Why didn't Claude Code save any files?

Why this catches people

When you watched the Module 1 video, Claude Code built a whole folder and populated it with organized research files. When you tried it yourself, you got a long text response in the terminal and nothing else. That feels broken. It isn't.

The difference between the video and your session is one line of prompt. Joe asked Claude to save the output. You asked Claude to do the research. Claude did exactly what each of you said.

This is the single biggest shift from browser AI to CLI AI: the model doesn't guess what you want on disk. You have to say so.

The three things your prompt needs

When you want Claude Code to leave files on your disk, tell it three things:

• 1.What — the task. "Research this person," "summarize this PDF," "extract every quote from the transcript."
• 2.Where — the file path. "Save the summary to summary.md." Or "save it to notes/research.md." A relative path works; Claude creates folders if they don't exist.
• 3.What shape — the format. "As a bulleted list." "In markdown with headings." "As a CSV with columns for name, source, date." Shape is optional but often the difference between a blob of text and something you can use.

If you only give Claude the what, you'll get a terminal reply. If you add the where, you get a file. Add the shape and you get a file that's useful.

A real example from the forums

Eva Wolfangel, a cybersecurity journalist in the course, hit this exact problem. Her first attempt produced nothing on disk. Her second attempt, written like this, produced exactly what she was after:

Please conduct the search for [person] now. You should know
that he is likely a fraud. [Context about what she already knew.]

Please go online and look for conflicting information he has
provided about himself, as well as potential interview subjects
who have had dealings with him.

Also: could you please organize/cluster your research results
in a logical way, ideally in a file that you save in the
Project folder?

Please be thorough, follow every lead, and evaluate each one
to see if there are further avenues for research.

Notice what she did. She described the task. She gave Claude the context she already had. She asked for clustering. And she said where the output should go: "in a file that you save in the Project folder." That single clause is what turned a terminal-only reply into a saved research file.

After Claude wrote the first version, she followed up:

Great research, thanks. What would help me: if you could
link directly to the sources for each one.

And Claude rewrote the file with source links inline. This is the pattern: ask for the work, ask for it to be saved somewhere, then iterate once you see what came back. You don't have to get the prompt perfect on the first try.

Common mistakes people make instead

• ›Asking for "a report" with no filename. Claude might write a report in the terminal and call it done. "Save it as report.md" is clearer.
• ›Assuming Claude will pick a good filename. It often will, but being specific removes the guesswork — and makes the file easier to find afterwards.
• ›Forgetting you're in a folder. Claude writes files into the folder you launched it from. If you ran claude from your Desktop, that's where files land. If you ran it from a project folder, that's where they land.
• ›Blaming the tool. Claude Code on Mac and Windows behaves identically here. If your files aren't landing, the fix is almost always in your prompt, not your operating system.