The short answer.
ChatGPT, Claude, Copilot and Gemini can each carry context from one chat to the next in three ways: memory, which picks up facts about you as you talk; standing instructions, which you write once and every chat follows; and shared spaces that hold the files and instructions for one piece of work. Let your assistant remember your role, your preferences and the background you would otherwise type again, and look at what it holds now and then. Do not let one person’s memory be the only home of a method the whole team follows: that method belongs in a shared workflow that everyone runs and one named person keeps current.
Most people use their assistant like basic Excel: type, get an answer, start again tomorrow. These features are an easy first step toward advanced use.
Three features carry context from one chat to the next.
An AI assistant forgets everything between chats unless one of three features carries the context across.
- Memory.The assistant notes facts about you as you talk, or when you ask, and uses them in later chats. Memory accumulates; you do not write it.
- Standing instructions.A short text you write once in settings: your role, your company, your tone and format. Every new chat follows it.
- Shared spaces for one piece of work.A space holding the files and instructions for one subject, such as a client or a recurring report. Every chat inside it starts from that material, and colleagues can often be invited in.
What each maker calls them.
All four makers offer the three features under their own names. Makers rename and move these features often; what follows was read on the makers’ own help pages in November 2026. Google, for one, is moving its Gems into a new feature it calls skills.
- ChatGPT.Saved memories and Reference chat history carry facts between chats, and you can correct them in settings. Custom Instructions set your standing preferences. Projects keep chats, files and instructions together, can be shared, and can use project-only memory.
- Claude.Memory is built from your chats, kept separately for each project, and you can read, edit or delete it. Instructions for Claude apply to every conversation. Projects hold project knowledge and project instructions and can be shared on a company account, where owners can switch memory off for everyone but cannot read anyone’s memories.
- Copilot.Saved memories keep what you ask Copilot to keep, and Copilot also learns from your chat history. Custom instructions set your standing preferences. Microsoft Copilot Notebooks gather reference files and their own instructions in one shareable place. Microsoft calls its memory features a preview and says administrators cannot restrict what kind of information goes into memory.
- Gemini.Memory, under Personal Intelligence, learns from your past chats. Instructions for Gemini apply to every chat on a personal account; on a work account the equivalent is Your instructions for Gemini in Workspace, which does not sync with it. Gems carry their own instructions and Knowledge files and can be shared. Google says telling Gemini to forget a topic does not always work perfectly; deleting the related chats does.
Developer tools have their own versions. In Claude Code, Anthropic’s coding tool, a team keeps CLAUDE.md files of instructions beside its code, and auto memory keeps notes from corrections. The other makers’ coding tools have their own versions. None of these live in the everyday chat product.
The advantage: you stop repeating yourself.
The first advantage of memory and standing instructions is that you stop retyping who you are, what your company does and how you like work set out. With standing instructions written once, the first reply already uses the house style.
The second advantage is consistent output. Standing instructions that say “start with a two-line summary, then the detail” produce replies in that shape every time, and a shared project with the current price list gives everyone the same starting material.
The trade-offs.
Memory and standing instructions bring four real trade-offs.
- Memories go stale or are wrong.Memory keeps what you said last spring after it stopped being true: a former role, a retired product, a lost client. The assistant then answers from an old fact. Read the memory settings now and then.
- Memory is hard to see.Standing instructions are a text you can read. Memory accumulates in the background, so you often cannot tell which remembered fact shaped a reply. ChatGPT, for one, shows the sources of a remembered fact beneath the reply that used it; read those notes when a reply surprises you.
- Some things should not be stored.Client confidences, staff matters, passwords and personal details about other people do not belong in an assistant’s memory. ChatGPT’s temporary chats and Claude’s Incognito chat leave memory out, Copilot and Gemini have their own controls for this, and on a company account an owner or administrator can switch memory off.
- One person’s memory is not the team’s method.Memory and personal standing instructions belong to one account. When that person is away or leaves, a method that lived in their memory goes with them, and a colleague doing the same task gets a different result.
A team’s method belongs in a shared workflow.
A method the whole team follows belongs in a shared workflow: written instructions the team agrees on, kept where everyone who does the work can read and run them, with one named owner who keeps them current. Memory is personal and invisible to colleagues.
Your own preferences, such as tone, format and the background of your role, go in your memory and standing instructions. Anything a colleague doing the same task should do the same way, such as the steps, the checks, the source records and who approves the result, goes in a shared workflow. The shared spaces in all four assistants are a good home for such a workflow and its reference files. When the know-how sits with one person covers getting a method out of an expert’s head, and Should we standardize on one AI assistant? covers why one company account makes sharing simpler.
What to check in your own assistant.
Work through this list in your own account, then ask each team lead to do the same.
- Read what memory holds.Open the memory settings, or ask the assistant what it remembers. Delete anything out of date.
- Write your standing instructions.Your role, the company in two sentences, who your customers are, the tone and the format you want.
- Remove what should not be there.Anything confidential about clients, staff or other people.
- List the methods that live only in someone’s memory.Ask each lead: which task does one person do well because their assistant “knows how we do it”?
- Move each of those methods into a shared space.Write the steps, attach the reference files, name the owner.
- Decide the company setting.Whoever administers the company account decides whether memory is on for everyone.
An insurance brokerage moves a method out of memory.
Take a 45-person insurance brokerage where one account manager writes the best renewal letters in the office. Over a year of chats, her assistant has learned how she sets out a renewal: policy changes first, then what the client should review, then the questions to answer.
Illustrative example.
- The problem.Two other account managers write renewals in their own assistants, each in a different shape, so clients receive three styles of letter from one brokerage.
- The check.The operations lead asks her to open her memory settings. Half is her own preference; the rest is the renewal method, including a carrier the brokerage no longer uses.
- The move.She writes the renewal method into a shared project, attaches the checklist and two approved letters, and deletes the stale fact. She is the named owner of the method.
- The new routine.Every account manager writes renewal letters from the same project and checks each letter before it goes out. Each keeps a personal tone in their own standing instructions; the method itself lives in one shared place.
Where FullyPowered AI comes in.
Helping a company level up what it gets from the AI it already pays for is what we do: we plan it with you, develop the workflows your teams share, and deploy them in your own assistant. The plan lives in FullyPowered Planner, and How organizations plan their use of AI sets out the whole method.