Guide
AI Twins vs Chatbots: What's the Difference?
They look similar. They work differently.
An AI twin and a chatbot share an interface. You type a question and get a reply. Under the surface they run on the same kind of language model too. Imora's twins use Anthropic's Claude.
The difference is what the model is given to reason from, and what it has to show you afterwards.
Chatbots: the most plausible answer
Ask a general chatbot whether to give a client a discount and you get a sensible summary of the considerations: margin, relationship, precedent, alternatives to a price cut. It is the answer a well-read stranger would give.
That is useful for research, drafting and quick lookups. It is no use when the question is "what would Sarah do?", because the chatbot has never met Sarah.
Chatbots with a persona: closer, but not there
The next step up is a custom GPT or a Claude Project with a persona prompt and your documents. It is an afternoon's work, and for many jobs it is enough. It picks up your tone and can quote what you have written.
It stops in three places. It cannot tell you why it answered as it did. It has your conclusions but not the conditions behind them, so it generalises badly. And a correction made in one conversation is usually gone by the next.
Twins: one person's recorded decisions
An Imora twin answers from a reasoning graph. You answer real scenarios in your own words. Those answers, together with your corrections and documents, are stored as evidence. Imora distils the evidence into principles with their conditions, such as "discount for commitment, never for complaint; one exception for clients in their first year".
When a question comes in, the twin reasons from the principles and evidence that fit that kind of decision. The reply opens with the twin's read, and "why this answer" shows the principles and evidence used and how confident it could be.
Four practical differences
It shows its work. A reader can see which of your principles an answer rests on, and the words of yours they came from.
Corrections stick. Mark a reply "Not what I'd say", give your version, and that correction outranks everything else from then on.
Only the owner teaches it. Nothing a visitor says changes how a twin thinks. A persistent client cannot argue it into a new pricing policy.
It admits gaps. When a shared twin has no recorded view, it says so. It offers an approach, clearly framed as an approach and not the owner's position, and the question goes to the owner's inbox.
The consistency test
Ask a chatbot the same judgement question on two days and the framing, and sometimes the recommendation, will move. Nothing fixed sits behind it.
Ask a twin twice and the wording will vary, but the position should not, because both answers come from the same principles. If the position does move, the owner can see why and correct it.
When to use what
Use a chatbot for general information, drafting and research.
Use a persona chatbot when tone and document lookup are all you need.
Use a twin when other people will act on the answer as your judgement: a team checking a client reply against how you would handle it, or clients who want your view between sessions.
We compare the options in more detail, including where other tools are ahead of us, on the comparison page. For how the graph is built, read introducing the reasoning graph or the product page.