Skip to content
Agentic Levels
Start hereLevelsTracksPromptsAbout
Get my levelGet my level

  • Start hereYour first useful task with AI.
  • Level quiz8 questions. No AI needed.
  • Fluency ScoreYour Level and a score, from your own AI.

  • LevelsThe 11 levels, in order.
  • LessonsShort lessons, filter by level.
  • TracksGuided paths by tool, role or goal.
  • Set up your AI tool
  • Starter packFree files that help AI remember.

  • PromptsReady prompts to copy and adapt.
  • ChatGPT vs Claude vs GeminiPick the right one for the job.
  • How-tos
  • ReferenceCommands for Claude Code, Codex, Antigravity.
  • ResourcesHand-picked reading and tools.

  • Your progress
  • Your next moveAn interview that picks your next skill.
  • Your next lessonsThe next lessons at your level.
  • Updates

  • Pricing
  • About
  • FAQ & Feedback
  • Work with us
  • How we handle your data
Sign inPreferences
AboutPricingHow-tosPromptsStarter packMy stackHow we handle your data

© 2026 Fuentes Studio

PrivacyTermsRefundsAccessibilityAI content policyCredits
yourCouncil
Ready to help
✦

What do you want to understand?

Ask anything about what you're learning.

←Claude Cowork
  1. Tracks
  2. ›Claude Cowork: Block 6
  3. ›Deepen your voice file over several sittings
3

Level 3 · Lesson 13 of 37 · Free

Deepen your voice file over several sittings

Optional: the 100-question version, in sittings of about 30 minutes

A short voice file gets you most of the way. The last 20 percent, your opinions, your hard nos, the small habits you cannot name, only comes out when someone asks you a hundred specific questions. You do not have to answer them in one go.

30 min a sitting · You'll finish one sitting and know how to complete the rest over days

You'll run one 25-question sitting, save the answers to a VOICE WORK folder, and see how the last sitting compiles everything into a richer voice.md.

Step 1 of 5: The idea

Your steps

Step 1 of 5

The idea

The one idea this lesson is about, and why it matters.

Before you start

Complete Start your voice file first. You need your short voice.md in ABOUT ME, because this lesson builds on it and keeps it as voice-short.md.

Quick recall · Start your voice file

From the short voice file: what was the rule for where a voice file may be stored, and what makes this longer version even more sensitive?

Keep it on your own computer, never in a shared or public place. The long version also holds your exact wording on many topics, so it is an even better guide for someone imitating you.

This lesson is optional. Skip it if your short voice.md from the last lesson already gives drafts you are happy with. It is for people whose drafts still miss: the long interview adds detail the short one cannot hold, such as your opinions, your hard nos and your small habits.

It takes about five sittings of 30 minutes: four of 25 questions each (100 in all), then one to compile. Spread them over days if you like. Nothing is lost between sittings because each one ends with a saved file.

Why sittings, not one afternoon. A very long chat has a limit: Claude can only hold so much of a conversation in mind at once.

Early answers fade as the chat grows. Your own answers also get thinner when you are tired. One sitting, one saved file and one fresh task for the next is the cleaner way.

Here is the plan. Sitting 1 covers how you write: openings, closings, rhythm and formatting. Sitting 2 covers words and taste, what you love and what makes you cringe.

Sitting 3 covers personality and tone: humor, disagreement, excited and serious. Sitting 4 covers what you believe, what you refuse, and how you judge whether something is good.

Where the answers go. Your instructions tell Claude to read every file in ABOUT ME before each task.

A 15,000-word archive there would be read every time and crowd out what matters. So make a second folder next to it called VOICE WORK. Keep the raw answers there, and keep only the finished voice.md in ABOUT ME.

A worked sitting. Dev is a freelance copywriter. In sitting 1 Claude asked: You said you never use exclamation marks. Show me a sentence where one would be tempting and what you wrote instead.

Dev wrote: The launch went well! became The launch hit its number. Claude saved that pair word for word in voice-archive.md in VOICE WORK. At the end of the sitting Dev had 25 answers saved and closed the chat.

Two days later Dev started a fresh task, told Claude to read the archive first, and carried on from question 26.

The compile. After sitting 4, a fresh task reads the whole archive and writes a new voice.md of about 800 words, a little past the 600 of the short file because it adds your hard nos and examples. The fuller version, 1,000 to 1,800 words, is saved in VOICE WORK for reference, and Claude reads only the 800-word one each task.

