the outcome. See Credits and billing. If you're out of credits, the ticket isn't analyzed.
5. Is it worth documenting?
An AI check decides whether the ticket holds knowledge other customers could use on their own: a common problem with a solution they can follow themselves.
Tickets about one customer's account, internal fixes ("we reset your account"), billing, refunds, access requests or feature requests are skipped, as are tickets with no clear resolution.
The check is conservative: when it isn't sure, the ticket is skipped. Skipped tickets are tagged akb_not_kb_worthy.
6. Matching against your help center
For tickets that pass, AutoKnowledgeBase summarizes the problem and the solution, then searches your help center for the articles that cover the same topic. The three closest matches, with how similar each one is, are passed to the next step and shown in your dashboard.
7. Create, update, or already covered
An AI model reads the ticket summary and the closest articles and picks one of three outcomes, with a confidence score and a short reason:
| Outcome | When | What happens |
|---|---|---|
| Already covered | An existing article fully covers the ticket | Nothing is written. The ticket is tagged akb_already_covered. |
| Update | An article covers the topic but is missing details from this ticket | An update draft is prepared for that article. |
| Create | No article covers the topic | A new draft article is created. |
Guardrails keep the decision on the safe side:
- If one article is a near-exact match, the ticket counts as already covered without asking the AI.
- An update is only made when the closest article is similar enough and the AI is confident. Otherwise a new draft is created instead, so new information is never forced into the wrong article.
- "Already covered" also needs high confidence. If the AI isn't sure, a new draft is created, so you can decide.
8. Writing the draft
New articles go into the section of your help center that fits the ticket best. The AI picks it from your section names and descriptions.
Updates depend on the article:
- Draft article: the update is written straight into the draft.
- Published article: your live article is never changed. The update goes into a separate draft in the same section, titled [Proposed update] article title and labelled
akb-review. A note at the top says it's a proposed update and links to the live article. - Published article with a proposed update already open: the open draft is left as it is, and the new suggestion waits in your AutoKnowledgeBase dashboard.
Section templates
Drafts follow the structure of the articles already in each section. During sync, AutoKnowledgeBase detects a template for each section from its existing articles. You can see and edit them in Settings → Section Templates. Templates you edit are marked Custom and are never overwritten by sync. Reset to Auto-Detected brings back the detected version.
9. Writing back to the ticket
Finally, AutoKnowledgeBase tags the ticket with the outcome and posts a private internal note explaining it, with links to the draft. Both are on by default and can be turned off. See Ticket tags and internal notes.
If analysis fails, the ticket is tagged akb_error and isn't retried automatically.
Seeing every decision
Your dashboard lists every analyzed ticket with its outcome. Open a ticket to see the decision, the confidence, the reasoning, the top three matching articles, and the difference between the original article and the suggested one.
Last updated November 21, 2026