Don't just hand over past articles to AI—what to keep and what to discard when auditing context

Past articles and scripts are useful materials for AI. However, simply handing over entire folders is not the right approach. If you include old policies, abandoned projects, or decisions that no longer reflect who you are, the AI will reference them with the same weight. Accumulation is not about increasing volume, but about organizing things into a usable state.

First, separate "originals" from "reference examples"

Transcripts, published manuscripts, primary sources, and project memos all have different roles. Do not put the originals that record what you said in the same box as the reference examples used as writing samples. When handing them to an AI, it becomes clear what should be treated as the basis for facts and what should be treated as a model for expression.

What should be kept are the decisions you make repeatedly

Beyond just how to write titles or place conclusions, there are items you always check in articles, primary sources you want to show readers, and assertions you should avoid. Extract these recurring decisions from individual manuscripts and store them separately. It is more accurate to provide them as short rules than to make the AI guess every time it reads a manuscript.

Don't delete them, mark them as "expired"

There is no need to delete old manuscripts. However, you should distinguish whether the material is suitable for current decision-making. Add dates to policies and figures, and mark premises that have already changed as "reference only." This allows you to keep them as a history while preventing them from being mixed into current drafts.

Label each manuscript with its role

• Original: What you actually said or wrote

• Basis: Public documents, meeting minutes, statistics, etc.

• Reference Example: Completed manuscripts used for structure or tone

• Expired: Things no longer in use but kept for historical context

The labels don't have to be perfect. The goal is to ensure that anyone looking at them later doesn't get confused about how to provide them to the AI.

Perform the audit in small steps before starting a new project

Trying to organize your entire archive at once will cause you to stall. Narrow the scope to just the five articles you will use for your next piece, or just the memos related to this month's theme. Each time you do this, you will see more clearly what to keep and what has expired.

It is fine to hand over only the "excerpts" after organizing them

Reading a massive amount of past manuscripts does not necessarily make the AI's answers sound more like you. Select and provide the decisions, evidence, and examples relevant to the current request. The quality of context is determined by how you select it, not by the volume you save.

The basic philosophy of accumulating context is summarized in this article.

The method for summarizing your decision-making criteria and tone on a single sheet is continued in How to create your own "AI instruction manual"—putting your decision-making criteria and tone on one sheet.

▼Read also

- Old context leads to AI errors: Review rules for updating premises

https://note.com/poliplus/n/n87d5e090259b

- AI operational design for repurposing a single video across multiple media: Decisions and responsibilities left to humans

https://note.com/poliplus/n/nfa84bc37af7a

▼ Summary of this theme

- Summary of AI x Content Creation Know-how

https://note.com/poliplus/n/ncd6194c20727