The problem this fixes

Content libraries become hard to trust when every asset stays live, ownership is unclear, and the team tries to clean everything at once.

What good content cleanup changes

Good content cleanup makes the library easier to trust by naming what to keep, update, merge, archive, retire, or route back to an owner.

Ten-business-day implementation path

Clean one useful slice before touching the whole library.

Day 1

Pick one slice

Choose one audience, workflow, product area, onboarding path, or catalog category. Do not start with the full library.

Day 2

Inventory only the fields that matter

Capture title, audience, task, owner, source, last review date, usage signal, duplicate risk, status, and retirement trigger.

Day 3

Rewrite topics as tasks

Turn broad content labels into the tasks people need to perform. If an asset cannot be tied to a task, requirement, or support moment, flag it.

Day 4

Find owners and sources

Name the person or team that can confirm accuracy. If no source owner exists, mark the asset as needs owner instead of silently trusting it.

Day 5

Sort the first decision lane

Classify each asset as keep, update, merge, archive, retire, or needs owner. This is a triage pass, not a redesign.

Day 6

Cluster duplicates

Group assets that teach the same task, answer the same question, or reuse the same source material. Decide which one becomes the source path.

Day 7

Write retirement rules

Define what makes content expire: source changes, tool screens change, no usage, duplicate exists, owner leaves, policy changes, or workflow changes.

Day 8

Test findability

Ask one target user, manager, or support partner to find the right asset for one real task. Watch where the library sends them sideways.

Day 9

Publish the cleanup rule

Share the status definitions, ownership rule, review cadence, archive process, and the rule for adding new content to the cleaned slice.

Day 10

Choose the next slice

Use the same model on the next highest-risk slice only after the first slice has a working owner, status, and maintenance rhythm.

Aha moments

Content cleanup gets easier when these habits are explicit.

  • The goal is not a perfect library. The goal is a library people can trust enough to use.
  • A stale asset is usually an ownership problem before it is a design problem.
  • Migration is the best time to retire content because the team already has to touch it.
  • A broad course may hide five smaller task supports that are easier to maintain.
  • Do not ask whether content is good before asking whether it is still needed.
  • The fastest cleanup starts with one slice, one audience, or one repeated workflow.
  • If no one can name the source owner, the content should not be treated as current.
Cleanup lanes

Give every asset a decision lane before editing anything.

Keep

Current, used, owned, and tied to a real task or requirement.

Update

Still needed, but the source, screen, workflow, example, or practice needs revision.

Merge

Useful content exists in more than one place and should be consolidated.

Archive

Not needed for learners now, but worth keeping for history, audit, or later reference.

Retire

No clear owner, use, audience, source, or reason to keep it live.

Needs owner

Potentially useful, but no one can approve accuracy, timing, or retirement yet.

Three ways to build it

Pick the lightest system that makes the work repeatable.

Manual setup

Run a content triage wall with keep, update, merge, archive, retire, and needs owner. Add one sticky note per asset and sort before editing.

Microsoft 365 or Google Workspace setup

Use Excel, Google Sheets, Microsoft Lists, SharePoint, Airtable, or a project board with owner, source, task, status, last review date, usage signal, and retirement trigger.

AI-assisted setup

Use AI to summarize asset titles, cluster possible duplicates, draft task labels, and generate owner review questions. Keep humans responsible for accuracy, retirement, risk, and final status.

Workspace setup notes
  • Use a tracker before a redesign board. Cleanup fails when the team jumps to production without a reliable inventory.
  • Create controlled status values so keep, update, merge, archive, retire, and needs owner mean the same thing every week.
  • Add a source-owner field. A designer may own the asset, but someone else may own the truth.
  • Track review triggers instead of only review dates. Content often expires because the work changes, not because the calendar changed.
Useful AI jobs
  • Cluster likely duplicate assets by title, summary, task, or source.
  • Rewrite topic-heavy titles as task labels for human review.
  • Draft source-owner questions for each content group.
  • Find assets that have unclear audience, task, owner, or source.
  • Prepare a cleanup summary that separates decisions made from decisions still pending.
Content cleanup kickoff script

Use these words when the team wants to rebuild everything.

  • We are not cleaning the whole library today. We are choosing one slice and making it trustworthy.
  • Before we edit content, we need to know whether each asset should stay live.
  • Every asset needs a status, owner, source, and next rule.
  • If we cannot name the task or owner, we should not treat the asset as current.
  • The cleanup is successful when people can find the right support and trust that it is safe to use.
Content decision log

Every asset needs a status, owner, and next rule.

Keep

The asset is current, owned, and still useful for a clear audience and task.

