Which tool to choose

Project spaces exist on the main assistants, under different names: Projects on ChatGPT and Claude gather instructions and files shared by all the project's chats; Gemini's Gems are custom assistants that can connect to your documents. Claude, in particular, is designed to hold a sizable knowledge base. If you use a tool without this feature, the substitute is a context document of your own, to be pasted into every new chat of the project.

How it's done

  1. Create a project dedicated to the work.
  2. Put in it the project instructions (goal, tone, rules) and the reference files (brief, documents, examples).
  3. Open a separate chat for each strand of the work, all inside the project: this way each starts already with the shared context in view.
  4. When in a chat you make an important decision, write it into the project files, so the other conversations inherit it.

A concrete example

Giulia was writing a novel with the AI's help. Instead of a single chat that lost the names and subplots after a few scenes, she created a project with the outline, the character sheets, and the world's rules inside. Then she opened a chat per chapter. Each conversation already knew the characters and plot because it read the project files, and when she defined an important new detail she added it to the sheets. The novel stayed coherent from start to finish, without her having to repeat the context for every chapter.

When it does NOT work (and how to fix it)

If one chat doesn't know what happened in another

A project's conversations share the common instructions and files, not each other's messages. If an outcome reached in one chat has to apply to all, don't leave it there: write it into the project files. The shared base is the channel that links the chats, not their direct content.

If the project is too loaded with material

Every project space has a limit to how much context it can hold. If you stuff everything in, it slows down or loses pieces. Select the documents that are truly relevant instead of loading the whole archive: better a few useful files than a mountain that smothers it.

If your tool doesn't have project spaces

Use a context document as the glue: a file with goal, decisions, and status, that you paste at the start of every chat of the project. It's more manual, but it achieves the same result and works everywhere.

A tip from someone who actually uses it

Treat the project files as the source of truth of the work: every important decision made in any chat, bring it there. That way the project isn't just a container of conversations, but a living memory that grows and that all the chats inherit. It's the difference between a messy archive and a worksite where every crew works on the same updated plans.

Frequently asked questions

Do the chats of the same project talk to each other?

Not directly: they share the common context — instructions and files — but they don't read each other's messages. To propagate a result from one chat to the others, you write it into the project files. It's that shared material that acts as the link.

How many conversations can I keep inside a project?

Many: that's precisely the advantage over the single chat. You can dedicate one conversation to each part of the work without losing the overall thread, because the common context stays a single one and all of them draw on it.

Better a project or a single big chat for a long job?

For a big job, the project: the single chat gets confused when it grows long, while the project separates the strands and keeps the context in one place. The mega-chat is fine for a short task, not for something that lasts weeks and has several parts.

Does a project remember everything I say in each of its chats?

No, and it's the misconception to avoid. The project shares the instructions and files you put in it, not every word of every conversation. What you want to keep and propagate, you must consolidate into the files: the project is a common base to curate, not a recorder that automatically retains everything that's been said.