Merakoi white paper 2026 · Where AI Patient Insight Stops
Who needs patients when AI can sound like one?
Ask any model what living with your condition is like. It will answer fluently, from whatever somebody once wrote down. Here is what it can never know.
Kevin Michels-Kim, co-founder, Merakoi
Sahara Fleetwood-Beresford, patient expert, IBDSeptember 2026
The uncomfortable numbers
Empathy is not where humans win any more.
79%
of the time, clinicians preferred a chatbot’s answers to real patient questions over a physician’s.
Ayers et al., JAMA Internal Medicine, 2023
6%
of the 341 pivotal trials behind recent FDA approvals enrolled a representative population. The record AI learned from was filtered before it was written.
Zaaijer and Groen, Communications Medicine, 2025
0
people you can call when a synthetic patient insight turns out to be wrong.
Gap 5, below
This is not a hypothetical
Somebody is already building your patients’ replacements
A patient engagement lead at a large pharma company described building patient digital twins at scale. Unprompted, she added that she did not want to leave real people behind.
The big unanswered question, who is actually accountable, remains open. An AI, no matter how intelligent, is not accountable.
Kevin Michels-Kim, co-founder, Merakoi
Where the synthesis stops
Five things your patients know that no model ever will
A bigger model reads the same record faster. It does not read what was never written.
-
1
Who wrote down the thing nobody wrote down?
Most of what patients know was never written down anywhere a model could read it.
-
2
Would you act on a patient insight that stopped updating last year?
Models are archives, and disease is current.
-
3
Whose voices trained the model? And whose never made it in?
The written record over-represents the connected, the confident, the English speaking and the digitally comfortable.
-
4
Can a chatbot lose sleep over your decision?
Knowing that a symptom is distressing is not the same as knowing what it makes someone do.
-
5
When the synthetic patient is wrong, who do you call?
Nobody can be accountable for a claim no person actually made.
Humans are the source. AI is the multiplier.
A multiplier with no source multiplies nothing.
This is not an argument against AI. It is about what you feed it.
-
From AI or patients
To AI times patients
-
From ad hoc recruitment
To standing relationships
-
From insight reports
To co-authored outputs
What the paper adds
This page is the argument. The paper is the evidence.
- The patient accounts, in a patient expert’s own words
- What to automate, and what to never automate
- The references and the method
Free PDF. One short form.
Who wrote this
Kevin Michels-Kim
Co-founder, Merakoi
Argues weekly with pharma teams about where AI belongs in patient engagement.
Sahara Fleetwood-Beresford
Patient expert, IBD
Lives with IBD and has run patient communities for over a decade.
AI helped write this paper, and we say where. No patient quote was generated by AI, and every sign-off is human. Our commitments: the Merakoi AI Editorial Charter.
