A prior art search tells you whether your invention has already been disclosed in the patent or scientific literature. The grader above screens yours in about 90 seconds. Below is what it measures, what comes back, and how a prior art search differs from a novelty search and an FTO search.

How this free AI prior art search works

Most patentability searches start with a claim set. This one starts with two plain sentences, because at the point where the decision actually matters you usually do not have claims yet.

You describe what the thing is — a product or a method — and then what is new about how it works. That second sentence does the work: it is the feature the search runs against, and the feature an examiner would test for novelty. From there the agent extracts your technical features, builds the search elements and classification scope, and runs semantic retrieval, three Boolean passes and citation tracking before comparing the closest documents against your feature set.

It runs on PatSnapGPT, which is pre-trained on 208M+ patents and papers rather than on the open web. No claim drafting, no classification codes, no Boolean syntax on your side. Your description is never used for model training.

What the patentability grade measures

The score runs 0–100, where 0 means fully anticipated and 100 means novel. It is a weighted composite, and every dimension is shown with its weight so you can see which part is dragging the result down instead of arguing with a single opaque number:

• Feature novelty — 40%. How many of your technical features appear in no reference at all. This carries the most weight because it is the question a novelty objection turns on: if the feature is disclosed somewhere, nothing else in the assessment rescues it.
• Closest-reference distance — 25%. How much of your invention the single nearest reference already covers on its own. A low score here is the hardest position to argue out of.
• Combination risk — 20%. How easily two or three references combine to cover you. This is the §103 obviousness question, and it is the one inventors most often miss searching alone — no single document looks threatening, but a pair of them does.
• Claimable space — 15%. How much unclaimed room is left around your differentiators. A high score here alongside a low novelty score usually means the invention is patentable but the claim scope needs to move.

A result in the middle of the range — say 56 out of 100 — is reported as partially anticipated: parts of what you described are already disclosed, and the useful output is not the number but which parts. Each dimension shows its own sub-score and the reasoning behind it, so you can check the arithmetic.

What comes back in 90 seconds

1. A graded assessment. The composite score plus each of the four dimensions above, weighted and shown separately.
2. The closest prior art, ranked. Each reference arrives with its patent figure, a similarity percentage, and the actual matching passage — so you read the disclosure yourself rather than trusting a relevance score. References carry citation-prediction tags: PS-X for a document that on its own defeats novelty or inventive step and is the kind of reference likely to be cited against claim 1, PS-Y for one that only challenges inventive step in combination with another, and PS-A for background context not used to assess novelty.
3. What to fix before filing. Claim-level moves rather than a verdict: which feature belongs in claim 1, which independent claim is too broad to survive the closest reference, and where there is unclaimed space worth taking. A recommendation to add a method claim because no reference claims your particular sequence is worth more than the score itself.

Prior art search vs. novelty search vs. freedom-to-operate

These get used interchangeably and they answer different questions.

A prior art search is the broadest: it looks for any disclosure — patents, applications, published papers, products, public web pages — that predates your filing date, whatever its legal status. It is the input to everything else.

A novelty search, also called a patentability search, is a prior art search aimed at one question: can this be patented? It cares about publication dates and disclosure, not about whether rights are still alive. An expired patent from 1985 defeats novelty just as effectively as one granted last year.

A freedom-to-operate search asks the opposite: can I sell this? It looks only at rights still in force in the markets you sell into, and an expired patent is irrelevant to it. FTO is a launch decision; novelty is a filing decision. A good result on one tells you nothing about the other.

The grader above is a novelty screen. If you are heading to market rather than to the patent office, run an FTO search instead.

How to describe your invention so the search actually works

Result quality depends almost entirely on the second sentence. The pattern that works is like X, but Y: name the familiar thing, then name the one mechanism that is different.

• Mechanical. Like a charging case for wireless earbuds, but a magnet beside each cavity self-aligns the earbud.
• Software. Like on-device personalisation of a speech model, but only encrypted weight updates ever leave the phone.
• Chemistry. Like a Li–P–S sulfide solid electrolyte, but ALD-coated at 2–5 nm and rare-earth doped.

Note what these have in common: the “but” clause names a structure or a step, not a benefit. “But it is more efficient” cannot be searched. “But the coating is applied at 2–5 nm” can. If your description is too thin to search at feature level, the tool asks a few targeted questions rather than guessing.

Where the free screen stops

The grade above is a screen on the top 5 documents. That is enough to decide whether to keep going, and not enough to file on. Two things it deliberately does not do: it does not compare every candidate reference feature by feature, and it says nothing about in-force rights.

Both of those are what the full agents produce. A Novelty Search report compares every retained candidate against every feature across 174 jurisdictions, source-linked throughout, and returns a feature comparison table, an overall reference matrix and the full reproducible search strategy. A Freedom-to-Operate report covers rights still in force in the markets you are launching into, with claim charting. Both take hours rather than the days a search firm needs, and both cite every document they rely on — so the attorney reviewing it can check the reasoning instead of taking it on trust.

Frequently asked questions

What does a prior art search include?

