Prepare the source packet
Clean files before upload or connection
After this, you'll be able to prepare documents, notes, tables, and links so a knowledge assistant can retrieve them without duplicate or stale context.
Before you start
Complete Define the source boundary.
The idea
Source prep keeps retrieval from drowning in clutter.
- The packet separates keepThe packet separates keep and retire lists
- names missing metadataIt names missing metadata
- flags conflictsIt flags conflicts
- Create one source packetCreate one source packet with keep list, retired list, missing metadata, and conflicts
Packet rule: each source needs title, owner, date, status, topic, and use. Remove duplicates, drafts, broken links, and files with unclear ownership.
AI can help audit a source list, but the source owner confirms what is current. The assistant should not decide which policy wins when two files conflict.
Worked example: six onboarding docs become one current checklist, one FAQ, one benefits PDF, and a retired-draft list that stays out of ingestion.
Clean the source packet before asking the tool to learn from it.
Source prep is retrieval work. The retrieval system can only find what the source packet makes findable. Good titles, dates, owners, topics, and status labels matter because they help the tool separate current policy from old drafts.
Do not judge the source packet by size. A smaller packet of current files usually beats a large folder of mixed drafts, screenshots, and duplicates. The goal is not to upload everything. The goal is to give the assistant the material a careful human would trust.
Try it (12 min)
Watch out for
- Uploading every file because it is easier.
- Leaving duplicate drafts in the source set.
- Using files with no date.
- Ignoring conflicts before ingestion.
Paste this into Claude
Prepare this source packet. Sources: [paste file names, links, dates, owners if known] Knowledge task: [what answers people need] Current owner: [name] Return: 1. Keep list. 2. Remove or retire list. 3. Missing metadata. 4. Conflicts to resolve before ingestion.
What a good response looks like
Keep: onboarding-checklist-current, benefits-2026, FAQ-approved. Retire: onboarding-draft-v2 and old policy PDF. Missing: owner for contractor FAQ. Conflict: PTO policy differs between handbook and HR update.
What good looks like
- The packet separates keep and retire lists
- It names missing metadata
- It flags conflicts
- It assigns source owner review
When this breaks
- Breaks when stale files outrank current files during retrieval.
- Breaks when duplicate sources create conflicting answers.
AI can help with this
Use your knowledge assistant to help you you can prepare documents, notes, tables, and links so a knowledge assistant can retrieve them without duplicate or stale context. Start with the exercise prompt and your real input. Ask for one draft, then check it against this proof: The packet separates keep and retire lists. Accept only the version you can verify yourself.
You can now
You can clean source lists
- You can retire stale docs
- You can name metadata gaps
- You can flag conflicts
Key takeaways
Source prep is the cheapest way to improve answer quality before any AI tool gets involved.
- Current beats complete.
- Dates and owners matter.
- Duplicates create conflict.
- Conflicts need resolution.