Your last five questions, and the AI that answers the sixth

Jay WestOut of My Head · about 13 min

By the end of this step you have:

Your Question Log started, with real questions in it — two is a start, five is a good first day
The reason behind one answer written down, the first piece of your judgment out of your head
The AI you already use answering a sixth question the way you would, or showing you exactly what it is missing

The idea

Your business does not run on what you know. It runs on what you are asked. Every day people bring you the same kinds of questions, and every answer you give lives in one place, your head, so the next person has to ask again.

"Asked" is wide — a team member's or a client's question, an escalation, a fire on your desk, a decision nobody asked you for, or one you keep putting off. Anything that only moved because it went through you counts.

Most of it repeats in shape. Written down, the shapes show within days, and an answer with its reason is one a person or an AI can use without you.

An example

Say you run a twelve-person agency. Before lunch your account lead asks whether a client can have an extra round of revisions, your designer asks whether to start a job before the deposit lands, and a client emails asking whether you can rush their launch.

You answer all three in under a minute each. None is hard. What makes the answers yours is the reason, and you never said it out loud — extra rounds are fine for retainer clients, never for one-off projects, because retainers renew; no work starts before a deposit unless the client has paid on time three times running.

Written down with their reasons and pasted into ChatGPT, they let it answer the deposit question for a different client the way you would.

Do this now

Came in through another step? The rows you added there are your first rows; add what reached you since, and go straight to the why. Ten minutes. Log the last five questions that reached you: who asked, what they asked, what you said. Look where they arrive — texts, Slack or Teams, email, your desk, the phone. If you cannot find five, two is a start. Other rows count too: a decision nobody brought you (me | the decision | what I chose), a lookup you or your team makes (who, what they looked up, where they found it), the questions you asked the people who run what you cannot judge, and what they said ("Get it out of their head while you can still ask" has the questions), each time you corrected an AI chat or agent, and a past week typed from a paper notebook. Many people asking? Skip to "Faster" below and paste a week. Decisions reach you through managers or directors? Your rows are what they bring up — escalations, slipped milestones, priority conflicts, a call you overruled — with their role as who asked. Then pick the one you answer most often and, in one sentence, add why you answered that way. Use roles, not names — "a client", "the office manager". Your situation not covered? Open the boxes below.

Log them here. They stay on this device.

Your Question Log: day 1 of 5 · 0 questions reached you · 0 rules written

If nobody asks you, and your rows are a journal, a second brain, meeting notes, or AI agents you correct

Nobody asks you? Your rows are the decisions you keep putting off, with who, if anyone, is waiting on them. A journal, a second brain or meeting notes? Paste them with "List every decision I made or put off in these notes, one per line: YYYY-MM-DD | me | the decision | what I chose."
Work with agents or AI chats? Each time one asked you, guessed wrong about what you needed, or you stopped, corrected or overrode one is a row — who is the agent, what it was about to do, and what you told it instead. To fill the days faster, paste the transcripts with "List every time I stopped, corrected or overrode the agent, one per line: YYYY-MM-DD | agent | what it was about to do | what I told it instead."

If you handed most of it off to your leads, are not the owner, just joined, are not open yet, or set this up for someone else

Handed most of it off, or the questions go to your leads? Log what still reaches you, and ask two leads to log theirs for five days; the single point may have moved to them. Have them press "Copy my log" on Friday and send it; paste theirs in with "Paste rows instead", one code per lead in who asked.
Not the owner, and the owner does not read email? Print one rule card and ask only yes or no.
Just joined? Log the questions you have had to ask this week — who you asked, what you asked, what they said.
Not open yet? Log the decisions you make while you set up, and the five questions a new hire will ask in week one, with your answers.
Setting this up for someone else? Log the questions that reach them, from their messages with permission, or sit beside them for a day.

Faster — fill all five days today

Your last five working days are already in your messages. Paste a week of chat, email, WhatsApp or ticket comments into your AI with this — "List every question someone asked me in these messages, and every decision I made that nobody asked for, one per line, as plain lines (not a table): YYYY-MM-DD | who | what they asked | what I answered. Use roles, not names." Then use "Paste rows instead" in the log. Rows with dates count as the days they happened, so a week pasted today is five days of log.

Test it

Ask the AI a sixth question now, or wait for the test box in "Give your AI your rules" — copy your rows, ask it the next question someone really asks, and mark the answer as you would, guessed, or wrong. Guessed or wrong is the list of what is still only in your head.

No AI yet? chatgpt.com or claude.ai, free — or pick "Not yet" in the questions after this step and the course sets one up with you first.

What never goes in

Passwords, card or bank numbers, health or personal details about customers or staff, and anything under a confidentiality agreement, unless your company's approved AI tool and the contract allow it. You are writing down how you decide, not who it was about.

This week

Keep logging for five days, two minutes at the end of each.

If your team already logs in a shared sheet, a form, Notion, an Obsidian vault or a repo, or you use several devices

Team already logs in a shared sheet, Notion, an Obsidian vault or a repo? Keep the same columns there (date, who, what, answer, because), or have a routine append rows in those columns, and paste them in with "Paste rows instead" when a step needs the counts; it also reads the Markdown table this log downloads, and "Download as Markdown" drops the log into a vault as a note.
On more than one device? Capture in a notes app that syncs — Apple Notes across iPhone, iPad and Mac — one line per item, and paste it in with "Paste rows instead" each evening, then delete what you pasted from the note, so tomorrow's paste holds only tomorrow. Capture there; review here.

If you track locations, units, people or two businesses by code, or have a client meeting coming

Tracking locations, units or people? Give each a short code in the "who asked" column (U14, M3). The key stays in your own notebook, never in the log or the AI, and the counts still show where the same issue keeps coming from.
Two businesses? One log, with the company as the first letter of the who-asked code (A-rep, B-ops), so the counts split by company.
Before a meeting? Put a short code for the client or project in the "who asked" column, press "Copy my log", and ask your AI for "Everything open, owed or decided about [code], newest first." Tasks and owners already in a tracker? If your tool offers a connector, ask your AI there; the log holds what no tracker does.

If you decide which AI your business is allowed to use

You decide which AI is allowed? Business plans (ChatGPT Business or Enterprise, Claude Team or Enterprise, Copilot signed in with a Microsoft 365 work account) say in their published terms that they do not train on your data by default — confirm it in your plan's admin settings, and check your client contracts allow it. If none is allowed, keep the log here and do the AI steps by hand or with a model on your own machine, such as Ollama with Open WebUI or AnythingLLM.

This step is written for this course from the research in the full curriculum. The research behind it, on howtothink.ai (L-0442, L-0012, L-0002, L-0164)

Step 1 of 8

Next: step 2 →