What Does Your AI Tool Believe About Your Project?
AI memory is now switched on in many of the tools project managers use at work, so your assistant is building a picture of your project from your chats and the files it can reach. To check that picture, ask the assistant you use most to list what it has stored or can see about the project, open its memory settings, and compare both with what you know is true. Fix the stale items in your own project record first, then in the tool. It takes about ten minutes, and it shows you what every draft the tool writes for you is built on.
AI Memory Is Already On in Your Work Tools
In Microsoft 365 Copilot, memory is on by default unless your IT team has turned it off. It keeps details you’ve asked it to remember, plus details it infers from your chat history, and uses them in later answers. Google’s new Gemini agent for work, announced on 8 October, keeps a knowledge base it builds from the documents and people it works with.
Memory saves you re-explaining the project every time you open a chat, which is why it’s on. But a detail the tool picked up in July sits beside one from last week, and nothing in a draft tells you which one it used. A tool still working from a sponsor priority that changed two months ago writes a reasonable-sounding update that argues the wrong case.
Memory import, which Claude and Gemini both added in March, moves a profile of you between assistants. The project record stays inside each tool, in whatever state you last left it.
Ask What Your AI Remembers, Then Open the Settings
Start with the assistant you use most for project work:
List what you have stored in memory, or can see in the files and instructions available to you here, about [project name]: scope, the people involved and what they care about, decisions made, open risks, and how I like things written. For each item, say where it came from and roughly when. Mark anything you’re unsure of, and write “not known” for anything you can’t point to, rather than guessing.
Asked what it “assumes”, a model can fill gaps with guesses that sound right, so the prompt asks for a source for each item and permission to say it doesn’t know. The chat answer is still the quick read. The record is in the settings, so open those too. As of October 2026:
Microsoft 365 Copilot: Settings, then Personalization. You can delete saved memories there. Details inferred from chat history only go when you delete the chats they came from. ChatGPT: the memory settings, plus each project’s own memory setting. Claude: Settings, then Memory, where each topic can be read, edited or deleted. On Team and Enterprise plans memory stays off until your organisation and you both switch it on. Gemini in Google Workspace: your conversation history, which is kept separately in each app.
If your tool has no memory switched on, run the same prompt inside the project or notebook you use. It shows which files the tool is drawing on, and an old plan sitting in a shared folder misleads it just as well.
Before you run this on a client project, check which account you’re signed in to. A personal assistant that holds client details holds them outside your organisation’s controls, and your policy may say they shouldn’t be there at all. ## Fix the Record First, Then the Tool
Correct your own project document first. Context Is the Skill sets out what belongs in it: a project brief, a stakeholder map, the risk history and a decision log, kept in Word, Google Docs or a wiki page you own. Then delete or correct the stale memory in the tool. Correcting the tool alone leaves the next tool, and the next person on the project, with the old version.
Then point the tool at the document so it starts from the current record. Most assistants have a place for this: a ChatGPT or Claude project, a Copilot Notebook (Microsoft says its references stay current as the files change), or saved instructions. For client work in ChatGPT, set the project to project-only memory so it doesn’t draw on your other chats.
What the tool believes about your project, and whether it’s current, is something only you can judge. Retention, audit and admin settings belong to IT. If the retention and audit gaps concern you, raise them there rather than working around them.
When AI Drafts the Status Report set a weekly check on what your tools produce. This one checks what they’re working from. Run it again when the project re-baselines, when a sponsor changes, or when you start using a new tool.
Frequently Asked Questions
How do I see what my AI assistant remembers about my project? Ask it to list what it has stored or can see about the project, with a source and rough date for each item and “not known” where it can’t point to one. Then open its memory settings, because the settings page is the record and the chat answer is a summary.
Should client project details be in my AI assistant’s memory? Only in the account and tools your organisation has approved for that client. Details saved in a personal account sit outside your organisation’s controls, so check your policy before running the check on client work.
Ask your assistant what it believes about your project this week. What’s the first thing on the list that’s out of date?
Ask your assistant what it believes about your project this week. What’s the first thing on the list that’s out of date?
Yes, AI helped me to write this :)
Unsubscribe
