Organising the literature
It helps us structure search strategies and sort large volumes of references. Someone on the team always makes the decision to include or exclude a study.
ES · EN
Book 30 min
Transparency statement
We use general-purpose AI assistants for some support tasks, and specialised evidence synthesis platforms where a project warrants it. What we do not delegate is the methodological judgement, which is precisely what the client is paying for. This page sets out where that line falls, so you do not have to ask.
Yes
On high-volume, low-judgement tasks, where we can check the output against a human standard.
It helps us structure search strategies and sort large volumes of references. Someone on the team always makes the decision to include or exclude a study.
First drafts, and clarity edits on material we have already written. We never draft a finding, a conclusion or a recommendation this way.
We produce deliverables in both languages and harmonise vocabulary across sources; a methodologist does the final technical review.
Summaries that help us get our bearings in a body of literature. We then check anything that enters the analysis against the original document.
No
These are judgement calls, and judgement is exactly what you are hiring us for. We do not delegate it.
A human makes the GRADE judgement on risk of bias, inconsistency, indirectness and imprecision, and signs it.
We do not delegate what an effect means for a patient and a health system.
The panel decides the direction and strength of a recommendation, and we document its deliberation.
No text that asserts a finding goes out without a named human author who answers for it.
A distinction
Lumping a general-purpose assistant together with an evidence synthesis platform would be convenient for us and untrue. There are platforms that run real R and JAGS, screen with models trained on biomedical literature, pre-rate the GRADE domains and log every step. They are good at what they do.
They still do not replace what you are hiring, and not for want of technical capability: the calculation is not the judgement. These three are the ones most often confused.
A platform computes I² and Cochran's Q. Neither decides whether to downgrade: that depends on whether the heterogeneity is explained by something that matters for your decision. An I² of 70% across studies in different populations means one thing; across near-identical studies, another.
The engine compares the confidence interval against a threshold. But choosing the threshold — the difference at which the effect starts to matter to a patient — is a judgement about values, not a computation. Change that number and the rating changes, and somebody has to defend why it is that one.
This is where it shows most. A platform compares the PICO fields. It does not know the trials were run in health systems whose access, comparators and patient profiles are not those of your country. That distance has to be judged by someone who knows the context of the decision.
Using one of these platforms strikes us as sensible, not objectionable, and we would say so in writing in the report as we say everything else. What we do not do is hand over their output as though it were a judgement.
The practical difference is who answers for it. When a committee, a peer or an auditor asks why that certainty came out low rather than moderate, the answer has to come from someone who will stand behind it by name. A platform does not appear before anyone.
It is a fair question to put to us and to anyone else you hire. Three concrete things are worth asking for:
Transparency
If we used automated tools on your project, the report says so. You should not have to ask.
Nothing a tool produces enters the analysis until someone on the team has checked it against the original source.
We do not upload patient data or confidential client information to services that do not meet the data processing agreement in force.
Speed and rigour
Automation without controls produces reports that do not survive peer review. Fully manual work produces answers that arrive after the decision has been made.
Our position is simple: we go faster where we can check the output against a human standard, and we cut no corners where judgement is exactly what you are paying for. Every project sets out in writing which side of that line each task fell on.
We will tell you what can move faster in your particular case and what cannot, with realistic timelines.