Methodology

How ClaimHit scores
patent infringement

ClaimHit does not collect AI outputs and rank them by confidence. Every result is evaluated across four independent factors before a risk level is assigned. A result only reaches HIGH when all four align.

The architecture

Nine models. Zero coordination.

When you submit a patent, ClaimHit runs nine frontier AI models simultaneously — Claude Sonnet, Claude Haiku, GPT-4o, GPT-4o Mini, Gemini 2.0 Flash, DeepSeek, Mistral, and Perplexity Sonar. Each model receives the same patent claims and searches independently without knowledge of what the others found.

A single model producing a confident result is hard to validate — confident hallucinations look identical to confident accurate results. Nine independent models converging on the same target is a qualitatively different signal. Independence is what makes the consensus meaningful.

Claude Sonnet
Anthropic
ACTIVE
Claude Haiku
Anthropic
ACTIVE
GPT-4o
OpenAI
ACTIVE
GPT-4o Mini
OpenAI
ACTIVE
Gemini 2.0 Flash
Google
ACTIVE
DeepSeek
DeepSeek
ACTIVE
Mistral
Mistral AI
ACTIVE
Perplexity Sonar
Perplexity
ACTIVE
Scoring

Four independent factors

Each factor captures a different dimension of infringement likelihood. All four must align for a HIGH result.

01
🤝

Model Consensus

How many of the nine independent AI models flagged the same target. Each model searches without knowledge of the others — convergence across models from different companies with different training data is a strong signal. A target flagged by 7 out of 9 models independently is categorically more significant than one flagged by 2.

This is why ClaimHit results hold up under scrutiny. Convergence is evidence.
02
📋

Claim Element Coverage

How many elements of the independent claim appear to be implemented — and which ones. Claim elements are not equal. Preamble language sets the context but doesn't define the invention. The novel core — the inventive steps — is where infringement actually turns. ClaimHit weights element matching toward the inventive steps, not the boilerplate.

Matching the preamble scores low. Matching the inventive core scores high.
03
🔍

Evidence Strength

The quality and specificity of citations underlying each match. A specific product datasheet URL with a version number and feature confirmation outweighs an inference from market position. HIGH risk requires at least three claim elements to have specific documented evidence — datasheets, standards clauses, FCC filings, developer documentation. Inference without documentation cannot produce a HIGH result.

No documentation = no HIGH. Specificity matters.
04
⚖️

Functional Equivalence

Whether the accused product performs the same function, in substantially the same way, to achieve the same result as the claimed invention. This is the doctrine of equivalents test applied at the preliminary screening stage. A product that achieves the same outcome through a technically different route may still infringe — and ClaimHit tests for this explicitly rather than requiring literal feature-by-feature matching.

Same function. Same way. Same result. The legal standard applied early.
Noise filtering

Why some high-consensus results still score MEDIUM

A noise penalty is applied when model agreement is high but documented evidence is thin. This filters a specific failure mode: many models agreeing that a company is relevant to a technology space without being able to find specific claim-element evidence.

Without this penalty, large companies in adjacent technology areas would consistently appear as HIGH risk — not because they infringe, but because AI models associate them with the technology. The noise penalty keeps HIGH meaningful.

Risk levels

What each level means

HIGH
Multiple models agree, core claim elements match, specific documented evidence exists for at least 3 elements. Warrants formal attorney analysis.
MEDIUM
Strong potential requiring further investigation. May reflect proprietary specifications, undisclosed implementations, or markets where public documentation is limited. Worth a closer look.
LOW
Filtered from results by default. Score based on inference only — treat as a directional lead at most. Not shown in standard results.
⚖️ Legal Disclaimer: ClaimHit Hit Scores are for preliminary research only. They indicate where attorney time is best focused — they are not infringement opinions, do not constitute legal advice, and should not be relied upon as the basis for legal action without formal analysis by a qualified patent attorney.

See it in practice

3 free searches. No credit card. Results in ~60 seconds.

Try ClaimHit Free →