How it speaks when someone is still sharp

Why there are two models

Every standard on our dementia page was written for dementia. Never correct. Validate before redirecting. Meet the person where they are. Treat each repetition as a fresh moment. Those are correct, they are well founded, and they are the reason families trust the product.

Applied to someone whose memory has only begun to change, every one of them is wrong.

A person in the early stage has insight. They know they are forgetting. They notice being agreed with when they are wrong, they notice a conversation that never goes anywhere, and they notice being spoken to slowly. What reads as gentleness to a family reads to them as being written off, and the person who feels written off stops picking up.

So there are two models on one product. A family chooses which one fits, and the choice is reversible.

The dementia model is called Meet them where they are. This one is called Speak plainly.

What this model is built on

The shape of these conversations comes from a randomized controlled trial called I-CONECT, which studied frequent structured conversation with socially isolated adults aged 75 and older. People living with dementia were not included in it.

What we took from it was the design, not a result. Thirty minute conversations rather than check ins. A large library of prepared subjects rather than whatever comes to mind. Three subjects offered at the start and the person chooses. A record of what has already been talked about, so the same ground is not walked twice. And a count of who did the talking, because a conversation where the visitor does most of it is not the thing worth having.

The trial used trained people. We are not people. That difference is the open question, and we do not paper over it.

What we do not claim

KindredMind does not claim that its conversations improve memory, protect cognition, slow anything down, or reduce anyone's risk of anything. We have not run that trial and we will not imply we have. What we can tell you is exactly what we do and why we chose to do it that way, which is the rest of this page.

If someone in your family is worried about their memory, the person to talk to is their doctor. We are a conversation, not a treatment.

We have written up the trial in full, including what it found, what it did not test, and where its limitations sit. Read our review of the I-CONECT study

The twenty two standards

These are not suggestions. Each one is enforced in the system that runs every conversation, on the phone and on the tablet alike.

Arriving

How it talks

Going deeper

Never a test

When something goes wrong in the conversation

Honesty

Their call, not ours

The subjects it brings

The conversation is built on a library of 150 subjects with 450 different ways into them, covering history people lived through, travel and places, pastimes and sport, music, books and film, ideas worth arguing about, how life together has changed, work and craft, food and home, the natural world, and this country in particular.

How it works in practice:

• One subject is chosen before the conversation starts, never during.
• Three ways into it are offered. They pick one, or they say something else entirely.
• A subject that has come up recently will not come up again. There is a record.
• The same ground is not revisited for two months, and then only from a different angle.
• If they pass on something twice, it stops being offered.
• If they say they would rather not talk about something, it never comes up again.

What is deliberately not in the library: politics, disasters, illness, death, and anything that invites someone to assess their own memory. Those subjects can be reached if the person raises them. They are never what it arrives with.

And none of it outranks their own life. If there is something real to pick up from the last conversation, the prepared subject is set aside and not mentioned.

If they tell it they are worried about their memory

This comes up, and it is scripted, because improvising it would be the worst possible place to improvise.

• It takes them seriously. It does not say everybody forgets things.
• It does not diagnose, predict, or reassure them with something it cannot know.
• It does not say the words dementia or Alzheimer's. Not once, under any circumstance.
• It points them toward a person rather than a service: their doctor, or the family member they are closest to.
• If they want their family told that they wanted to talk, it passes that on as a request for a conversation, not as an alert.

What they can see, and what they can stop

On this model the person is the customer, not the subject.

They agree to it themselves. Their family cannot arrange these calls on their behalf. The person hears, in plain language, that the calls come from a service their family set up, that it is AI and not a person, what it knows about them, who sees the summaries, and that they can stop it whenever they like. Then they say yes in their own words. The system will not start without it.

It never knows something about them they cannot see. Anything the conversation draws on can be read back to them in ordinary language whenever they ask. A family member can keep private notes of their own, and those never reach the conversation and are never spoken.

They can add to it themselves. Most of what makes a conversation good comes from the person, not from a form their daughter filled in. What they add goes in as they wrote it.

They can turn it off. Not by asking their family to. Themselves, at any time, and the family is told afterward.

Safety and escalation

Unchanged from the dementia model, because distress does not care which model is running.

• It listens for signs of distress and alerts the family straight away.
• It stays calm and stays on the line.
• If the system fails mid conversation, the call routes to the family's phone.
• It never gives medical advice and never interprets a symptom.
• It never says anyone is on their way or about to arrive unless a real visit is on the calendar.
• It never speaks for a real family member on anything that matters: no promises, no medical or money statements, no family decisions.

After the conversation

The family receives the same three things they receive today: the full conversation, stored encrypted; a short summary of what was talked about; and any distress alerts with the time they happened.

What the family does not receive is a score. No engagement rating, no participation percentage, no report on how well the conversation went. We measure the quality of our own side of it, closely, and that is our problem to fix rather than a performance review of someone's parent.

Conversations are deleted after ninety days by default, and the family can change that.

Languages

Every standard on this page applies identically in all fifteen supported languages. The prepared subjects are adapted rather than translated, because a conversation about the country someone grew up in cannot be a word for word translation of a conversation about this one.

If memory changes further

Families do not stay in one place, and the product should not pretend otherwise.

If the time comes when the plain spoken model no longer fits, the family can move to the dementia model. It is a deliberate step, not a settings change: the family has to say that their loved one can no longer decide this for themselves, and it is recorded. From then on the dementia standards govern, and a voice they know becomes available for the first time, with the permission of the person it belongs to.

It can be moved back. First judgments are sometimes wrong.

Nothing about that move is ever automatic. We do not read conversations for signs of decline and we do not suggest to a family that it might be time. That would be a clinical judgment about a named person, made by a company, from a phone call. We will not do it.

For clinicians and care teams

The evidence

We are not going to restate the trial here, because we have already written it up properly and doing it twice is how two versions of the same number end up on one website.

Our full review of the I-CONECT study covers the design, the primary outcome and its actual sample size, the secondary outcomes and why the multiple comparison threshold matters when reading them, the exploratory imaging and why fifteen participants is not a finding, and the pandemic disruption that shapes how all of it should be read. It also sets out, in a list, the nine things the trial did not establish.

Read that first. This section is about what we built from it.

What we replicated, and what we did not

That last row is the honest one. The trial studied trained people. We are not people. Whether a conversation held to the same standards produces anything like the same result is unproven, by us and by anyone else.

The claim we make, and the one we do not

We make no claim that KindredMind improves memory or cognition, protects against decline, or reduces anyone's risk of anything. We have not run that trial.

What we claim is narrow and checkable: our conversations follow a documented set of standards modeled on a published trial's design, every standard is enforced in the system rather than suggested to it, and we measure our own adherence to them after every conversation.

Partners evaluating this should ask us for the adherence data. We will show it.

Related reading