Each line must pass one test: if this line disappeared, would a draft change? If not, it is cut. Keep your short file from the last lesson as voice-short.md, so you can compare the two on the same message.

Keep it private, more than ever. The archive and the final file together describe how to imitate you very convincingly.

Keep both on your own computer. Do not put them in a shared drive, a public online folder or a post.

One sitting, one saved file, one fresh task for the next. The compile comes last.

That is the idea. Next you watch it work.

Step 2 of 5

Watch it

See one real prompt do one real job, then see which part of it made that work.

You type this

This is sitting 1 of 4 in a long interview about how I write. Ask me 25 questions about how I open, close, shape and format my messages, one at a time, numbered. Push back on vague answers and ask for a real sentence. When I say stop, add my answers, word for word, to voice-archive.md in a folder called VOICE WORK, and tell me the next question number.

Claude does this

Claude goes next.

Step 3 of 5

Your turn

Write one part of the prompt yourself, then compare with one good answer.

What to do before asking

This is sitting 2 of 4. Read VOICE WORK/voice-archive.md first, then . Ask 25 questions about my words and taste, one at a time, and add my answers to the same file.

Sitting 2 starts in a brand new task, so Claude remembers nothing of sitting 1. What should it do so it does not repeat questions and numbers them correctly?

Your answer is saved with your progress, so leave out anything private.

Try it before you go on.

Getting it wrong first is what makes it stick.

Step 4 of 5

On your own

Do the task on your own work, in your own tool.

Run sitting 1 on your own writing now, 25 questions, and save voice-archive.md in VOICE WORK. Then write the prompt for sitting 3 yourself (personality and tone) in two lines, using the sitting 2 prompt as your shape, and put a date in your calendar for the next sitting.

  • Breaks when you compile before all four sittings are done, because the file then reflects only the early topics. Finish the sittings, or accept a smaller file and say so.
  • Breaks when the archive is so long that one task cannot hold it comfortably, and the compile skips parts. If that happens, ask Claude to compile each sitting into a half-page summary first, then compile the four summaries.
  • Breaks when a richer file is not better for your work. For short replies, the 300-word file can win. That is why you compare the two on the same message instead of assuming.

Go do it for real.

Come back with a result.

Leave out anything private. This uses one of your Course Council messages.

Step 5 of 5

Remember it

Say it from memory first, then check it.

Try it from memory

Pulling this out of your own head now is the part that actually moves it into long-term memory. Re-reading only feels like it does.

Without scrolling up: why does the raw archive live in VOICE WORK instead of ABOUT ME, and what do you tell a fresh task so sitting 2 carries on from sitting 1?

One good answer: 'Claude reads ABOUT ME before every task, and a huge archive there would crowd out the rest. In the new task I say: read voice-archive.md first and number new questions from where it ends.' If you mention both the crowding and the read-first line, you have it.

You can now

✓

You can run one long interview sitting and save the answers where Claude will not over-read them

  • ✓You can resume a sitting in a fresh task by pointing Claude at the archive
  • ✓You can compile an archive into a richer voice.md and keep the short version
  • ✓You can compare two voice files on the same message

Key takeaways

A long interview works when it is cut into sittings, each ending with a saved file. Compile last, keep the short version, and compare before you trust the longer one.

  1. 1This lesson is optional. If the short voice.md already works for you, stop here.
  2. 2Cut the interview into four sittings of about 25 questions, about 30 minutes each with saving, and save after every one.
  3. 3Keep raw answers in VOICE WORK, outside ABOUT ME, so Claude does not read them before every task.
  4. 4A fresh task has no memory. Tell it to read the archive and number from where you stopped.
  5. 5Compile last, keep voice-short.md, and compare both on a real message.

AI can help with this

Open a new task and say: This is sitting 1 of 4 of a long voice interview. Ask me 25 questions one at a time and save my answers word for word in VOICE WORK/voice-archive.md. Claude runs the sitting and tells you where to resume.

Go deeper

  • Ruben Hassid's article that started this idea (our version is rewritten from the vendor docs)ruben.substack.com (opens in a new tab) ↗
  • Claude Cowork: get started (folders and instructions)support.claude.com (opens in a new tab) ↗

That is the lesson. Mark it complete, and its question comes back to you in three days.

Was this helpful?

Up nextManage your Cowork context files without a terminal→

Related lessons

Start your voice fileSet up your Cowork folder so Claude knows you on day oneManage your Cowork context files without a terminal
← Back to Claude Cowork