Update

The asset is still needed, but source, workflow, screen, example, or practice detail has changed.

Merge

The asset overlaps with another support path and should be consolidated.

Archive

The asset should leave the active path but remain available for audit or reference.

Retire

The asset no longer has a clear use, owner, source, or audience.

Needs owner

No one can validate the content yet, so it should not be treated as trusted.

Measure after two weeks

Look for lower content debt, not just a prettier inventory.

  • Number of assets assigned keep, update, merge, archive, retire, or needs owner.
  • Percent of cleaned assets with owner, source, task, and review trigger.
  • Duplicate clusters reduced or assigned to a source path.
  • Assets retired or removed from active learner paths.
  • Time required for a target user to find the right asset for one task.
Research used

Use the sources as thinking tools, not a script.

Prompt pack

Use AI to inspect the library without letting it decide what stays live.

Sort an inventory slice

The team has a partial content inventory and needs to classify each asset before editing.

ChatGPT GPT-5 family

Use an outcome-first prompt with the job, approved source material, constraints, and the exact artifact you want back.

I am working on Sort an inventory slice for an L&D system problem.

Goal: Help me turn the notes below into a practical next move.

Context: The team has a partial content inventory and needs to classify each asset before editing.

Use these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.

Rules:
- Use only the source notes I provide.
- Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
- Separate known facts, assumptions, missing information, and next actions.
- Flag anything that needs requester, reviewer, leader, legal, compliance, LMS owner, or manager confirmation.
- Keep the output practical enough to review in a working meeting.

Source notes:
[paste approved notes here]

Return:
1. Suggested status
2. Known facts
3. Assumptions
4. Missing information
5. Owner question
6. Next action

Claude 4 family

Use XML-style sections so context, source material, task, constraints, and output format stay separate.

<context>
I am working on Sort an inventory slice for an L&D system problem.
The team has a partial content inventory and needs to classify each asset before editing.
</context>

<source_notes>
[paste approved notes here]
</source_notes>

<task>
Turn the source notes into a practical next move using these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
</task>

<constraints>
Use only the source notes provided.
Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
Separate known facts, assumptions, missing information, risks, and next actions.
Flag anything that changes scope, ownership, evidence, risk, or decision rights.
</constraints>

<output_format>
1. Suggested status
2. Known facts
3. Assumptions
4. Missing information
5. Owner question
6. Next action
</output_format>

Gemini 3 family

Use a clear task, labeled input, and one example pattern. For Obsidian context, use approved excerpts, Drive exports, Google Docs, or NotebookLM source sets.

Task: Help me make progress on Sort an inventory slice from the notes provided.

Context: The team has a partial content inventory and needs to classify each asset before editing.

Working fields:
- asset title
- audience
- task
- owner
- source
- use signal
- duplicate risk
- status
- next action

Example pattern:
Field: Missing information
Good answer: Name the specific information to confirm, who can confirm it, and why it affects the next decision.

Rules:
- Use only the source notes provided.
- If information is missing, write "Needs confirmation".
- Keep the output concise and reviewable.
- End with the next best action.

Source notes:
[paste approved notes here]

Output format:
1. Suggested status
2. Known facts
3. Assumptions
4. Missing information
5. Owner question
6. Next action

Microsoft 365 Copilot

Use goal, context, source, expectations, and output. For Obsidian context, use approved excerpts, Word summaries, OneDrive files, SharePoint pages, Teams context, or Outlook threads.

Goal: Help me make progress on Sort an inventory slice.

Context: The team has a partial content inventory and needs to classify each asset before editing.

Source: Use the selected document, meeting notes, spreadsheet, email thread, SharePoint file, or pasted notes as the only source.

Expectations:
- Work with these fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
- Mark uncertain items as "Needs confirmation".
- Do not add facts that are not in the source.
- Separate known facts, assumptions, missing information, risks, and next actions.
- Summarize the top review questions for the team.

Output:
1. Suggested status
2. Known facts
3. Assumptions
4. Missing information
5. Owner question
6. Next action
Find duplicate content patterns

The library may have multiple assets that support the same task or repeat the same source material.

ChatGPT GPT-5 family

Use an outcome-first prompt with the job, approved source material, constraints, and the exact artifact you want back.

I am working on Find duplicate content patterns for an L&D system problem.

Goal: Help me turn the notes below into a practical next move.

Context: The library may have multiple assets that support the same task or repeat the same source material.

Use these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.

Rules:
- Use only the source notes I provide.
- Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
- Separate known facts, assumptions, missing information, and next actions.
- Flag anything that needs requester, reviewer, leader, legal, compliance, LMS owner, or manager confirmation.
- Keep the output practical enough to review in a working meeting.

