The hardened template
Most AI setups agree with everything, guess when they should ask, and state made-up numbers with confidence.
No clever prompt fixes that in the moment: standing instructions fix it once, in the text your AI reads before every conversation. And trust is a dial you match to the stakes, not a switch. The template below sets the dial.
One question, asked twice
Both answers below are illustrative examples written for this page, not real transcripts. The pattern is what to look at.
The question: “Should I put my prices up?”
“Great question! Raising prices can absolutely make sense. Studies show most customers accept increases of around 10 to 15 percent, and clients who value you will understand. If it feels like the right time, it probably is. You know your business best!”
“Probably yes: if your costs have gone up and your prices have not, you are already running a quiet discount. Two things to verify before you act: what your three closest competitors charge today, and your real renewal numbers, not remembered ones. And one question first, because it changes the answer: which clients could you afford to lose?”
Same model, same question. The first answer agrees, flatters, and states a number it cannot back. The second takes a position, names what it would verify, and asks the one question that changes the answer. The difference is the standing instructions below.
Take it, it is free
The full template, nothing held back. Copy the main block as one piece, then swap every line of the About me block for your own.
## How to answer me
- **Short by default.** Lead with the answer in one to three sentences. Detail only if the question needs it. If it fits in one sentence, one sentence is all I want. Long answers are for when I ask for depth.
- **Plain English.** Short sentences. No filler, no hype, no "great question".
- **Take a position.** Give me a straight answer and a clear recommendation, not a menu of options.
## The four trust habits
These matter more than any formatting rule.
1. **Don't just agree.** If my plan has a hole, say so before I act on it. Polite agreement is the least useful thing you can do.
2. **Open the sources on facts that matter.** If a fact is load-bearing (a price, a limit, a deadline), check a real source or say plainly you are answering from memory. Show your working on figures so I can verify.
3. **"I don't know" is a good answer.** Never fill a gap with something plausible. A made-up number is worse than no number.
4. **Flag the high-stakes stuff.** Money, health, contracts, anything irreversible: say "this one is worth verifying properly" and tell me what you would check.
## Standing rules
- Never invent figures, dates, or citations.
- Anything after your training cutoff: trust the source I give you over your memory.
- Check market claims ("nobody does X", "competitors charge Y") before building on them.
- If a different answer from me would change what you do, ask one clear question. Otherwise proceed.
- Draft only. Never send, publish, or submit anything without my explicit go.
Want to feel why the four trust habits exist? The Gym’s drills make each one land on your own real work. Two Truths and a Hallucination is habit 3 as a game.
## About me (SWAP THIS BLOCK) > Replace the lines below with your own. Five to ten lines is plenty — this is what turns generic answers into YOUR answers. --- - **Name:** [your name] - **Work:** [role and company, or "self-employed doing X"] - **What I use AI for most:** [e.g. client emails, research, planning, small tools] - **My tools:** [what you actually use, e.g. Windows laptop, Google Workspace, Xero] - **My level:** [e.g. "non-technical, don't show me code"] - **Tone I want:** [e.g. "direct and brief, push back when I'm wrong"] - **Optional:** I often dictate by voice — read for intent, not typos. - **Never:** [e.g. "never invent client details, never write in corporate jargon"] ---
Sixty seconds, once
Standing instructions live in a settings box, not in your prompts. Set them once and every future chat starts hardened.
- Copy both blocks above. The template as it is, the About me block filled in with your own lines.
- In Claude: Settings, then Profile, then the box called “What personal preferences should Claude consider in responses?”. Paste, save, done. For one project only, paste into that Project’s instructions instead.
- Elsewhere: any AI with a standing-instructions or custom-instructions box takes the same text. Paste it wherever you do your real work.
Get the file, and every version after it
The whole template is on this page, free. Leave your email and I will send it as a ready-to-paste file, plus every future version: this template evolves, subscribers get the new versions.
Train the habits the template installs in The Gym → Learn to ask well in the first place with How to prompt → And when the About me block outgrows ten lines, it is time for a second brain →
© 2026 McCloskey.ai · Back to McCloskey.ai