Assisted critical reading

How can you turn a scientific corpus into a traceable synthesis to support decision-making, without delegating critical judgement to AI?

1960 € per session, excl. VAT

In short

This 7-hour day, excluding breaks, provides a structured approach to critical reading supported by generative AI: framing a question, selecting articles, extracting results, examining biases and comparing conclusions. Participants produce an operational briefing with source references and measure its fidelity to the publications, explicitly stating the uncertainties.

Why this training?

A fluent summary can conceal a weak protocol, an inappropriate comparison or an overstated conclusion. The value of a scientific analysis depends on the verifiable link between question, method, results and interpretation. AI helps to organise information and compare articles, but may invent a reference or confuse association with causation. The training requires participants to return to the source passages and explicitly characterise the limitations. Running case study: The fictitious organisation Aveltrine Matériaux must decide whether a coating warrants a laboratory trial. Six fictitious publications designed for teaching present partly contradictory protocols, results and conclusions. The team prepares a reasoned recommendation without extrapolating to industrial conditions that have not been studied. Transfer to the workplace: Reuse the assessment grid on a small, authorised corpus from your field, have a suitably qualified colleague review the briefing and record extraction errors before adjusting the analysis instructions.

Objectives of this training

  • Defining an analysis question and inclusion criteria
  • Extracting methodological characteristics with their source locations
  • Identifying biases and limitations to transferability
  • Comparing results without erasing differences in their protocols
  • Writing a recommendation and measuring its fidelity to the source documents

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Scientific monitoring officers and specialist information professionals
  • R&D engineers, researchers and study managers preparing technical decisions

Prior skills

  • Use a generative AI assistant and formulate structured instructions.
  • Read an article in your field and understand the concepts of sample, variability and comparator group.

What you will work on

  1. Framing the scientific decision

    Objective: Define the question and the scope of the documentation. Population or material, intervention, comparator, expected outcomes; inclusion and exclusion criteria.

  2. Preparing a traceable reading process

    Objective: Safeguard the sources provided to the assistant. Versions, appendices, pagination, usage rights; table extraction errors and limitations on access to the text.

  3. Extracting methods and results

    Objective: Build a verifiable factual matrix. Protocol, sample size, comparator, measurement, effect size, uncertainty; separation of results from commentary.

  4. Assessing the strength of the evidence

    Objective: Characterise the limitations without an artificially precise score. Selection and measurement bias, confounding factors, missing data, practical relevance and conflicts of interest.

  5. Comparing the publications

    Objective: Explain convergences and contradictions. Heterogeneity of conditions, comparability of measurements, absence of evidence versus evidence of absence.

  6. Writing and auditing the recommendation

    Objective: Produce a briefing that can be used to decide whether to conduct a trial. Source-based reasoning, conditions for validity, fidelity checks and information still to be sought.

What you take away

  • Corpus selection protocol
  • Extraction matrix with source references and critical reading grid
  • Decision briefing distinguishing evidence, hypotheses and uncertainties
  • Extraction checks and corrections log

Method and assessment

  • Short instructional inputs followed by immediate practice on the running case study.
  • Individual work followed by work in pairs, with reciprocal review of the outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Describe a work-related decision that could be informed by publications, without sharing confidential information. Check access to a tool authorised by the organisation; the exercises use only the fictitious corpus provided.

Observable criteria: Each inclusion or exclusion is justified against the stated criteria. ; Each retained result refers to a page, table or figure. ; Key figures, units and denominators are checked in the corpus. ; Each recommendation explicitly states the methodological limitations and limitations to transferability. ; The agreement rate for the checked fields is calculated and errors are corrected..

Frequently asked questions

Does AI replace reading the full articles?

No: the key passages, methods and results must be reread in the sources.

Can you share an article that requires a subscription?

Only if the usage rights and the organisation’s rules allow it.

Will a meta-analysis be conducted?

No. The day covers a critical synthesis; statistical pooling requires a specific protocol.

How should information that cannot be found be handled?

Mark it as missing or inaccessible, without asking AI to reconstruct it.

Last updated: