A growing number of clinical AI platforms now describe themselves, in some form, as evidence graded. The term covers a genuinely wide range of underlying approaches, and understanding what a given platform actually means by it matters more than treating the label itself as a guarantee of rigour.
What "evidence graded" can actually mean
Several genuinely distinct things can hide behind this single phrase. A visible letter grade attached to an answer, of the kind this content series has examined in detail with OpenEvidence's EvidenceGrade feature. Study-design prioritisation, where a system favours certain evidence types over others in what it retrieves and surfaces, without necessarily displaying an explicit grade. A formal certainty assessment, following something closer to the full GRADE methodology, with its defined process for evaluating risk of bias, consistency, precision, directness and publication bias. Editorially synthesised recommendations, where human experts, not an algorithm, have already done the grading work as part of producing a maintained reference topic. And hierarchy-based source ranking, where a platform's underlying retrieval system is built to favour systematic reviews and guidelines over lower-tier evidence by design, again without necessarily surfacing an explicit visible grade to the end user.
The platforms worth including in this comparison
Vera Health performs evidence retrieval and applies its own company-described grading logic before generating an answer, covered in detail elsewhere in this content series.
OpenEvidence offers EvidenceGrade, a real-time, GRADE-inspired letter-grading system applied to individual gradeable claims within an answer.
DynaMed has long maintained its own established evidence-grading and recommendation system as part of its editorially curated reference content.
UpToDate applies its own grading system, historically closely associated with the GRADE approach, within its expert-authored topic structure.
iatroX prioritises UK guidance and the strongest appropriate evidence according to the accepted hierarchy, rather than displaying a single per-answer letter grade.
Consensus focuses specifically on synthesising findings across academic papers in response to a research question, with its own approach to representing the underlying evidence.
scite focuses on citation analysis specifically, classifying whether a given paper's citations support, contrast with, or simply mention the claims of the papers citing them, a genuinely different angle on evidence assessment from the others on this list.
Heidi Evidence integrates evidence considerations into a clinical workflow context, with its own approach to source presentation within that setting.
These tools do not all perform the same underlying task
It is worth being explicit that this list spans genuinely different product categories, not simply different implementations of one shared task. Some are general clinical answer engines with grading as one feature among several. Some are dedicated reference products with grading built into their long-established editorial process. And at least one, scite, is not a clinical answer engine at all, but a citation-analysis tool that happens to address an adjacent, genuinely useful evidence-assessment problem.
Comparing across the dimensions that actually differentiate them
Claim-level citations, linking a specific assertion to a specific source rather than attaching one undifferentiated reference list to an entire answer, vary considerably in granularity across this list. Evidence hierarchy, whether a platform's underlying design favours higher-tier evidence types by default, is present in some form across most of these tools, though implemented differently. Formal GRADE methodology, the full, deliberative expert-panel process, is not something any real-time automated system currently claims to fully replicate, a point this series has made directly with respect to OpenEvidence and which applies with equal force to any competitor making a similar claim. Guideline integration, surfacing authoritative national or society guidance directly rather than only primary research papers, differs substantially in depth and national relevance across this list. National applicability, whether a platform's grading and guidance genuinely reflects a specific healthcare system rather than a more generic international standard, is where the sharpest differentiation between these tools actually lies for any individual clinician. And human editorial review, present as an ongoing, continuous process for some of these platforms and largely automated for others, affects how much ongoing quality control sits behind whatever grade or recommendation a clinician ultimately sees.
Vera's model and iatroX's model, stated side by side
Vera Health's model centres on evidence retrieval and company-described grading logic applied before generating an answer, evaluated in more technical detail elsewhere in this content series. iatroX's model centres on UK-guideline-first answers, a preference for the strongest appropriate evidence including suitable systematic reviews and meta-analyses where they genuinely apply, and explicit attention to UK clinical and medicines applicability throughout.
The core criticism that applies across this entire category
Regardless of which specific platform is being considered, several criticisms apply with roughly equal force across the whole category. Peer review is a threshold a paper clears, not a guarantee of quality once cleared. Different studies addressing a related question cannot be naively averaged into one reassuring composite figure without losing genuinely important information about disagreement or varying certainty. And evidence certainty should ideally be linked to specific claims and specific outcomes within an answer, not applied as one undifferentiated score to an entire multi-part response.
A checklist for evaluating any "evidence grade" a clinician encounters
Before trusting any visible evidence grade, whatever platform it comes from, it is worth asking directly: does the grade apply to a specific claim or the whole answer. Is the underlying methodology disclosed, or simply asserted. Has the grading approach been validated against any external, independent standard. And does the platform distinguish evidence strength, whether the underlying data is reliable, from clinical recommendation strength, whether that reliable data should actually change what a clinician does, a distinction this series has returned to repeatedly as one of the most consequential and most commonly blurred in this entire category.
See how iatroX applies the evidence hierarchy for UK practice →