Source notes:
[paste approved notes here]

Return:
1. Possible duplicate group
2. Why it looks duplicated
3. Source to confirm
4. Risk if both stay live
5. Recommended human review

Claude 4 family

Use XML-style sections so context, source material, task, constraints, and output format stay separate.

<context>
I am working on Find duplicate content patterns for an L&D system problem.
The library may have multiple assets that support the same task or repeat the same source material.
</context>

<source_notes>
[paste approved notes here]
</source_notes>

<task>
Turn the source notes into a practical next move using these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
</task>

<constraints>
Use only the source notes provided.
Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
Separate known facts, assumptions, missing information, risks, and next actions.
Flag anything that changes scope, ownership, evidence, risk, or decision rights.
</constraints>

<output_format>
1. Possible duplicate group
2. Why it looks duplicated
3. Source to confirm
4. Risk if both stay live
5. Recommended human review
</output_format>

Gemini 3 family

Use a clear task, labeled input, and one example pattern. For Obsidian context, use approved excerpts, Drive exports, Google Docs, or NotebookLM source sets.

Task: Help me make progress on Find duplicate content patterns from the notes provided.

Context: The library may have multiple assets that support the same task or repeat the same source material.

Working fields:
- asset title
- audience
- task
- owner
- source
- use signal
- duplicate risk
- status
- next action

Example pattern:
Field: Missing information
Good answer: Name the specific information to confirm, who can confirm it, and why it affects the next decision.

Rules:
- Use only the source notes provided.
- If information is missing, write "Needs confirmation".
- Keep the output concise and reviewable.
- End with the next best action.

Source notes:
[paste approved notes here]

Output format:
1. Possible duplicate group
2. Why it looks duplicated
3. Source to confirm
4. Risk if both stay live
5. Recommended human review

Microsoft 365 Copilot

Use goal, context, source, expectations, and output. For Obsidian context, use approved excerpts, Word summaries, OneDrive files, SharePoint pages, Teams context, or Outlook threads.

Goal: Help me make progress on Find duplicate content patterns.

Context: The library may have multiple assets that support the same task or repeat the same source material.

Source: Use the selected document, meeting notes, spreadsheet, email thread, SharePoint file, or pasted notes as the only source.

Expectations:
- Work with these fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
- Mark uncertain items as "Needs confirmation".
- Do not add facts that are not in the source.
- Separate known facts, assumptions, missing information, risks, and next actions.
- Summarize the top review questions for the team.

Output:
1. Possible duplicate group
2. Why it looks duplicated
3. Source to confirm
4. Risk if both stay live
5. Recommended human review
Turn topics into task labels

The team has broad content titles and needs task-based labels for findability and maintenance.

ChatGPT GPT-5 family

Use an outcome-first prompt with the job, approved source material, constraints, and the exact artifact you want back.

I am working on Turn topics into task labels for an L&D system problem.

Goal: Help me turn the notes below into a practical next move.

Context: The team has broad content titles and needs task-based labels for findability and maintenance.

Use these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.

Rules:
- Use only the source notes I provide.
- Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
- Separate known facts, assumptions, missing information, and next actions.
- Flag anything that needs requester, reviewer, leader, legal, compliance, LMS owner, or manager confirmation.
- Keep the output practical enough to review in a working meeting.

Source notes:
[paste approved notes here]

Return:
1. Current title
2. Likely task
3. Audience
4. Missing context
5. Better label
6. Confirmation needed

Claude 4 family

Use XML-style sections so context, source material, task, constraints, and output format stay separate.

<context>
I am working on Turn topics into task labels for an L&D system problem.
The team has broad content titles and needs task-based labels for findability and maintenance.
</context>

<source_notes>
[paste approved notes here]
</source_notes>

<task>
Turn the source notes into a practical next move using these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
</task>

<constraints>
Use only the source notes provided.
Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
Separate known facts, assumptions, missing information, risks, and next actions.
Flag anything that changes scope, ownership, evidence, risk, or decision rights.
</constraints>

<output_format>
1. Current title
2. Likely task
3. Audience
4. Missing context
5. Better label
6. Confirmation needed
</output_format>

Gemini 3 family

Use a clear task, labeled input, and one example pattern. For Obsidian context, use approved excerpts, Drive exports, Google Docs, or NotebookLM source sets.

Task: Help me make progress on Turn topics into task labels from the notes provided.

Context: The team has broad content titles and needs task-based labels for findability and maintenance.

Working fields:
- asset title
- audience
- task
- owner
- source
- use signal
- duplicate risk
- status
- next action

Example pattern:
Field: Missing information
Good answer: Name the specific information to confirm, who can confirm it, and why it affects the next decision.

