Skip to main content

    Research · 2026

    The documents nobody has studied take an hour each.

    The Documentation Minute · Aisel Research, 2026

    We timed 202 documentation tasks across eight psychiatric clinics in Denmark and the UK, spanning 16 document types. The time is concentrated in long-form reports - Mental Health Act, medico-legal and court documents at 46 to 57 minutes each. AI assistance has been evaluated on the consultation note only: in the published literature, and in our own data. This paper locates the target. The next one tests it.

    Lotte Kjær Svalberg · Cand.psych.; stud.med.; Clinical Product Manager, Aisel··12 min read

    Open access · No email required to read

    Declaration

    This analysis was conducted by Aisel, which develops one of the tools evaluated. The dataset, methods and limitations are published in full so readers can weigh the findings accordingly.

    The centrepiece

    Where the time goes.

    Sixteen document types were timed across eight clinics. Each row below shows the mean unassisted time and the number of entries it was drawn from. Rows drawn from fewer than five entries are shown as provisional.

    Mental Health Act report
    57.1 minn=7
    Medico-legal report
    49.5 minn=5
    Court reportprovisional
    46.2 minn=2
    Risk assessmentprovisional
    45.0 minn=1
    Benefits report
    44.5 minn=5
    Discharge summary
    24.9 minn=27
    Certificateprovisional
    17.5 minn=2
    Referral letter
    15.5 minn=25
    Clinical letter to GP
    14.6 minn=17
    Clinical notes
    14.5 minn=43
    Clinic letterprovisional
    13.8 minn=2
    Status certificateprovisional
    12.5 minn=2
    Advice and guidanceprovisional
    10.0 minn=1
    Medication passportprovisional
    7.0 minn=1
    Medical certificate (sick leave)provisional
    5.2 minn=4
    FIT note
    4.9 minn=8
    Unassisted mean timeConsultation note (the task most studied)Provisional (n < 5)
    Fig. 1. Mean unassisted time per document type. Sorted descending. Sample sizes vary; interpret provisional bars accordingly. Aisel documentation timing dataset, Denmark and United Kingdom, 2025-26.
    Mean unassisted time per document type, with sample size.
    TaskMean minutesSample sizeConfidence
    Mental Health Act report57.17indicative
    Medico-legal report49.55indicative
    Court report46.22provisional
    Risk assessment45.01provisional
    Benefits report44.55indicative
    Discharge summary24.927robust
    Certificate17.52provisional
    Referral letter15.525robust
    Clinical letter to GP14.617robust
    Clinical notes14.543robust
    Clinic letter13.82provisional
    Status certificate12.52provisional
    Advice and guidance10.01provisional
    Medication passport7.01provisional
    Medical certificate (sick leave)5.24provisional
    FIT note4.98indicative
    Of 23 studies in the most recent systematic review, 20 evaluated only the post-consultation note. The report types at the top of this chart appeared in none. They do not appear in our own AI-assisted data either. That is what this programme of work exists to close.

    The gap

    One task studied. Fifteen left standing.

    What has been evaluated

    • Clinical notes (consultation)14.5 min

    What has not

    • Mental Health Act report57.1 min
    • Medico-legal report49.5 min
    • Court report46.2 min
    • Benefits report44.5 min
    • Discharge summary24.9 min

    These documents follow statutory or pro-forma structures that language models handle well. They also demand synthesis across long clinical histories and carry legal consequences if summarised inaccurately.

    Which factor dominates - the structure that AI is good at, or the synthesis and consequence that make errors costly - is unknown. It has not been studied.


    Clinical notes - the one task with data on both sides

    14.5 minutes to 5.5 minutes.

    Only one document type in our dataset has enough matched entries to compare unassisted and assisted times. It is the same task almost all the published literature has looked at.

    Unassisted (n=43)14.5 min
    Aisel-assisted (n=37)5.5 min

    Means

    14.5 → 5.5 min

    62% reduction

    Medians

    10.0 → 5.0 min

    50% reduction · more representative given the skewed distribution

    How to read this figure

    • The assisted entries come from a single clinic and were stopwatch-timed.
    • 39 of the 43 unassisted entries are retrospective estimates from other clinics.
    • Retrospective estimates tend to run high.
    • This is an observational cross-site comparison, not a controlled trial, and does not establish that AI assistance caused the difference.

    An independent NHS child and adolescent psychiatry pilot reported AI-assisted documentation at 45% of manual time (median 27 → 10 min). Stanton et al., BJPsych Open 2025 - a conference abstract reviewed by the RCPsych Academic Faculty, not peer-reviewed.


    Why this matters

    Four consequences, honestly tagged.

    Clinical scalability

    Documentation time is measured directly here, and it is the resource that competes with patient contact. How much converts into additional appointments is not measured.

    Supported by this data

    Staff wellbeing

    Documentation burden is one of the most consistent predictors of clinician burnout in the published literature. This dataset measures time, not wellbeing.

    Hypothesis from published mechanisms

    Service economics

    Medico-legal and report work is among the best-paid clinic work; preparation time is the bottleneck.

    Hypothesis from published mechanisms

    Patient experience

    Faster documentation is associated with more attention returned to the room; direct evidence is limited in psychiatry.

    Hypothesis from published mechanisms


    Limitations

    A real one, not a disclaimer.

    • Observational cross-site design. No randomisation and no controlled counterfactual. Differences between clinics may explain some of the observed effect.
    • Measurement method is confounded with condition. Assisted entries are stopwatch-timed; most unassisted entries are retrospective estimates. The two are not directly comparable.
    • Clustering. Assisted entries come from a small number of clinicians at one site; the effective sample is smaller than the raw n implies.
    • No entry dates. We cannot control for time-of-day, day-of-week or learning curves.
    • Convenience sample. Clinics were recruited through Aisel's clinical network. They are not necessarily representative of psychiatric services overall.
    • Author conflict of interest. Aisel develops one of the tools evaluated. Findings should be read with that in view.

    What comes next

    What this paper is for.

    Mapping where the time goes was the prerequisite. The documents at the top of that chart - medico-legal, Mental Health Act and benefits reports - have not been evaluated with AI assistance by anyone, including us. That evaluation is the study we are running now.

    GDPR-compliant. We send the study and nothing else unless you opt in.

    Full paper

    The Documentation Minute: A Multi-Site Analysis of Time Spent on Clinical Documentation in Psychiatry, and the Effect of AI-Assisted Notetaking.

    Download the PDF - charts, tables and references included.

    Cite this

    Svalberg, L. K. (2026). The Documentation Minute: A Multi-Site Analysis of Time Spent on Clinical Documentation in Psychiatry, and the Effect of AI-Assisted Notetaking. Aisel Health. https://www.aisel.co/research/documentation-burden-psychiatry

    Aisel is built for the documents at the top of that chart.