Help-centre article gaps from ticket themes

Use a tagged ticket sample to identify the self-service content that would remove repeat contacts. You will leave with a ranked article backlog, each item supported by customer wording, a proposed title and an outline ready for assignment.

This workflow is for a support content manager or support lead who owns help-centre coverage. Run it after a product change, during a content planning cycle, or whenever one ticket theme is growing.

1. Set the review period and sample

Choose one support queue, one market or language, and a fixed review period. Do not mix different queues unless their customers, product access and support policies are the same.

Export or copy a ticket sample into a review sheet. Include resolved tickets and tickets escalated to another team. Exclude spam, duplicate records and contacts with no usable customer question.

Create these columns in the sheet:

• Ticket ID: the record you can return to when checking a finding.
• Created date: lets you spot a recent issue rather than an old one.
• Existing tag: the queue tag or topic tag already applied.
• Customer question: the customer's own wording, with personal data removed.
• Support outcome: what the agent told them to do.
• Current article: the article linked, if any.
• Resolution status: resolved, reopened, escalated or unresolved.
• Content observation: missing article, weak article, wrong article, or no content issue.

Keep a separate column for a short, anonymised quotation. Do not paste names, email addresses, account numbers, order details, internal notes or attachments into the review material.

2. Make the sample usable

Review enough tickets from each material tag to see the underlying task, not just the label. If one tag contains very different questions, split it in the sheet. For example, a tag called billing may contain failed payments, invoice requests, refund timing and plan changes. These need different article decisions.

For each ticket, write one plain-language intent in the Content observation field. Use a verb and object:

• change payment method
• find VAT invoice
• cancel a renewal
• understand why verification failed

Mark whether the answer is already available in the help centre:

What you find	Mark it as	Next action
No article covers the task	Missing article	Propose a new article
An article exists but omits a required step	Weak article	Propose a revision
The article answers a different question	Wrong article	Fix title, links or navigation
The answer depends on an individual account review	No content issue	Route to an escalation or contact path

Use the final category carefully. A help-centre article can explain eligibility, expected evidence and next steps, even when support must make the final decision. Do not label a repeatable explanation as account-specific merely because an agent handled it manually.

3. Group questions into article candidates

Sort the sheet by intent. Group tickets only when a single article could answer them without becoming vague. Keep distinct customer goals separate, even if the same policy or product area is involved.

Create one row per candidate in an article backlog. Add these fields:

• Candidate ID: for example, HC-01.
• Customer task: the job the customer is trying to complete.
• Content status: missing, weak or wrong.
• Evidence count: the number of sampled tickets in the group.
• Evidence quotes: two or three short anonymised question fragments.
• Existing article and gap: its URL or title, then the missing step, unclear wording or misleading scope.
• Affected tag(s): the tags found in the sample.
• Support outcome: the common reply or macro used.
• Owner: the team that can verify the process.

A candidate should describe one customer outcome. Billing help is too broad. Download an invoice for a completed payment is specific enough to write and check.

4. Rank the backlog before drafting

Score each candidate from 1 to 3 on four factors. Record the scores, rather than relying on a meeting discussion later.

Factor	Score 1	Score 3
Ticket presence	Seen rarely in the sample	Seen repeatedly across the sample
Customer friction	Simple clarification	Blocks access, payment, setup or a key task
Self-service fit	Requires individual investigation	A customer can follow documented steps
Content weakness	Existing content is mostly adequate	No usable content, or current content sends customers wrongly

Add the four scores. Sort highest first. Then apply two checks:

1. Move down any item that depends on an unconfirmed product or policy decision.
2. Move up an item where a clear article would prevent unnecessary escalation or repeated contact.

Do not use the score as a prediction of reduced ticket volume. It is a transparent way to decide what your team should examine first.

5. Draft the proposed article package

For the top candidates, produce a short content brief. This is the handover to the writer, subject expert or reviewer.

Use this format for each candidate:

Candidate ID: HC-01
Proposed title: [Start with the customer task]
Content status: Missing article
Customer task: [What they need to do]
Why now: [Score and short reason]
Evidence: [Ticket IDs] and [two anonymised question fragments]
Current support answer: [Steps or macro summary]
Article outline:
1. When to use this article
2. What you need before starting
3. Steps in order
4. What happens next
5. Common errors or limits
6. When to contact support
Links to add: [related articles or contact route]
Reviewer: [process owner]

Write titles in the words customers use, while keeping the product term where it matters. Prefer Change the payment method for your subscription over Payment settings. The first tells the reader whether the page solves their task.

Keep the outline procedural. If an agent reply contains three actions, the article should normally show those actions in the same order. Add screenshots, interface labels or policy wording only after the process owner checks them.

If you use an AI workspace to cluster questions or draft outlines, first check the current product guidance and applicable limits in the xAI documentation overview. Features and handling options are version-dependent. Give it anonymised ticket text and ask it to return ticket IDs beside every theme, so the findings remain auditable.

6. Check that the recommendations are real

Read the top five candidates against the original tickets. This is where weak analysis usually appears.

Look for these failure signals:

• Several evidence quotes use the same wording because they came from one long conversation, not separate contacts.
• The proposed title combines two different tasks, such as changing a plan and requesting a refund.
• The outline repeats an internal macro but does not tell the customer where to click, what they need, or what result to expect.
• The current article was not actually read, so a revision is wrongly labelled as a missing article.
• The candidate is common only because an outage or temporary defect generated a short-lived cluster.

Ask the process owner to verify every factual step in the highest-ranked items. Ask a support lead to check whether the title matches the language agents see in tickets. Record disagreements in the backlog rather than silently changing the score.

7. Hand over and run the next review

Share the ranked backlog with the content owner and the relevant product, policy or operations reviewer. For each of the first items, assign an owner, reviewer and target review date in your usual planning system.

After publishing or revising an article, add its title to the original candidate row. In the next review, check whether agents still send the same macro, whether customers still ask the same question, and whether the new article creates confusion elsewhere in the journey.

If the workflow does not produce clear priorities, do not force a list. Recheck the sample boundaries, split broad tags, and collect more tickets from the unclear themes. If the evidence is clear but the answer is not stable, pause the article and ask the process owner to settle the process first.