A thorough prior art search includes granted patents and published patent applications across all relevant jurisdictions, plus non-patent literature: journal articles, conference papers, theses, standards documents, product manuals, datasheets and public web pages. Anything publicly available before your filing date can be prior art — it does not have to be a patent. The free Patentability Grader runs semantic retrieval, three Boolean passes and citation tracking, but compares only the top 5 documents. Eureka IP’s Novelty Search Agent compares every retained candidate across 208M+ patents and 216M+ non-patent documents in 174 jurisdictions.

How much does a prior art search cost?

A professional prior art or patentability search from a law firm or search vendor commonly runs to several hundred dollars and takes about a week; our own benchmark uses $600 and 7 days as the reference point. That cost is why many inventors skip the search and find out about the blocking reference in an office action instead. The Patentability Grader is free and takes about 90 seconds. Eureka IP’s Novelty Search Agent brings that same $600, 7-day third-party search down to roughly $100 and half a day.

Can I use ChatGPT for a prior art search?

Not reliably. Benchmarked on X-document retrieval — finding the single reference that independently defeats novelty — PatSnap’s Novelty Search Agent reaches an 81% top-100 hit rate, against 32% for ChatGPT-o3 and 9% for DeepSeek-R1 (September 2025). General-purpose chatbots have no patent database behind them, no examination-rule knowledge and no reproducible search strategy, so they tend to produce plausible-looking patent numbers that do not hold up. They also log your inputs; PatSnap commits contractually to zero use of your data for model training.

How is the patentability grade calculated?

It is a weighted score over four dimensions, all shown on the results page: feature novelty (40%), how many of your technical features appear in no reference at all; closest-reference distance (25%), how much of your invention the single nearest reference covers on its own; combination risk (20%), how easily two or three references combine to cover you, which is the §103 question; and claimable space (15%), how much unclaimed room is left around your differentiators. Each dimension shows its own sub-score and the reasoning behind it, so you can check the arithmetic rather than trusting a single opaque number.

What is the difference between a prior art search and a novelty search?

In practice the terms are used interchangeably. Prior art search describes the activity: finding everything published before your filing date. Novelty search, or patentability search, describes the purpose: assessing whether your invention can be patented. A search run to invalidate someone else’s patent is usually called an invalidity or validity search, and one run to check infringement risk before launch is a freedom-to-operate search.

Can this tool tell me whether my invention is patentable?

It answers the first and most decisive question: novelty under 35 U.S.C. §102. If a single prior-art reference already discloses your invention, nothing else matters — you cannot patent it as drafted. A complete patentability assessment also needs §103 non-obviousness, §101 eligibility and §112 sufficiency, which depend on legal judgment about your specific claims. Use the grade as a go / no-go signal before spending money on drafting. It is not a legal opinion.

Is a free prior art search as good as a professional patentability search?

No, and it is not meant to be. A free search is a screening layer: it tells you whether an obvious blocking reference exists so you can decide whether to keep going. A professional patentability search, or Eureka IP’s Novelty Search Agent, produces a defensible, reproducible record across jurisdictions and non-patent literature, with element-level comparison you can hand to a patent attorney or put in a due-diligence file.

Is my invention description kept confidential?

Yes. PatSnap commits contractually to zero use of user data for model training, with encryption at rest and in transit, and you retain ownership of everything you enter or edit. A genuine public disclosure elsewhere can destroy your novelty, so search before you publish, not after.

Which jurisdictions and languages are covered?

Eureka IP covers 174 jurisdictions and 1.62B+ data points, with cross-language semantic search, so a Chinese utility model or a Japanese application can be surfaced against an English-language invention description.

What is an example of a prior art search?

Say the invention is a phone case with a magnet ring moulded into the wall. A prior art search would split that into its technical features — a case, a magnet, the moulded-in position, the attachment to a metal surface — turn each into search terms and classification codes, run those against patent databases and non-patent literature, then compare the closest documents feature by feature. The output is not a list of hits: it is a statement of which features are already disclosed and which are not.

Where can I search prior art?

Free options include Google Patents, Espacenet, the USPTO Patent Public Search and WIPO Patentscope. All four are keyword and classification tools: they will find documents, but you build the queries and read the results yourself, and none of them covers non-patent literature well. This page is different in kind — you describe the invention in plain English and an AI agent builds the search strategy, runs the retrieval and does the feature-by-feature comparison for you.

What is the difference between a prior art search and an FTO search?

A prior art search asks whether your invention is new enough to patent, and looks at everything published before your filing date. A freedom-to-operate search asks whether selling your product would infringe someone else’s rights, and looks only at patents still in force in the markets you sell into. Different question, different date logic, different result. Run a patentability search before you file and an FTO search before you launch.

How do I know if my idea is patentable?

Four tests have to pass: it must be novel (35 U.S.C. §102), non-obvious (§103), eligible subject matter (§101) and described well enough to be enabled (§112). Novelty is the one you can check cheaply and the one that decides whether the rest is worth paying for, because if a single earlier document already discloses the invention, nothing else matters. That is what this tool measures. Eligibility and sufficiency need a patent attorney’s judgment about your specific claims.

Related reading

• What is a patent novelty search?
• Prior art search tools for patent attorneys
• How to search design patents