Showing posts with label ChatGPT. Show all posts
Showing posts with label ChatGPT. Show all posts

Monday, November 16, 2026

Teach ChatGPT and Claude Once, Instead of Every Single Time

You have probably had this conversation with an AI chatbot more than once. You explain who you are, what you are working on, how you like things written, and the tool gives you something useful. The next day you open a fresh chat and start the whole explanation over, because the assistant has forgotten everything. That repetition is the single biggest time sink in how most people use ChatGPT and Claude, and both tools have a feature built to end it. The feature has different names and a few different rules, but the idea is the same: tell the assistant your context once, and have it apply automatically to everything you do afterward.

The feature is called Projects. In both ChatGPT and Claude, a Project is a workspace that groups many separate chats together and gives all of them the same set of background instructions and uploaded files. Claude shipped the idea first, and ChatGPT now has a close equivalent. Think of a Project as a labeled folder that also remembers the ground rules. Every conversation you start inside it already knows what you told the folder.

Set it once. It sticks.

The two things worth setting up

There are two separate controls, and people tend to confuse them. The first is standing instructions. ChatGPT calls these Custom Instructions, and Claude handles the same job through its Preferences and writing Styles. These are short notes about you that apply across your chats, things like "I am a high school biology teacher," or "write in plain English, no jargon, and keep answers under 200 words unless I ask for more." You write them once in your account settings, and the assistant reads them every time.

The second control lives inside a Project: the Project instructions and the files you attach to it. This is where you put context that matters for one body of work but not your whole life. If you are planning a wedding, a wedding Project might hold the guest list, the budget spreadsheet, and an instruction that says "help me stay under 12,000 dollars and flag anything that pushes past it." If you are job hunting, a Project might hold your resume and a note about the kinds of roles you want. Claude reads the files you attach to a Project on every conversation in it, so you never re-paste your resume again.

This is the practical difference people miss. Custom Instructions are about who you are in general. Project instructions and files are about one specific thing you are working on. Use the first for your durable preferences and the second for the context that would be noise anywhere else.

How the two assistants differ

They behave differently in ways that matter once you lean on them. Claude's approach is more document-driven and more under your control. The files and instructions you put in a Project are the memory, and Claude reads them each time, so what persists is exactly what you decided to put there. ChatGPT leans more on automatic memory, where the model itself decides what to remember from your conversations over time. Neither is better in the abstract. If you want predictability and you like knowing precisely what the assistant is working from, Claude's explicit style fits. If you would rather the tool quietly pick things up as you go, ChatGPT's automatic memory fits.

Size is the other practical gap. As of 2026, Claude can hold roughly 200,000 tokens of context in a conversation, which works out to around 150,000 words. ChatGPT works with about 128,000 tokens, or roughly 96,000 words. A token is just a chunk of text, a little smaller than a word on average. The number only becomes real when your uploaded files get long. More room means you can give Claude a bigger pile of reference material before it has to start skipping parts, and that matters if your Project holds long documents.

That leads to the one mistake worth avoiding: dumping everything in. It is tempting to upload every file you own, but a focused Project beats a crowded one. One analysis suggested keeping a Project's documents to something like 25 pages, which is a tiny fraction of the available space and leaves the assistant plenty of room to reason. The goal is not to feed the tool your entire hard drive. It is to give it the handful of documents that actually bear on the work, so its answers stay on target.

Writing instructions that actually work

Vague instructions produce vague help. "Be helpful" does nothing. A useful pattern, suggested by people who set these up for a living, is three layers. First, the role: tell it who to be, like "you are my editor for a monthly community newsletter." Second, the output: say what you want back, like "give me a headline, a 150-word summary, and three bullet points." Third, the boundaries: say what to avoid, like "no exclamation marks, no buzzwords like 'synergy' or 'circle back,' and ask me a question if a date or name is missing." Those three lines will change the quality of everything that Project produces.

A concrete example. Suppose you manage a small nonprofit and you are tired of rewriting donor thank-you emails. Make a Project called Donor Emails. Attach your organization's one-page style note and a couple of past emails you liked. Set the instructions to: "You help me write warm, specific thank-you emails to donors. Keep them under 150 words. Always mention the specific program the gift supports. Never sound like a form letter. If I don't tell you the donor's name or gift amount, ask." From then on, every email you start in that Project begins with all of that already loaded.

One honest caveat. These features sit behind the paid tiers for the deeper capabilities, and some sharing options, like team workspaces in Claude, cost more per seat. The free versions of both tools still let you do a lot, but Projects as described here are mostly a paid-plan feature as of late 2026, so check what your plan includes before you build a workflow around it.

If you take one thing from this, make it a single Project this week for the task you repeat most. Spend 20 minutes writing its instructions and attaching two or three documents. The next time you sit down to that work, notice how much you did not have to explain. That gap, between re-explaining yourself every time and never again, is the whole point, and it is sitting in a feature you may already be paying for.

References

Lett, J. (2026, June 10). ChatGPT vs Claude vs Gemini: Feature and terminology map (2026). Bootstrap Creative. https://bootstrapcreative.com/chatgpt-vs-claude-vs-gemini-terminology/

Velani, S. (2026, April 12). Claude Projects vs ChatGPT Custom GPTs: 2026 verdict. Builts AI. https://builts.ai/blog/claude-projects-vs-chatgpt-custom-gpts/