From update alert to useful briefing
What changed in a medical guideline? Build a traceable comparison with AI
Compare two dated guideline versions with an evidence trail. Use a change table that preserves wording, population, grading and source locations.
The short answer
A useful guideline comparison shows the old wording, the new wording, where each appears and what still needs interpretation. Give AI the two actual versions and ask it to map changes before explaining them.
TRY THE WORKFLOW
Put a reference PDF to work.
Add a paper or guideline you have permission to use to AI Library, ask about a specific passage, and check the answer against the document.

Make sure you are comparing the right documents
An announcement that a guideline has changed is the beginning of a reading task. Obtain both versions from the issuing organization and record their titles, dates, jurisdictions and scope. Check whether the newer document replaces the whole guideline, updates one section, or is an addendum. An executive summary and a full guideline are different documents; label them accordingly.
Store the documents in MedSafeAI’s AI Library with clear names such as ‘Society — topic — previous version — date’ and ‘Society — topic — current version — date.’ Include relevant appendices when available. If you only have an abstract or announcement, limit the comparison to that material and keep the full-document review pending.
Use one row per recommendation or topic
Begin with extraction rather than a broad ‘what is new?’ summary. The following is our comparison worksheet, not a claim that a particular guideline changed. Keep source locations even when the AI reports no change. A row without a readable source remains unverified.
| Field | Record for each version | Why keep it? |
|---|---|---|
| Document identity | Title, publication date and section | Prevents mixing an update with a different publication |
| Population and setting | The group and situation the wording addresses | Shows whether the scope is actually comparable |
| Recommendation | Short exact passage or faithful paraphrase plus location | Makes the claimed change checkable |
| Strength and certainty | The guideline’s own labels, where reported | Preserves the authors’ distinctions |
| Change status | Changed, unchanged, added, removed or not located | Separates findings from gaps in retrieval |
Worked example: separate the recommendation category from the evidence type
This historical source-reading exercise compares two labels in Recommendation 11 of the CDC’s 2016 and 2022 opioid-prescribing guidelines. It is not a summary of current clinical guidance.
The documents have different scopes. Matching recommendation numbers does not establish that the populations, settings or full recommendations are equivalent.
| Field in Recommendation 11 | 2016 label | 2022 label | What to record |
|---|---|---|---|
| Recommendation category | A | B | Changed label: A → B. Do not call this stronger or weaker evidence. |
| Evidence type | 3 | 3 | Unchanged label: 3 → 3. This does not establish identical evidence or unchanged wording. |
Clinical meaning and application to an individual patient are not assessed from these two fields. Read the full recommendation, scope, implementation considerations and rationale in each document before interpreting the difference.
Check the exact source locations
CDC Guideline for Prescribing Opioids for Chronic Pain — United States, 2016
Publication date: · MMWR Recommendations and Reports 65(1):1–49
Recommendations, item 11: printed pages 31–32; PDF file pages 33–34, counting the first file page as 1. The recommendation begins on printed page 31.
CDC Clinical Practice Guideline for Prescribing Opioids for Pain — United States, 2022
Publication date: · MMWR Recommendations and Reports 71(3):1–95
Recommendation 11: printed pages 52–53; PDF file pages 54–55, counting the first file page as 1. The recommendation begins on printed page 52.
The source documents are available free on the CDC website. Use of these materials and links does not imply endorsement of MedSafeAI by CDC, ATSDR, HHS or the U.S. Government.
Sources checked for this example:
Ask for a comparison that can be audited
Check the largest claimed changes first. Confirm that a relocated paragraph has not been labelled a deletion and that differently numbered sections address the same question. Preserve qualifiers such as setting, exceptions and eligibility; shortening those away can change the meaning of a recommendation.
Read the rationale before writing the headline
After verifying the wording, inspect the guideline’s explanation and supporting references. Evidence Search can help you locate cited research; Strict Mode can support a citation-focused review. Distinguish ‘the authors explain the change with this evidence’ from ‘this is our interpretation.’ AGREE’s reporting checklist is a useful companion when checking whether a guideline describes its scope, methods and other essential information.
If you use Ask All AI and Consensus, give each model the same two documents and scope. Investigate differences in extraction or interpretation by returning to the source. Agreement between models does not resolve an ambiguous passage or establish that the update applies to a particular patient.
Publish a briefing with a visible boundary
Your team briefing can now state the versions reviewed, confirmed changes, unchanged points that matter, and outstanding questions. Link to both issuing-body documents and date your review. Keep local implementation decisions separate from this document comparison. When a correction or addendum appears, update the relevant rows and explain what changed in your own briefing.
Sources & further reading
Follow the original sources to check the details. Product pages describe their own products; they are not independent clinical evaluations.
- AGREE Enterprise: AGREE Reporting Checklist agreetrust.org
- MedSafeAI: medical AI workspace and product features medsafeai.com
For healthcare professionals. AI responses and agreement between models do not replace source verification or clinical judgment.