5 min read

AI & Agents

Claude Gives You Generic Answers Because It Knows Nothing About You

Every new Claude chat starts with a stranger. Here is the four-folder setup that gives it your voice, your projects and your templates once, so you stop re-explaining yourself.

why Claude gives generic answersClaude Cowork folderClaude context

Claude Gives You Generic Answers Because It Knows Nothing About You · Products Decoded
By Debashis Nayak · Products Decoded20 Mar 2026

If you have ever pasted the same brief into Claude for the third week running, you have already felt the problem. You re-explain your tone, re-upload the old context, remind it what you do, and then the answer comes back sounding like everyone else’s.

I used to blame the model for that. I now think the model was doing its job. It had nothing about me to work with.

If Claude’s response is too generic, what should you try?

Three things, in this order.

  1. Tell it who you are and what the answer is for, such as your role, your audience and the decision it feeds.
  2. Give it a constraint or an example of what good looks like: a length, a format, or one finished piece to copy the shape of.
  3. Ask it to put questions to you before it answers, and list what you do not want to see.

Those three fixes patch one chat. The setup below makes them permanent.

Why are Claude’s answers so generic?

Claude does not carry you from one session to the next. It does not know your name between chats, your writing style, your brand voice, your active projects or the formats you like your output in. Each new chat is the first time it has met you.

So the generic answer is what you get from a stranger. If you keep paying for that with a long paste of background at the top of every session, you have built an expensive habit loop and called it a workflow.

A better prompt does not fix that. What fixes it is a better setup, and it is a small one.

What is the Claude Cowork folder system?

It is nothing exotic. It is the oldest rule in software projects applied to your own work: keep your context, your execution and your outputs apart. Mine looks like this.

Claude-Cowork/
├── ABOUT ME/
├── PROJECTS/
├── TEMPLATES/
└── CLAUDE OUTPUTS/

Four folders, one job each.

FolderJobWhat goes in
ABOUT METeaches Claude who you areAn about-me.md with your role, tone, audience and preferences, plus an anti-AI-writing-style.md listing the words, phrases and patterns you hate
PROJECTSHolds one subfolder per active projectThe brief, references and working drafts for each
TEMPLATESHolds the shape of your best workFinished pieces you want repeated: a proposal that landed, a report layout stakeholders liked, a post structure that kept performing
CLAUDE OUTPUTSThe only folder Claude writes toEverything it produces

ABOUT ME is what stops the generic default

Without this folder Claude defaults to generic. With it, Claude defaults to you. The anti-AI style file does a lot of the work, because it tells Claude what you would never say before it says it.

PROJECTS moves context from the session to the task

When you open a session you are not explaining anything. You point Claude at the project folder and treat it as already briefed. As a PM I find this the part that scales: context per task rather than context per session.

TEMPLATES holds the shape, not the content

Claude does not need your old content. It needs the structure of your best work. Hand it the layout that worked and let it fill in what is new.

CLAUDE OUTPUTS keeps the rest clean

Everything else in the system is read-only context. Outputs go in one place, so they stay searchable and separate from your source material. There is nothing technical in this. It is hygiene, and clutter in the output folder is what kills a reusable system.

How do you connect the folder to Claude?

This is the step people skip. Building the folders and stopping there gets you a tidy desktop and not much else.

The step that actually changes things, at least as Claude works in March 2026, is connecting the folder to Cowork:

  1. Open Cowork.
  2. Click Add Folder.
  3. Select the Claude-Cowork folder.

Now Claude works inside a system you built for it, so it has much less to guess. For any session with real complexity, I pair this with Opus 4.6 and Extended Thinking. Treat the model names as a snapshot of that month, since they change.

Why does this work?

From a PM’s seat, I read it as dependency injection for your AI workflow. You are not changing the model. You are changing what it has access to at runtime, and good context in gives much better output out. I do not think that is a hack. It follows from how context windows work.

The common mistake is treating Claude like a smart search engine you prompt on demand. I think it behaves more like a collaborator whose work tracks the quality of the environment you give it. Two examples of what I mean:

  • One well-structured about-me.md beats fifty random uploads.
  • One project brief beats twenty messages explaining background.

I cannot give you a benchmark for either. It is what I have noticed, and it is easy to test on your own work: run the same task once cold and once from the folder, then compare.

What if you work from the command line instead?

The same principle holds on a different surface. If you run a CLI-based Claude setup, this folder structure maps onto a context.md plus a project-scoped RAG pipeline. The idea underneath is identical: structured context treated as a first-class artifact, not an afterthought you retype.

Your setup checklist

  1. Create the four folders, ideally in one parent folder you can point Claude at.
  2. Write a short about-me.md: role, tone, audience, preferences.
  3. Write down the words and patterns you refuse to see in your own writing, and save them as anti-AI-writing-style.md.
  4. Add one subfolder per live project, each with its brief.
  5. Drop in two or three finished pieces as templates.
  6. Connect the folder through Add Folder in Cowork.
  7. Update the files when your work changes.

Build it once and revise it as your work evolves. You will spend less time re-prompting and more time shipping, and that is the whole point of doing it.