25 Years of Files

What this is for

You have an old external drive, a graveyard Dropbox folder, or a laptop from three jobs ago full of decks, reports, and templates. Somewhere in there are frameworks you've reused without writing down. This is for pulling that material into a tool you can search and question, before deciding what to build from it.

Before you start

• Locate the actual files — this takes longer than the rest of the process; start it first.
• Decide which tool you'll use before uploading (see below) — switching later means re-uploading everything.
• Set aside a separate folder for anything client-confidential, before you upload anything.

What to gather and what to skip

Gather: proposals and SOWs you wrote, internal frameworks or process docs, training materials, retrospectives, presentations reused across projects, anything you wrote twice because it worked the first time.

Skip: raw meeting notes with no synthesis, financial or HR records, anyone else's personal data, one-off emails, superseded drafts (keep the final only), and anything you don't have the right to reuse — see the warning below.

The folder structure

Flat and by-type beats deep and by-project — you're building for retrieval, not archiving.

/legacy-archive
  /frameworks-and-processes
  /client-facing-templates (SOWs, proposals, decks — anonymized)
  /training-and-onboarding-materials
  /case-studies-and-retros
  /reference-only (background material, not for extraction)

Tools retrieve better from a shallow structure with descriptive filenames than a deep nested one — filenames and folder names are part of the retrieval signal.

Which tool to use

• NotebookLM — best for pure research and querying. Handles many source documents well, cites the exact source file, free. Weakest at turning findings into new deliverables.
• Claude Projects — best for the extraction and rewriting work below. 200K-token context handles a substantial batch. Costs a Pro or Team subscription. Less precise retrieval than NotebookLM across very large file sets.
• ChatGPT Projects — best if you already live in ChatGPT. File search has improved but is still less precise than NotebookLM for large sets, and per-project storage has limits.

First time doing this: start in NotebookLM to confirm what's valuable, then move the good material into Claude or ChatGPT Projects for the writing and extraction work.

De-duplication and naming pass

Do this before you upload to your chosen tool — it makes everything downstream better.

Use in: Either, but this works best as a local step: list your filenames first, then paste the list in.

ROLE: You are an archivist who cleans up messy personal file collections for retrieval.

CONTEXT: A list of filenames from my old work archive: [PASTE FILENAME LIST, e.g. "Proposal_FINAL_v3.docx, Proposal_FINAL_v3_ACTUALFINAL.docx, Q3 review (1).pptx, Copy of Training deck.pptx"]

CONSTRAINTS:
- Identify likely duplicates from filename patterns (version numbers, "final," "copy of," date stamps) — flag them, don't assume which is authoritative without asking.
- Propose a new filename per file: [type]-[topic]-[year].[ext], e.g. "framework-onboarding-process-2019.pptx"
- Group output by folder: frameworks-and-processes, client-facing-templates, training-and-onboarding-materials, case-studies-and-retros, reference-only.
- Where you can't tell what a file is from its name, list it under "Needs manual review" instead of guessing.

OUTPUT FORMAT: A table: Original filename | Likely duplicate of (or "none") | Proposed new name | Proposed folder. Then a "Needs manual review" list.

The prompt that extracts your repeatable frameworks

Use in: Claude Projects, after uploading a batch of your actual documents (not summaries of them) — this needs to read full source material to find patterns across files, which benefits from Claude's longer context.

ROLE: You are a knowledge management consultant who extracts reusable frameworks from a career's worth of work product.

CONTEXT: The uploaded files are a sample of my work over [NUMBER] years in [FIELD/ROLE, e.g. "22 years in supply chain consulting"]. I've likely repeated the same approaches or decision frameworks without writing them down.

CONSTRAINTS:
- Only surface a pattern if it appears in at least two uploaded documents — cite which files.
- Don't invent a framework that sounds plausible but isn't evidenced in the files.
- Distinguish a true repeatable framework from a one-off solution — label each finding as one or the other.
- Where a pattern is only partially formed, describe what's there and what's missing rather than completing it for me.

OUTPUT FORMAT: Numbered list of candidate frameworks: working name, files it appears in, 2-3 sentence description, and "fully formed" or "partial — needs my input."

Warning: client-confidential material and NDAs

Before uploading anything from a past job or client engagement, check whether it's covered by an NDA, a confidentiality clause, or your former employer's data policy — most consulting and corporate work is, and the obligation typically doesn't expire when the engagement ends. Uploading client-identifiable material to any AI tool can breach that agreement even if you never publish the output. When in doubt: strip identifying details, rewrite from memory instead of uploading the original, or leave it out. If you're unsure whether something is covered, treat it as covered.

Where this goes wrong

• You upload everything at once and get generic, unusable summaries. Volume without curation produces mush. Do the gather-and-skip pass first.
• You extract a "framework" that was really one project's specific solution. If you can't explain it to someone outside that project, it isn't a framework yet.
• You treat old client work as automatically yours to reuse. Check your agreements before you check your drive.

The 2-minute version

Find the drive. Sort files into gather/skip, erring toward skip for anything client-named. Run the de-duplication prompt on your filename list. Upload the gather pile to NotebookLM first to see what's there before committing to a full build.

---

Where you stand with AI, in 10 minutes. The free AI Readiness Quiz gives you a personalized roadmap instead of another tool list: genxcelerate.com

Weekly, on Tuesdays. The Gen Xcelerator — practical AI intelligence for people who already know how to do the job. Subscribe free

Experience Is the API. — GenXcelerate