The reference files you re-upload constantly

There is a small set of documents you attach over and over: a style guide, a product spec, a terminology sheet, last quarter’s numbers. Inside the old account they feel like part of the tool. They are not — they are copies, and the account is about to stop being a place you can get them from.

The job is to assemble one canonical folder of these outside any product, then upload from it. It takes an afternoon and it never has to be done again.

Find them by looking at what you attached

Go through your recent conversations and note every file you uploaded. If the export is already flattened and searchable, the filenames appear in the text and the list assembles itself.

You are looking for repeats. A document uploaded once was a task; a document uploaded eleven times is part of how you work, and that is the set worth curating.

Expect the list to be shorter than you assume and different from what you would have guessed. The files you think of as important are often uploaded rarely, and something mundane — a template, a glossary — turns out to be in half your conversations.

Then find the real copies

For each repeated file, locate the authoritative version, which is almost never the one in the assistant.

That is the point of the exercise. The upload was a copy of something living in shared storage, an email attachment, or a folder on your machine, and copies drift. If you have been uploading a style guide for eight months, there is a real chance the version in the last conversation is older than the version your colleagues are using.

Whether attachments came through in your export varies by product and is worth checking rather than assuming — but even when they did, the exported file is a snapshot of a copy. It is evidence of what you used, not the document of record.

Assemble one folder, with one version of each

A single folder, next to your instruction file and prompt library. One file per thing. No -v2, no -final, no two candidates you will have to choose between under time pressure.

Where a document genuinely has versions, keep the current one at the top level and put the rest in an old/ subfolder. The rule is that the folder can be used without deciding anything.

Add a one-line note per file if the purpose is not obvious from the name: what it is for, and where the authoritative version lives if it is maintained elsewhere. That second half is what stops the folder becoming another stale copy in six months.

Convert to formats that will still open

Prefer plain text or Markdown for anything you wrote yourself. Keep the original format alongside it where the formatting genuinely matters.

Two reasons, both practical. Text files are what every assistant handles most predictably, so uploads behave consistently. And text is what still opens in a decade, which is the same reason it is the right format for the archive.

Anything that is a spreadsheet or a design file stays in its own format — converting those loses the thing they are for.

Upload from the folder, not from wherever

The habit that makes this stick: the folder is the source, and every upload comes from it.

That means when a document changes, it changes in the folder, and the next upload is current automatically. It also means the question “which version did I give it?” has an answer, which it does not when uploads come from your downloads folder and your email interchangeably.

If the new assistant has some persistent place to attach reference material, populate it from the folder — and treat that as a cache, not as storage. It is inside a product, and products are things you leave.

What moves

WHAT MOVES — reference documents

  · The documents themselves
                    → move, because you own them. The
                      upload was always a copy.

  · Attachments in the export
                    → varies by product. Present,
                      name-only, or STAYS BEHIND.

  · The association between a file and the
    conversation that used it
                    → STAYS BEHIND except as text.

  · Whatever the old account did with an
    uploaded file — indexing, extraction,
    persistent attachment
                    → STAYS BEHIND. Re-upload and
                      re-establish it.

  · Deleting the account before locating the
    real copies
                    → IRREVERSIBLE for any file whose
                      only surviving copy was the
                      upload.

The file that exists only as an upload

Occasionally there is one: something you generated in a conversation, downloaded nowhere, and have been re-attaching ever since.

Find these before you close anything. They are the only genuinely at-risk item in this whole exercise, because everything else is a copy of something you still have. Download them, put them in the folder, and note that they came from the assistant rather than from a source you can go back to.

This is worth a deliberate pass through the conversations where you know you produced a document. Not a long job, and it is the one item on the list that cancellation destroys.

What this does not solve

The new assistant has no idea what any of these documents are, or that a particular one applies to a particular kind of task. That association was accumulated context on the old side and it is gone — seeding it back is a deliberate act, not something the folder does for you.

And you will be attaching files manually again for a while, in situations where the old setup had stopped requiring it. That friction is temporary, and it is the honest cost of having the documents live somewhere you own rather than somewhere you rent.