Originality is estimated from the evidence examined

Is this original?

Compare your idea with what public search and base AI models already produce.

IsThisOriginal doesn't claim to search all human knowledge. It shows where your idea matches—or moves beyond—the evidence examined.

The accessible idea frontier

An idea is hard to place on its own. It becomes tractable when you ask a narrower question: how far does it sit beyond what someone could reach cheaply? We measure two such baselines, and report where your idea falls relative to each.

The public-search frontier

What an ordinary person could find with basic web searches. We plan 10–20 queries across your exact position, its paraphrases, its mechanism, the relationships between your concepts, adjacent positions, and the positions that contradict you.

Every result is classified by its relation to your idea — not by keyword overlap. Two texts using the same words are not the same claim.

The model frontier

What a base language model produces from ordinary, low-effort prompts. We sample the model dozens of times across escalating prompt tiers, split the answers into atomic claims, cluster them, and check whether your idea is among them.

The prompts never mention your position or your vocabulary. Feeding those in would manufacture a match and make the comparison worthless.

Model-accessibility ladder

Each part of your idea gets placed on a ladder describing how much effort it took the baseline to reach it.

Level 0Produced directly by a basic prompt
Level 1Produced through repeated basic sampling
Level 2Produced only after requesting unconventional alternatives
Level 3Produced only through multi-step reasoning
Level 4Reconstructable only after supplying key assumptions
Level 5Not produced within the tested baseline

What Level 5 does and does not mean

Level 5 means a facet was not produced during this analysis. It is not a claim that the model does not know it, could not produce it with different prompting, or that no human has had the thought.

Explainable scoring

Every number is produced by a deterministic function over stored evidence. Language models supply observations — relation labels, ordinal ratings, counts — and never author a score. The full arithmetic is shown on every report, so you can disagree with a specific term rather than with an opaque total.

Scores are heuristic estimates. They describe where your idea sits relative to the evidence this run examined, and nothing more.

Originality Estimate

Public-search deviation: 30%
Model-baseline rarity: 25%
Mechanism & structural novelty: 20%
Consequential difference: 15%
Specificity: 10%

Qualified Originality

Estimate × Coherence Factor × Testability Factor. An unusual idea that can't be tested is reported as unusual, not as promising.

What this tool will not tell you

Whether your idea is truly novel. Absence of a match is weak evidence. It means these queries did not surface a prior statement, not that none exists.
Who or what wrote something. This is not an AI-writing detector and makes no attribution claims.
Anything about unsearched sources. Paywalled journals, books, theses, patents, non-English work and private documents are outside every run.

Analyze an idea Read the method in full

This deployment is in Demo Mode

Search results and model generations come from fixtures. The pipeline, the durability, the clustering and the scoring are all real — only the evidence is scripted. Add a model provider key and a search provider key to run live research.