What an AI twin is

The offer

A verified twin of you, live on the internet, answering in text chat, in your own voice, and on a live talking avatar. One build gives you all three. It runs on your material, in your manner, at the hours you are not available, and it carries your name because your identity is checked before it goes live.

Building it is free. There is no subscription, so a quiet month costs you nothing. You set your own per-minute price above a platform base and keep 80% of your markup, with commission on the markup only. You choose what a first-time visitor gets free. You approve what it may say before it says anything, and you can switch it off at any moment.

Going live needs no plan at all. Consent and a checked identity are the only gate. Paid seats add the things you want once it is working: embedding your twin on your own site, keeping it out of the public directory, usage analytics and response controls, and an allowance of sponsored minutes to give your visitors.

The short definition

An AI twin represents one specific, named, living person. It is built from material that person supplied, published with their recorded consent, and it stays under their control.

Four properties separate a twin from software with a face on it. The identity is real and checked. The source material came from the person. The person consented on the record. The person can switch it off. Everything else is implementation. Whether it speaks, whether it appears on video, whether it runs in a browser or inside someone else's website, those are surface choices that can change while the thing remains a twin.

How it works on AIMayDo

Five lines cover the arrangement.

• Building a twin is free.
• There is no subscription, and nothing is billed in a quiet month.
• The expert sets a per-minute price above a platform base rate.
• The expert keeps 80% of that markup, and commission applies to the markup only, never to the base.
• Going live needs no plan. Paid seats (from EUR 149/mo) add embedding, analytics, directory control and a sponsored-minute allowance.
• Visitors are billed only while a conversation is actually running, and the expert can revoke the twin at any time.

If you want the arithmetic instead of the summary, we set out how per-minute pricing and the revenue split work in full on its own page.

How a twin differs from a chatbot

A chatbot answers from a general model plus whatever documents it was given. It makes no identity claim. When it gets something wrong, nobody is embarrassed, because it was not speaking for anyone in particular.

A twin makes an identity claim, and that claim is the product. When a twin of a named psychiatrist answers a question about sleep, the answer carries that person's reputation. That changes what the software has to do, in ways that look small on a feature list.

• The source material has to be traceable to the person, not scraped from the open web on the assumption that it is theirs.
• The identity has to be checked before publication. On AIMayDo that check runs through Stripe before a twin goes live.
• There has to be an off switch the person operates themselves, without asking support and waiting.
• The distance between the twin's answers and the person's own answers has to be measured and shown, because a visitor paying for proximity to a named individual deserves to know how close they are getting.

Text-only chatbot builders need none of that, which is why a good deal of what is sold as a clone builder is a chatbot builder with a photograph in the corner.

How a twin differs from a video avatar

Tools that generate a talking video avatar solve rendering. They produce a face that moves and lips that track generated speech, and the good ones solve it well. That is the presentation layer, and it is the layer a buyer can see in a demo, which is why it gets sold hardest.

Put a general model behind a rendered face and you have a puppet wearing someone's cheekbones. Take the face away from a twin and a twin remains. The substance is the half that carries the value.

A complete twin is layered. The person's material and voice sit underneath, the conversation layer sits on top, and rendering comes last. AIMayDo produces all three layers from one build, so a visitor can type or talk, the live avatar is available in the same session, and it is the same twin answering behind each of them.

Adjacent terms that get confused with this one

Digital twin. An older industrial term for a simulation of a physical asset such as a turbine or a factory line. Same words, unrelated field.

AI persona or character. A fictional personality with no real person behind it. No identity claim, so no verification problem and no consent problem.

AI agent. Software that takes actions on your behalf, booking things and calling other systems. A twin talks. An agent acts. Blurring the two is how conversational software ends up holding authority it should not have.

Voice clone. One component. It reproduces how someone sounds without reproducing what they know.

Deepfake. The same rendering technology used without consent. Consent and verification are the whole of the difference, which is why a twin platform that treats them as paperwork is closer to the wrong category than it thinks.

