Claim Verification API
Stop your AI agents from acting on false information. Cross-reference any factual claim against fact-check databases, news sources, and LLM evaluation. Returns a confidence-calibrated truth score.
Cross-reference factual claims against news sources, fact-check databases, and academic content.
How it works
Submit a claim and SignalStack retrieves supporting and contradicting evidence from multiple sources, evaluates it with LLMs, and returns a calibrated truth score.
1. Submit claim
Send a textual claim — a sentence, headline, or statement. Optionally provide context (topic, entities, date range) for higher-precision evidence retrieval.
2. Evidence retrieval
SignalStack queries fact-check databases, news archives, academic papers, and web sources in parallel. Retrieved evidence is deduplicated and ranked by source authority and relevance.
3. Multi-LLM evaluation
An LLM evaluates the claim against the retrieved evidence and produces a calibrated truth score (0.0–1.0) with supporting evidence citations, reasoning, and confidence estimation.
from signalstack import SignalStack
client = SignalStack(api_key="sk_...")
result = client.verify.claim(
claim="Global average temperatures have not risen in the past decade",
context={"topic": "climate change"},
)
print(f"Truth score: {result.score}")
# Truth score: 0.08
for eval in result.evaluations:
print(f"LLM ({eval.provider}): {eval.verdict}")
print(f" Confidence: {eval.confidence}")
for source in result.evidence:
print(f"{source.name}: {source.support_level}")
# "NOAA: contradicts"
# "NASA GISS: contradicts"Comprehensive claim analysis
Multi-source analysis combines fact-check databases, news signals, and LLM evaluation for robust claim verification.
Real-time fact checking
Cross-reference claims against Google Fact Check Tools, Snopes, PolitiFact, Full Fact, and 20+ fact-check organization databases. Results include claim status, rating, and source article links.
News signal correlation
Analyze claim coverage across GDELT, news APIs, and social media signals. Assess claim propagation patterns, source authority, and cross-source agreement to detect coordinated misinformation.
Academic & scientific validation
Cross-reference claims against Semantic Scholar, CrossRef, PubMed, and arXiv. Verify scientific claims with paper citations, study methodology analysis, and consensus tracking across the literature.
LLM-based truth scoring
Multiple LLMs independently evaluate claims against retrieved evidence. Cross-model agreement scoring and confidence calibration reduce hallucination risk. Supports configurable LLM providers including Groq, OpenAI, and Anthropic.
Source authority scoring
Each source is scored on authority, recency, bias rating, and historical accuracy. Sources from Media Bias/Fact Check, domain trust signals, and citation graph analysis inform the overall evidence reliability score.
Temporal claim tracking
Track how claims evolve over time with version history, first-seen timestamps, and propagation patterns. Detect claim mutation across sources with textual similarity analysis and named entity drift tracking.
Frequently asked questions
We query Google Fact Check Tools (aggregating 20+ fact-check organizations), Wikipedia, GDELT (Global Database of Events, Language, and Tone), World News API, Semantic Scholar (200M+ academic papers), CrossRef (130M+ scholarly works), PubMed (36M+ biomedical citations), and arXiv (2M+ open-access papers). Each source is searched independently and results are cross-referenced for consistency.
Each claim is evaluated by an LLM (Groq Llama-3) that receives the claim plus retrieved evidence and produces a verdict with reasoning. Additional LLM providers can be configured for multi-model evaluation.
Yes. We support claim verification in 50+ languages. Our fact-check database covers claims in English, Spanish, French, German, Portuguese, Hindi, Arabic, Chinese, Japanese, and more. The LLM evaluation pipeline supports all major languages with native prompt templates.
Temporal tracking is a core feature. Every claim check includes a timeline of related fact-checks, news coverage first-seen dates, and a recency-weighted confidence score. You can also set min_recency parameters to only consider sources published within a specified timeframe.
Yes. With the option to configure which sources to query, you can prioritize faster sources (GDELT, Wikipedia) for time-sensitive applications and set strict response timeouts.