← Back to this week

How we choose what is worth knowing

GI Signals is not an alert and not a chatbot. It is a short, scored, physician-vetted list, and it shows the math. Here is exactly how a paper gets from the literature to a card.

The pipeline, in plain language

01

Scan

Every week we pull new research across 25+ leading gastroenterology, hepatology, and general-medicine journals, the outlets a practicing GI actually needs to keep up with.

02

Screen out the noise

Most of what publishes is not decision-relevant. First we drop what is not a study at all: letters, editorials, comments, corrections and retraction notices, and case reports. A correction carries its original paper's title and design, so it reads like the study it corrects unless it is caught deliberately. Then we drop thin observational work and papers that do not bear on care, so the list stays short. Reviews, meta-analyses and guidelines stay: they carry no primary data, but they are how a standard actually changes.

03

Score

Each surviving paper is scored on five things: is there something to do, how many of your patients it touches, what is at stake, how strong the evidence is, and how likely you are not already doing it. Every card opens to show where the paper lands on each axis, Strong to Limited, against the rubric below. Evidence strength is capped by the study design, so a retrospective study never displays as strong evidence however well it scores otherwise.

See the scoring rubric ↓

The impact factor shown on each card comes from a single hard-coded table (Journal Citation Reports 2026 release (2025 Journal Impact Factor), as of 2026-07-18), refreshed on a fixed cadence rather than fetched per paper, so the same journal always reads the same way within an issue. It is context, not one of the five axes: where a paper was published does not change what it found. Hover any IF on a card to see its source.

04

Anchor to the standard of care

New is judged against current. Each paper is matched to the guideline that governs its clinical question, drawn from a maintained reference set of current standards, each with its source society and the date it was last reviewed. The paper is then placed by how it moves that standard: reinforces it, refines it, changes it, or is not yet actionable. Every card quotes the standard it was measured against, with a link to the source, so the verdict is something you can check and argue with rather than take on trust. A finding earns attention only when it moves the standard, not because it is recent.

05

Physician review

Each issue is reviewed by a gastroenterologist, Simon Mathews, MD, before it publishes: the selection, the leads, the verdicts against the standard, and anything the checks flagged. The review is of the issue as a whole, not a signature on every card, so read each card as an AI-assisted summary that a physician has looked over, not as a clinical opinion.

The first pass is AI-assisted; the judgment is human. Every issue is human-reviewed before it is surfaced or emailed. We show the funnel, the evidence grade, and a primary-source link on every paper because a source you can check is the only kind worth trusting. What was read but not selected is listed at the foot of each issue, with its score, so the filter is inspectable rather than assumed.

The scoring rubric

Every card's score opens to show where the paper lands on each of the five axes. Each axis is rated on its own, so you can compare quality directly. Here is what the levels mean.

Actionabilityis there something to do
StrongA specific action: order, stop, choose or counsel
GoodAn option to weigh for some patients
ModerateShapes a decision, no specific action yet
LimitedNothing to act on yet
Prevalencehow much of a general GI practice it touches
StrongSeen weekly in a general GI clinic
GoodSeen monthly in a general GI clinic
ModerateA defined subgroup of a common condition
LimitedRare disease or tertiary-only care
Stakeshow consequential the decision is
StrongMortality, cancer, surgery or major drug harm
GoodHospitalisation or major morbidity
ModeratePersistent symptoms or quality of life
LimitedComfort, convenience or cost alone
Evidence strengthhow far the evidence supports the action, capped by study design
StrongRandomised trial, meta-analysis, systematic review or guideline
GoodProspective cohort, or a trial with limitations
ModerateRetrospective, observational or narrative review
LimitedCase series, case report or preclinical work
Noveltyhow likely you are not already doing it
StrongFirst of its kind, or stop doing it
GoodKnown but not yet routine practice
ModerateExtends known findings to a new setting
LimitedConfirms what is already done

Evidence strength is capped by the study design: a randomised trial, meta-analysis or systematic review may reach Strong; a prospective cohort, Good; an observational or retrospective study, Moderate; a case series or preclinical work, Limited. A paper never displays an evidence strength its design cannot support, however the rest of the read scores it.

Why this beats an alert or a chatbot

A keyword alert gives you everything and sorts none of it. A general chatbot will summarize a paper you already found, with no accountability for whether it is right or whether it matters.

GI Signals does the opposite: a short, ranked, scored list; each card anchored to the standard of care; each one shaped by a named physician's editorial judgment, not an anonymous model; and the scoring shown so you can weigh it for yourself. That is the difference between a feed and a signal.