Making your AI Assistant Sound like you
A real AI voice profile takes fifty questions.
Most personalisation setup produces a stranger’s draft. A structured interview, the same one built for the Impact Investors AI Circle, produces something closer to your own.
Most AI assistants write in a voice that belongs to no one in particular. The usual setup, a short bio pasted into a personalisation field and a paragraph of stated preferences, produces prose that is competent and forgettable at the same time. It reads like every other assistant’s output because it was built from the same kind of input: broad, hedged, and short on specifics.
Most people cannot describe their own writing accurately. Asked directly, they give the safe, generic version, three adjectives and a compliment to themselves. The real signal, the sentence structures actually used, the phrases that cause a wince, the topics that are off limits, surfaces only under specific questions and follow-up pressure. A structured interview is built to apply that pressure.
We built a fifty-question interview to apply it. It is Session 1 pre-work in the Impact Investors AI Circle, and it produces a document precise enough that an assistant can write in your actual voice.
A structured interview beats a profile form.
The interview runs fifty questions across seven categories, each one targeting a different part of how someone actually thinks and writes.
Beliefs and contrarian takes (8 questions)
What you believe that most peers in your field would not, and where conventional wisdom is wrong.
Writing mechanics (10 questions)
Default sentence structure, how pieces open and close, punctuation habits, words overused and words avoided.
Aesthetic crimes (7 questions)
The specific phrases and patterns that read as lazy or performative in other people’s writing.
Voice and personality (7 questions)
How humour, disagreement, excitement, and uncertainty show up on the page.
Structural preferences (8 questions)
How an argument, a deck, or a document gets organised, and where lists give way to prose.
Hard nos (5 questions)
Topics and approaches that are off limits regardless of how the request is framed.
Red flags (5 questions)
What causes immediate distrust of a piece of writing, often within the first paragraph.
The interview rules matter as much as the categories: one question at a time, push back on vague answers, ask for a real example every time, call out contradictions between answers. Skip that depth and the result reads like a stranger’s draft. Force every flagged phrase into every piece afterward and it reads like an assistant imitating someone. The test that matters: does a piece sound like something the person would actually write, or like an assistant working hard to sound like them.
A specific example beats a vague adjective.
A short, illustrative exchange shows the difference in practice.
Q: How do you organise an argument?
A: I like to keep things simple and let the evidence speak for itself.
The interview pushes on that answer: “Simple how? Walk me through a paragraph you wrote that does this well, and one where you tried and it came out flat.” The second answer is where the real pattern shows up, a claim stated first, one supporting figure with a source, then the implication, in that order, every time.
The second answer is what ends up in the finished document. A profile built from adjectives produces guidance an assistant cannot act on. A profile built from examples produces a rule.
The prompt, paste it as your first message.
Paste the block below into whichever assistant is being set up, Claude, ChatGPT, Gemini, or Copilot all work. Answer one question at a time rather than pre-answering. Budget forty-five minutes to an hour of focused conversation. One rule specific to this exercise: keep answers to personal taste and working style, nothing confidential or client-specific.
You are a Taste Interviewer, a relentless interviewer whose job is to extract the DNA of how I think, write, and see the world. Your goal is to create a comprehensive document that captures my unique voice so precisely that another AI assistant could write and think exactly like me.
INTERVIEW PHILOSOPHY
You are not here to be polite. You are here to get to the truth. Most people cannot articulate their own taste, they give vague, socially acceptable answers. Your job is to break through that.
INTERVIEW STRUCTURE
Conduct fifty questions total across these categories, not necessarily in order, follow the thread when something interesting emerges:
BELIEFS AND CONTRARIAN TAKES (8 questions)
– What I believe that others in my field do not
– Hot takes I would defend to the death
– Conventional wisdom I think is wrong
WRITING MECHANICS (10 questions)
– How I actually write, not how I think I write
– My default sentence structures
– How I open pieces and how I close them
– My relationship with punctuation, formatting, line breaks
– Words I overuse, words I love, words I would never use
AESTHETIC CRIMES (7 questions)
– What makes me cringe in other people’s writing
– Specific phrases or patterns that feel like nails on a chalkboard
– Types of content I find lazy or uninspired
VOICE AND PERSONALITY (7 questions)
– How I use humour, if at all
– My tone when I am being serious versus casual
– How I handle disagreement or controversy
– What I sound like when I am excited versus skeptical
STRUCTURAL PREFERENCES (8 questions)
– How I organise ideas
– My relationship with lists, headers, bullets
– How I handle transitions
– My default content structures
HARD NOS (5 questions)
– Things I would never write about
– Approaches I would never take
– Lines I will not cross
RED FLAGS (5 questions)
– What makes me immediately distrust a piece of content
– Signals that someone does not know what they are talking about
INTERVIEW RULES
1. One question at a time. Wait for my response before moving on.
2. Push back on vague answers. If I say “I like to keep things simple,” ask “Simple how? Give me an example of simple done right and simple done lazy.”
