Guide
How to Build an AI Version of Yourself
Documents are not enough
The usual first attempt is to upload your documents to a chatbot and tell it to answer as you. What comes back knows your content and sounds like everyone else.
Your documents record what you concluded. They rarely record how you got there, or what would have changed your mind. A twin your team or clients can rely on needs decisions.
Step 1: answer real scenarios
Start with concrete cases, not a questionnaire about your values. People describe their judgement badly in the abstract and well when a case is in front of them.
Good scenarios have a trade-off in them:
- A client asks for a 20% discount to renew. They are profitable but demanding.
- A project is drifting past the agreed scope, one small favour at a time.
- A strong candidate wants more than the band allows.
- A client is rude to a junior on your team for the second time.
Say what you would do and why, in your own words. Include the conditions. "I would hold the price, unless they commit to twelve months" is worth ten statements about valuing fairness.
On Imora this is the quick start: five scenarios, about 10 minutes.
Step 2: check what it learned
Before you chat, look at what was taken from your answers. Imora distils them into principles, each with the conditions it applies under, and shows your own words beside them. Together they form your reasoning graph.
Read them critically. If a principle says "never discounts" and you would discount for a long commitment, a condition is missing.
Step 3: ask it something new
Chat with your twin, and ask about situations you did not cover. Repeating you is easy. This is the real test.
Each reply opens with the twin's read of the situation. Open "why this answer" to see the principles and evidence it used and how confident it could be. A good answer on thin evidence should say the evidence is thin.
Step 4: correct it
It will get things wrong. When it does, mark the reply "Not what I'd say" and write what you would have said. That correction outranks everything else, and it sticks. When a reply is right, "Sounds like me" confirms it.
This matters more than a perfect first pass.
Step 5: add depth where it is missing
Two optional things add depth. The guided interview has six sections and takes 30 to 40 minutes. It covers your context, expertise, communication style and career stories, and we describe it in six layers of context. Your own documents add knowledge, and answers that use them cite them.
Do these when you notice a gap: documents if the twin lacks background, the interview if it sounds unlike you.
Step 6: share it carefully
Send one link to your team or clients. Two rules protect you once it is shared.
Only you teach your twin. Nothing a visitor says changes how it thinks.
When a shared twin has no recorded view on something, it says so, offers an approach clearly framed as an approach and not your position, and puts the question in your inbox. Those questions are the best guide to what to teach it next.
Step 7: measure it
Do not rely on whether it "feels like you". On Imora, Mirror Score shows how much your twin has to go on. It is not an accuracy score. For accuracy there is the fidelity check: we hide some of your real answers from your twin, then check which version gets closest to what you actually said.
Short calibration sessions keep it current as your thinking changes. What makes a good AI twin has tests you can apply to any tool, and the product page shows the whole process.