AI Agent Cherry-Picks Evidence and Ignores Contradictions: Causes and Fixes

Issue: AI agent selectively cites supporting evidence from retrieved context while ignoring information that contradicts it.

Frequency: Common

Symptoms

Answer presents one-sided view when context is balanced
Caveats, exceptions, or limitations omitted
User gets incomplete picture
Decisions made on partial information
Commonly reported in RAG pipelines (LangChain, LlamaIndex) where the model favors a coherent narrative over surfacing every retrieved passage

Root Cause Model may select evidence that best fits a coherent narrative, ignoring qualifying information or exceptions that would complicate the answer.

Example

Retrieved context:
"Clinical trials showed the treatment was effective in 67% of 
patients. However, 23% experienced significant side effects 
including nausea and headaches. The treatment is not recommended 
for patients with heart conditions or those over 65."

Query: "Is this treatment effective?"

Agent: "Yes, clinical trials showed the treatment was effective 
in 67% of patients."

Omitted: Side effects, contraindications, age restrictions

Result: User not informed of risks and limitations

Mitigation Strategies

Balanced response instructions: “Include caveats and limitations”
Structured extraction: Require pros/cons, conditions, exceptions
Comprehensiveness scoring: Measure coverage of context
Devil’s advocate check: Ask “what did you leave out?”
Multi-turn clarification: Follow up on omissions
Template responses: Ensure sections for limitations

Detection

Compare answer coverage to context content
Track caveat/limitation inclusion rates
User feedback on incomplete answers
Measure answer comprehensiveness

How to fix it: force the model to explicitly weigh contradicting passages before answering and check answer comprehensiveness against all retrieved evidence — see Mitigation Strategies below.

Mitigations

Prevention

Implement query-answer consistency validation: Decompose complex queries into atomic components and verify each component is addressed in the answer before returning. Use RAGAS Answer Relevancy metric (target: >0.75) with automatic re-generation for scores below threshold. Root cause mitigation: Prevents context-anchoring by explicitly binding answer generation to parsed query intent.
Apply multi-source consensus verification: Require answers synthesized from multiple sources to explicitly cite which sources support each claim and flag unresolved contradictions. Implementation: Use semantic similarity checks across source fragments to detect cherry-picked evidence patterns. Root cause: Ensures balanced representation of evidence when sources conflict.
Enforce comprehensive coverage checks: Implement structured extraction requiring explicit treatment of caveats, limitations, exceptions, and counterevidence for each claim type. Use template responses with mandatory caveat sections. Root cause: Prevents omission of qualifying information that would change user decision-making.

Detection & Response

Answer completeness monitoring: Measure coverage of query intents in generated answers. Track query decomposition rate (% of query components explicitly addressed) and flag responses with coverage <85%. Instrument RAG pipeline to log query-answer similarity scores per component. Alert on sustained scores <0.70.
Evidence balance scoring: For each answer, compute evidence distribution across sources and flag one-sided responses (>70% from single source on multi-source queries). Implement automated extraction of caveat/limitation mentions and track inclusion rates by query type. Target: >80% of medical/financial answers include relevant caveats.

Architecture Patterns

Query Intent Decomposition Graph: Parse complex queries into a DAG of atomic intents before retrieval. Each retrieved document is mapped to specific intent nodes. Answer generation must satisfy all leaf nodes. Validation layer computes coverage before response generation.
Evidence Consensus Engine: Maintain a fact graph where each claim is attributed to specific sources with confidence scores. Multi-source claims require consensus computation (intersection of sources supporting claim). Flagging layer surfaces contradictions to generation model.
Structured Response Templates: Use task-specific response schemas that enforce inclusion of: primary answer, supporting evidence, relevant caveats/exceptions, alternative interpretations, confidence bounds. Auto-flag template violations before user delivery.

Key Metrics

Metric	Target	Alert Threshold	Measurement Method
Answer Relevancy Score	>0.75	<0.70	RAGAS metric on generated answers vs. original query
Query Coverage Rate	>90%	<85%	Percentage of query components explicitly addressed in answer
Evidence Balance Index	>0.6	<0.4	Distribution of citations across sources (Gini coefficient, 0=balanced, 1=single-source)
Caveat Inclusion Rate	>80%	<70%	Percentage of medical/financial answers including relevant limitations
User Clarification Rate	<5%	>10%	Percentage of answered queries requiring follow-up clarification

Alerts & Escalation

Alert	Condition	Severity	Response
Low Answer Relevancy	Answer Relevancy Score < 0.70 for >5% of queries in 1-hour window	HIGH	Page on-call; trigger re-generation with query reinforcement prompt
Single-Source Dominance	Evidence Balance Index < 0.4 on multi-source queries for >3 consecutive queries	MEDIUM	Log event; audit cherry-picking patterns in retrieval/synthesis
Rising Clarification Demand	User Clarification Rate exceeds 10% (vs. 5% baseline) over 24-hour window	HIGH	Investigate query decomposition or answer template effectiveness

References

Stanford Legal RAG Hallucinations - Selective evidence use
Mindee: RAG Hallucinations Explained - Incomplete synthesis