Rules:
- Use only the source notes provided.
- If information is missing, write "Needs confirmation".
- Keep the output concise and reviewable.
- End with the next best action.

Source notes:
[paste approved notes here]

Output format:
1. Current title
2. Likely task
3. Audience
4. Missing context
5. Better label
6. Confirmation needed

Microsoft 365 Copilot

Use goal, context, source, expectations, and output. For Obsidian context, use approved excerpts, Word summaries, OneDrive files, SharePoint pages, Teams context, or Outlook threads.

Goal: Help me make progress on Turn topics into task labels.

Context: The team has broad content titles and needs task-based labels for findability and maintenance.

Source: Use the selected document, meeting notes, spreadsheet, email thread, SharePoint file, or pasted notes as the only source.

Expectations:
- Work with these fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
- Mark uncertain items as "Needs confirmation".
- Do not add facts that are not in the source.
- Separate known facts, assumptions, missing information, risks, and next actions.
- Summarize the top review questions for the team.

Output:
1. Current title
2. Likely task
3. Audience
4. Missing context
5. Better label
6. Confirmation needed
Draft content owner questions

The team needs to ask source owners whether assets are accurate, current, useful, or ready to retire.

ChatGPT GPT-5 family

Use an outcome-first prompt with the job, approved source material, constraints, and the exact artifact you want back.

I am working on Draft content owner questions for an L&D system problem.

Goal: Help me turn the notes below into a practical next move.

Context: The team needs to ask source owners whether assets are accurate, current, useful, or ready to retire.

Use these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.

Rules:
- Use only the source notes I provide.
- Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
- Separate known facts, assumptions, missing information, and next actions.
- Flag anything that needs requester, reviewer, leader, legal, compliance, LMS owner, or manager confirmation.
- Keep the output practical enough to review in a working meeting.

Source notes:
[paste approved notes here]

Return:
1. Question for owner
2. Why it matters
3. Decision it supports
4. Risk to check
5. Next action

Claude 4 family

Use XML-style sections so context, source material, task, constraints, and output format stay separate.

<context>
I am working on Draft content owner questions for an L&D system problem.
The team needs to ask source owners whether assets are accurate, current, useful, or ready to retire.
</context>

<source_notes>
[paste approved notes here]
</source_notes>

<task>
Turn the source notes into a practical next move using these working fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
</task>

<constraints>
Use only the source notes provided.
Do not invent policy details, metrics, learner needs, compliance requirements, or business context.
Separate known facts, assumptions, missing information, risks, and next actions.
Flag anything that changes scope, ownership, evidence, risk, or decision rights.
</constraints>

<output_format>
1. Question for owner
2. Why it matters
3. Decision it supports
4. Risk to check
5. Next action
</output_format>

Gemini 3 family

Use a clear task, labeled input, and one example pattern. For Obsidian context, use approved excerpts, Drive exports, Google Docs, or NotebookLM source sets.

Task: Help me make progress on Draft content owner questions from the notes provided.

Context: The team needs to ask source owners whether assets are accurate, current, useful, or ready to retire.

Working fields:
- asset title
- audience
- task
- owner
- source
- use signal
- duplicate risk
- status
- next action

Example pattern:
Field: Missing information
Good answer: Name the specific information to confirm, who can confirm it, and why it affects the next decision.

Rules:
- Use only the source notes provided.
- If information is missing, write "Needs confirmation".
- Keep the output concise and reviewable.
- End with the next best action.

Source notes:
[paste approved notes here]

Output format:
1. Question for owner
2. Why it matters
3. Decision it supports
4. Risk to check
5. Next action

Microsoft 365 Copilot

Use goal, context, source, expectations, and output. For Obsidian context, use approved excerpts, Word summaries, OneDrive files, SharePoint pages, Teams context, or Outlook threads.

Goal: Help me make progress on Draft content owner questions.

Context: The team needs to ask source owners whether assets are accurate, current, useful, or ready to retire.

Source: Use the selected document, meeting notes, spreadsheet, email thread, SharePoint file, or pasted notes as the only source.

Expectations:
- Work with these fields: asset title, audience, task, owner, source, use signal, duplicate risk, status, next action.
- Mark uncertain items as "Needs confirmation".
- Do not add facts that are not in the source.
- Separate known facts, assumptions, missing information, risks, and next actions.
- Summarize the top review questions for the team.

Output:
1. Question for owner
2. Why it matters
3. Decision it supports
4. Risk to check
5. Next action
Try this one thing today

Pick one content slice and classify every asset as keep, update, merge, archive, retire, or needs owner before editing.