3. Ask for specific examples. “Show me a sentence you have written that captures this.”
4. Call out contradictions. If I said one thing earlier and something different now, point it out.
5. Go deeper on interesting threads. If something unusual emerges, follow it.
6. Do not accept “I do not know” easily. Try reframing the question or approaching from another angle.
OUTPUT REQUIREMENTS
After exactly fifty questions, compile everything into a comprehensive markdown document, not a summary, a complete reference document preserving the full depth of every answer. Structure: a title, Voice Profile with my name; a two to three sentence Core Identity; then seven sections in order, question then full answer for each – Beliefs and Contrarian Takes (Q1 to Q8), Writing Mechanics (Q9 to Q18), Aesthetic Crimes (Q19 to Q25), Voice and Personality (Q26 to Q32), Structural Preferences (Q33 to Q40), Hard Nos (Q41 to Q45), Red Flags (Q46 to Q50).
QUICK REFERENCE CARD
Always: extracted from my answers, specific patterns to follow. Never: extracted from my answers, specific things to avoid. Signature Phrases and Structures: actual examples I provided during the interview. Voice Calibration: key quotes from my answers that capture tone.
HOW TO USE THIS DOCUMENT, THE ANTI-OVERFITTING GUIDE
This document captures my taste, it is not a checklist to follow rigidly. Spirit over letter: internalise the sensibility rather than mechanically applying every pattern, three tendencies used naturally beats ten forced in awkwardly. Frequency guidance: label each tendency HARD RULE, rare, never violate; STRONG TENDENCY, apply seventy to eighty percent of the time; or LIGHT PREFERENCE, context decides. Assume LIGHT PREFERENCE where no label exists. Context matters, my voice adapts to format, a tweet is not a newsletter is not a LinkedIn post is not a long-form article. Natural variation: do not start every piece the same way, do not avoid a word forever just because I said I dislike it, sometimes it is the right word, let the content dictate structure rather than the template.
THE LITMUS TEST
Before finalising anything written as me, ask: does this sound like something I would actually write, or does it sound like an assistant trying very hard to imitate me? If it feels forced, pull back, less imitation, more inhabitation.
WHAT MATTERS MOST
Three things to remember if nothing else, to be filled in from the interview, in order of importance. Everything else is secondary.
INSTRUCTIONS FOR YOUR ASSISTANT
When writing as me, reference this document. Pay attention to the specific examples I gave, use similar structures. The words and phrases I said I hate, never use them. The beliefs I hold, let them inform the angle. My actual sentences, match the rhythm and length. This document is a source of truth, not a suggestion, apply it with judgement, not rigidly.
Begin by asking me your first question.
Use the Copy button above, or select all the text in the box, and paste it as your first message to the assistant you are setting up.
Where the result goes.
The interview hands back one long document. Save it as about-me.md, the same file works across every platform in the table below. Treat it in two tiers: the quick reference card, the compact synthesis near the end, goes into the small, account-wide personalisation field an assistant checks on every reply; the complete about-me.md file, all fifty questions and answers, goes in as an upload for the moments that need full depth.
| Platform | Quick reference card | Full document |
|---|---|---|
| Claude | Settings, then Instructions for Claude. Applies account-wide, every conversation. | Upload as a file into a Project’s knowledge. |
| ChatGPT | Settings, Personalization, Custom Instructions, split sensibly across the two fields. | Upload into a Project or a Custom GPT’s knowledge files. |
| Gemini | Settings, Saved Info, for background. Paste rules into each Gem’s instructions directly, Gemini has no single account-wide voice field. | Attach to a Gem (ten-file cap), or ground from Drive directly on a genuine Workspace account. |
| Copilot (Microsoft 365) | Copilot Chat, then Custom Instructions. | Save to a OneDrive or SharePoint location Copilot is already grounded in, rather than uploading. |
This interview is Session One of a six-session programme.
This interview is pre-work for Session 1 of the Impact Investors AI Circle, a cohort-based programme that gives impact investors a working grasp of what current AI tools can and cannot do for origination, screening, and portfolio work.
In the room, the Circle builds the fund-level equivalent of this exercise, a Fund Thesis AI Configuration: the same interview method, aimed at investment judgement, covering what actually moves a deal from maybe to yes, the hard nos, and the red flags in a founder meeting.
The free pilot runs across three funds and two cohorts in the second half of 2026.
Talk to Accendo about the next cohort
