How TwentyThird treats evidence.
We cite other people’s studies and say what they do not show. We publish nothing of our own until it can bear the weight.
TwentyThird publishes no efficacy data of its own today. The studies cited on its science page concern psychodynamic therapy delivered by people and AI tools built by others, not TwentyThird. TwentyThird would publish aggregate results from its own members only when the sample supports them, the method and limits are stated beside every figure, and members have given consent that covers research.
· Published under the responsibility of the founder, John Muhammadi, M.D..
Other people’s evidence, labelled as theirs.
The science page names each study, says who it was about, and says that none of them studied TwentyThird. The methodology page explains how the ideas are used and what a language model cannot do with them.
- Each study is cited with its authors, publisher and year
- Each one says whom it studied: people in therapy, or other AI tools
- A figure on this site has a source, or it says it is an example
- Example insights in the product are demos, never quoted as results
Read the science and the methodology.
Nothing published yet, and the conditions for when.
TwentyThird has not been tested in a trial. If we run an evaluation, we will publish it with its method. Until then, no result from TwentyThird appears on this site.
Member data is used for the member
Today it produces the member’s own readings, conversations and reports, and keeps them safe. It is not sold, not used for advertising and not used to train models, as the privacy policy states.
Research needs its own consent
The consent members give today covers delivering the service. Using anyone’s data for research would need a separate consent that can be withdrawn, and no data is used for research without it.
We publish when n supports it
Aggregate results only, never an individual, and never a group so small that a person could be picked out. The minimum size is set before the analysis, not after. Every figure carries its n, its method and its limits.
No number before it exists
We do not estimate, round up or borrow a figure to fill a gap.
Details of what is processed, by whom and for how long are in the privacy policy.
Six questions we want answered.
These are questions, not findings. None has an answer yet.
Do members recognise the loops Oneli names?
A reading is useful only if the person can check it against their own life. We want to know how often a named loop is recognised, rejected or corrected.
What does the Circle add?
Does bringing in the view of people who know a member change what the member notices, and when does it only add noise?
What do psychologists want from an AI between sessions?
What would a psychologist want to see, and what would they rather never see? We would ask psychologists before building more.
Where does the risk check fail?
It is an AI model and can miss a serious message or flag a harmless one. How often, and in which kinds of message, is a question for review by clinicians.
Do readings survive translation?
TwentyThird is published in 27 languages. Does a reading keep its meaning when it moves between them, including in languages where the clinical terms differ?
Does it change what happens in the appointment?
When a member brings a reading to a psychologist, does the session go somewhere it would not have gone? This needs the psychologist’s view as well as the member’s.
Psychologists and academics, write to us.
If you research psychodynamic work, AI in mental health, or measurement in either, we would like to hear your critique as much as your offer.
Psychologists
Licensed psychologists can apply to design and supervise an agent, and can tell us what they need to see. Read for psychologists.
Academics
A collaboration would be a conversation first. We cannot promise access to member data. Any study would need member consent and ethical review where the law requires it. Write to info@day-23.com with the subject “Research collaboration”.
Sources checked, pages dated, errors corrected.
The rules every explainer, comparison and fact sheet on this site is held to.
Who is responsible
Content is published under the responsibility of the founder, John Muhammadi, M.D., a physician. There is no outside clinical reviewer today, and we do not write “reviewed by a psychologist”. When there is one, the page will name them.
Sources are opened, then cited
Every external claim names its author or publisher and year. A source that could not be opened is not cited. Statements about other products come from that vendor’s own public pages or reputable reporting, carry a date, and are written neutrally.
Pages are dated
Each page shows when it was last updated, and the date changes only when the content does. The sitemap reports the same date.
No superlatives, no invented examples
We do not use “first”, “only”, “best”, “leading” or “disruptive”. Examples are described as types of situation, never as a real person’s story or quote. Where TwentyThird is weaker, a page says so.
Found an error? Tell us.
Write to info@day-23.com with the subject “Correction”. Give the page address, the sentence you think is wrong and, if you can, a source. We reply, and we correct the page when you are right.
A correction changes the text and the page’s date. If an error could have misled a reader about safety, data or price, the page says what was changed and when. We do not remove an error silently.
Read the evidence, then decide.
If you are in danger, contact local emergency services now — 112 anywhere in the EU. TwentyThird is not an emergency service. How safety works