Organize your files so AI can find anything

Most people's file organisation is a strata of "Downloads," "Documents," and "Desktop" with occasional folders called "Important" or "Misc." That structure is bad for humans and worse for AI retrieval.

A small amount of upfront discipline makes both you and the AI dramatically more effective. Here's the system.

Principle 1 — Topic, not date

"Q1 2026" tells you when. It doesn't tell you what. Folder by topic instead, with date as a file attribute:

• ✅ Vendor Contracts/acme-msa-2025-03.pdf
• ❌ 2025/March/contracts/contract.pdf

Topic-folders are how you'll think about retrieval ("show me Acme's MSA"), not how you'll think about archiving.

Principle 2 — Descriptive filenames

The filename is metadata. AI retrieval often picks up on filenames as a signal. Use:

• ✅ acme-msa-signed-2025-03-14.pdf
• ❌ contract.pdf
• ❌ IMG_4521.pdf
• ❌ Final_FINAL_v3_signed.pdf

A good filename template: {counterparty}-{doc-type}-{date}.pdf.

Principle 3 — Flat-ish hierarchies

Don't nest 5 levels deep. 2–3 levels max. Deep nesting makes scope selection annoying and adds zero retrieval value.

• ✅ Contracts / Vendors / acme-msa.pdf
• ❌ Business / Legal / Contracts / 2025 / Q1 / Vendors / IT / acme-msa.pdf

Principle 4 — One folder per assistant

When you build a SeekFiles Assistant, you scope it to a folder (or specific files). A 1:1 relationship between folders and assistants keeps everything clean.

• "Bar Reviewer" assistant → "Bar Reviewer" folder.
• "Tax 2025" assistant → "Tax 2025" folder.
• "Client X Case" assistant → "Client X" folder.

Principle 5 — Separate raw from processed

If you OCR a scan into a clean PDF, keep the original and the cleaned version, but in different subfolders:

• Receipts / raw-scans/
• Receipts / cleaned-pdfs/

Point your assistant at the cleaned folder. The raw scans stay for audit if needed but don't pollute retrieval.

Principle 6 — Tag the unusual

Most files are routine. The unusual ones (lapsed contracts, exceptional cases, legal flags) get a tag in the filename:

• acme-msa-2025-03-EXPIRED.pdf
• case-doe-v-roe-FLAG-conflict.pdf

You can then filter chats with "find all EXPIRED contracts" and get clean retrieval.

Cost of NOT organising

A folder with 500 mixed-topic files retrieves worse than five folders of 100 topic-scoped files each. The math is just retrieval noise: more irrelevant chunks crowd out the relevant ones. Organisation isn't aesthetic — it's accuracy.

Bonus — AI-assisted reorganisation

If your existing files are messy, build a one-time "Organiser" assistant scoped to everything. Ask:

"Group these files into topic-based folders. Suggest a folder structure and assign each file."

You won't follow the suggestion verbatim but it's a strong starting outline. Then you reorganise once and reap the benefits forever.