Quick answer
To stop AI using an old version of your document, keep only the current version in a “current” folder, move older versions into an archive folder, name files with a version number and date (for example proposal_v15_2026-11-22), and tell the AI exactly which file to use. Start each new version as a copy rather than editing over the old one, and check prices and dates before anything goes out.
Key takeaways
- Out of sight is safest. If the AI can see five proposals, it may copy from the wrong one.
- Names should make the latest version obvious: a version number plus date beats “final_FINAL2”.
- New version = new copy. This avoids two people or agents overwriting each other.
- Name the file in your prompt. Do not leave the AI to guess.
In this guide
- Why does AI pick up old versions?
- Step 1: Give each project one “current” folder
- Step 2: Use a clear version naming rule
- Step 3: Archive old versions out of reach
- Step 4: Copy to a new version, never edit over
- Step 5: Tell the AI which file is current
- Step 6: Check the key figures before sending
- Which naming rules work best?
- Frequently asked questions
- Related posts
This guide follows on from Long-Running AI Agents: Why Yours Still Needs Everything Explained Again, our write-up of a video by Nate B Jones. One of his warnings is easy to recognise: an AI drafts a polished proposal using an old example and quietly includes last month’s price. Here is a simple folder and naming routine that prevents it.
📘 Main article: See why long-running AI agents need current decisions and version rules in Long-Running AI Agents: Why Yours Still Needs Everything Explained Again.
Why does AI pick up old versions?
Because it cannot tell which file you think is current. AI tools look for the most relevant example they can find, and an older proposal with a similar title often looks just as relevant as the new one. If both sit side by side, the choice is down to chance.
Step 1: Give each project one “current” folder
For each project, create two folders:
- 01-current – only the latest version of each document
- 99-archive – everything older
The numbers keep “current” at the top of the list. If you use Google Drive, Dropbox or OneDrive, the same structure works.
Step 2: Use a clear version naming rule
Use one pattern for every file: name_vNumber_YYYY-MM-DD. For example, clientx-proposal_v15_2026-11-22.docx. The year-month-day date sorts correctly, and the version number removes any doubt.
💡 Tip: Write the rule in a short README or decision log in the project folder, so teammates and AI agents follow the same pattern. Our guide on How to Keep a Decision Log Your AI Agents Will Actually Follow shows how.
Step 3: Archive old versions out of reach
When a new version becomes current, move the old one into 99-archive straight away. If your AI tool is connected to a folder, consider giving it access only to 01-current and the decision log, not the archive.
Step 4: Copy to a new version, never edit over
To make a change, duplicate the current file, raise the version number and edit the copy. Nate describes this as the knowledge-work version of what engineers call separation of concerns: the agent starts version 15 as a fresh copy rather than reworking version 14, then flags it for review.
This also stops two people, or two agents, editing the same file at once. If several bots share the work, see how to split work between several AI bots.
Step 5: Tell the AI which file is current
Be specific in every prompt:
“Use only clientx-proposal_v15_2026-11-22 as the source. Ignore any other proposals. The current price is £450. If you find a different price anywhere, stop and ask me.”
⚠️ Warning: Do not rely on the AI to remember which version was current in an earlier chat. Each new chat or agent run should be told again, or read it from the decision log.
Step 6: Check the key figures before sending
Before sending, check the five things that most often carry over from old versions: price, dates, client name, scope and terms. Use your browser or editor’s find tool to search for the old price or old client name. It takes a minute and catches the most expensive mistakes.
Which naming rules work best?
| Naming style | Example | Verdict |
|---|---|---|
| Version + date | proposal_v15_2026-11-22 | Best: clear and sorts well |
| Date only | proposal_2026-11-22 | OK for one-person projects |
| Words | proposal_final_v2_NEW | Avoid: impossible to tell which is latest |
| Initials added | proposal_v15_2026-11-22_GK | Handy for teams reviewing each other’s work |
Next step: pick one active project, create 01-current and 99-archive folders, rename the latest file with a version number and date, and move everything else to the archive. Then read Long-Running AI Agents: Why Yours Still Needs Everything Explained Again for the bigger picture.
Frequently asked questions
Why does ChatGPT or Claude use an old version of my document?
Usually because the old file is still visible alongside the new one, or because an earlier chat referred to it. Archive old files and name the current file in your prompt.
Should I delete old versions?
No. Archive them. You may need them for reference or to show a client what changed. Just keep them out of the AI’s working folder.
What is the best way to name document versions?
A short name, a version number and a year-month-day date, such as proposal_v15_2026-11-22.
Does this work for non-technical work?
Yes. It works for proposals, price lists, course outlines, email sequences and slide decks.
Related posts
- Long-Running AI Agents: Why Yours Still Needs Everything Explained Again
- How to Keep a Decision Log Your AI Agents Will Actually Follow
- How to Split Work Between Several AI Bots Without Overlap