Consent, verification and the off switch

Verification means somebody checked that the person in the twin agreed to be in it, and that they are who the page says they are. On AIMayDo the expert uploads their own material, records a consent statement, and passes an identity check through Stripe. Only then does the twin go live.

Order matters more than mechanism here. A platform that publishes first and checks identity later has, for the whole window in between, made an unverified public claim about a real person. A thorough check afterwards does not recover that window.

Consent should also be withdrawable without a negotiation. On AIMayDo the expert revokes their twin whenever they choose, and leaving deletes the face and the voice at the underlying vendors as well as in our own database. Consent you cannot withdraw is a licence under another name.

What a fidelity score measures

Fidelity is how closely the twin's answers match what the person would say. It is not accuracy and it is not answer quality. A twin can produce an excellent answer the real expert would never give, and that counts as a fidelity failure.

We built our own engine to measure it. It profiles the person across five dimensions, voice and tone, reasoning style, lexicon, positions and guardrails, each with its own confidence, and drafts the decision rules the expert approves. Scoring holds back questions the expert answered themselves, puts those questions to the twin cold, and grades the answers against a rubric we wrote, averaged over several runs and re-measured as the material changes. The result is published on the twin's own page.

We do not market any twin as indistinguishable from the person, and we could not defend that claim if we made it. Publishing a number sets a ceiling as well as making a claim, and the ceiling is the reason to publish it. If you are assessing twins anywhere, ask whether a fidelity number is published and how it was produced.

What the expert controls

The decision rules pulled out of the material go to the expert for approval before anything is published. A twin repeating a position its expert abandoned two years ago is doing so under that expert's name, so nothing goes live on their behalf that they have not read.

The free taster is set per twin by the expert, as two caps that work independently of each other: an amount of free time, and a number of free questions. Either can be set to zero, and whichever runs out first ends the free tier. An expert who wants visitors to have a long look sets both caps generously. An expert who wants the meter running from the first second sets them to nothing. There is no blanket platform allowance underneath this; the twin's owner decides.

The twin can also be embedded on the expert's own site through an allowlisted iframe. That is the version of this we would point at first. Your twin sits on your own domain, answering people who already know you and already came looking, in text or voice or on the live avatar, at the price you set.

The limits, stated plainly

A twin is not the person. It has no memory of your history with them and no duty of care towards you. It cannot notice that you have gone quiet because something has upset you.

A twin can be wrong. It runs a language model over supplied material, and language models produce confident errors, particularly at the edges of what a person has written down. A high fidelity score means the errors sound like the person, which is a different thing from being right, and harder to catch, because an error in the right voice is the one least likely to be questioned.

A twin is not regulated advice. Talking to a twin of a psychiatrist is not treatment, and no score changes that. If the question underneath is one you would take to a professional, take it to a professional.

Where AIMayDo is today

AIMayDo is live and running. Dr Evangelos Katsioulis, a psychiatrist and a co-founder, has his twin on the platform now, with its measured fidelity score published on the page. You can talk to it before you decide anything.

We are not going to publish what a twin earns, because we have no evidence that would make such a figure true, and anything we printed would be a guess wearing the clothes of data. Apply the same suspicion to everyone in this category, including us.

Reasons people build one

The reasons we hear are not all commercial, and they do not have to be. Curiosity is one of them. Being early is another. Some people want to hear what a version of themselves, trained on their own words, says when a stranger asks it something difficult. Some like the idea of their material answering a question at two in the morning, in their own manner, while they are asleep. Those are legitimate reasons to want one, and we would rather say so than pretend the only respectable motive is revenue.

Building your own

Building one is free, and it comes apart again on your own say-so. You upload your own material, record consent, clear the identity check through Stripe, approve the decision rules, then set your taster and your price. One build gives you text, voice and the live avatar, and you can put the result on your own site.

If you would like to watch one behave first, the live twin of Dr Evangelos Katsioulis is on the site with its fidelity score on the page. When you are ready to make your own, start building it on our page for experts. It costs nothing to find out what yours sounds like.