Knowledge Base Policy

ChatIBD answers questions using retrieval grounded in a curated knowledge base of clinical sources, rather than from the open-ended memory of a general language model. This policy sets out how sources are selected, sourced, versioned, and retired, so that every source in the corpus is either core IBD guidance from a recognised body or carries a recorded reason it changes IBD management.

Our aim is a corpus that is bounded, explainable, and auditable: not "everything that might be relevant", but a defined set of sources whose inclusion can be justified.

Last Updated: 2 October 2026

What the knowledge base contains

The knowledge base holds the professional sources used to ground ChatIBD's answers. These fall into two tiers:

  • Core sources — IBD-specialist guidance from recognised authorities.
  • Adjacent sources — non-IBD-guideline material admitted only where it changes how IBD is managed, such as vaccination guidance, medicines safety alerts, and the product labels (Summaries of Product Characteristics, or SmPCs) of co-prescribed drugs.

Research datasets, ethics-study materials, general website content, and user chat content are not part of the retrieval knowledge base and are governed separately.

Two questions we ask of every source

Each source is assessed on two independent questions, recorded separately:

  • Eligibility — should this source be in scope?
  • Sourcing and licensing — can we lawfully store, index, and reuse it?

A source can be eligible but not lawfully sourceable. Keeping these separate means such a gap is recorded in our internal source register ("in scope, not indexed for licensing reasons") rather than going unnoticed.

Generally, an eligible source is one that is published by a recognised professional, academic, regulatory, or medicines-information body; is date-stamped and attributable to a publisher; and is available in a stable format suitable for citation and review.

Core sources: recognised IBD guidance

A core source is admitted if either:

  1. It is published by a recognised IBD guidance body — for example BSG, ECCO, ESPGHAN, ACG, AGA, or GETECCU. Recognising a body does not automatically admit everything it publishes; each document is still assessed on its own merits.
  2. A named clinical reviewer admits it with a recorded rationale — preserving legitimate specialist judgement for high-quality sources outside the list above, while keeping every such decision explainable and attributable.

Our internal register records each core source's grade (guideline, consensus/position statement, or review) and its jurisdiction, because these carry different authority. ChatIBD does not currently use grade or jurisdiction to rank or weight sources.

Adjacent sources: the management-change gate

A non-IBD-guideline source is eligible only where the material it contains changes IBD management. It must satisfy at least one of the following, each backed by a citable source:

  • Interaction with an IBD therapy — an actionable pharmacological interaction (dose change, avoid, or monitor). For example, allopurinol with a thiopurine.
  • Peri-procedural implication — it changes management around endoscopy or surgery. For example, anticoagulants before endoscopy, or GLP-1 agonists and aspiration risk under sedation.
  • Effect on disease activity — it can precipitate, worsen, or mimic IBD activity. For example, NSAIDs triggering a flare, or checkpoint-inhibitor colitis.
  • Change to preventive care driven by immunosuppression or IBD therapy — vaccination, infection screening, or malignancy surveillance altered by IBD treatment. For example, live-vaccine contraindications under biologics or thiopurines, and pre-biologic TB and hepatitis B screening.

Prevalence alone is not a criterion. That a condition or medicine is common among IBD patients does not, by itself, admit a source. Without this line the corpus would expand without limit into general cardiology, endocrinology, and beyond. How common something is affects only how we prioritise work — never whether a source is in scope.

Drug product labels (SmPCs)

ChatIBD indexes the official product labels of IBD therapies and of a small, curated set of co-prescribed drugs. A co-prescribed drug's label is included only where co-prescribing it changes IBD management, through one of the mechanisms above. The first such drugs are the GLP-1 medicines Ozempic, Wegovy, Rybelsus and Mounjaro.

We source labels from the European Medicines Agency's centrally authorised products. Many older, nationally authorised medicines, for example mesalazine, corticosteroids and NSAIDs, have no such label, so their labels are not indexed. This is a structural sourcing limit, not an oversight.

When ChatIBD answers from a label, it says which product label it used. Where no label is indexed for a drug, it says so rather than treating the missing label as evidence that no warning exists.

Jurisdiction

Sources span jurisdictions, for example UK, European, US and international guidance. ChatIBD does not currently put one jurisdiction first. Answers cite each source by title and year, so the issuing body, and with it the jurisdiction, can be checked in the citation.

Superseded and conflicting guidance

  • Superseded guidance is retired: it is removed from active retrieval and the source list, and kept internally so that citations in earlier answers still resolve. A replacement is added where available.
  • Conflicting guidance is retained where each source is authoritative and current. Rather than hiding disagreement, ChatIBD is instructed to present the differing positions side by side.

What we exclude

  • Patient forum and social-media content.
  • Unattributed or undated content.
  • Superseded guidance, unless deliberately retained and clearly marked.
  • Promotional material, unless used solely as clearly identified product information and reviewed for bias.
  • Sources that cannot be lawfully stored or indexed.
  • Adjacent sources that fail the management-change gate.

Review and change control

  • Additions and retirements are approved by the founding team and recorded in an internal change log with the reason and impact.
  • The corpus is reviewed against this policy, and after any significant user-reported error or superseding publication.
  • Each answer records the version of the knowledge base it was generated from.

Feedback

If you believe an important source is missing, out of date, or incorrectly cited, please tell us. User feedback is reviewed as part of the process above, and clinically significant reports are escalated for review.

For queries about this policy, contact us at contact@chatibd.com.