Prior Art Explained: What Counts, and What a Search Really Covers

"Prior art" is the phrase that decides every patent's fate, and it is broader than most inventors expect. Here is what actually counts, how examiners use it, and what a search (human or AI) can and cannot promise.

What counts as prior art

Under US law (35 U.S.C. § 102, as amended by the America Invents Act), prior art is essentially anything made available to the public anywhere in the world before your effective filing date. In practice that includes:

• Granted patents and published patent applications, from any country, in any language. This is the corpus searches concentrate on, because it is well-indexed and where examiners look first. Note that applications count even if they were later abandoned or rejected.
• Non-patent literature (NPL): journal papers, conference proceedings, textbooks, standards documents, theses sitting in university libraries.
• Public disclosures: talks, posters, trade-show demos, YouTube videos, blog posts, forum threads, Kickstarter pages.
• Products and public use: something sold or publicly used can be prior art even if nobody ever wrote it down properly.
• Sales and offers to sell — the "on-sale bar." An offer to sell your own invention more than a year before filing can bar your patent, and under current case law even certain confidential commercial sales can trigger it. This one catches startups constantly: pre-orders, pilot agreements, and LOIs deserve a conversation with an attorney before they are signed.

Two things do not matter: obscurity and success. A 1970s West German utility model with three readers is exactly as lethal as an iPhone patent.

The one-year US grace period

The US gives inventors a grace period: your own disclosures within one year before filing generally do not count against you (§ 102(b)(1)). Two hard caveats. First, most of the world has no grace period, so disclosing before filing typically forfeits European, Chinese, and Japanese rights. Second, the grace period protects you from your disclosure, not from an independent third party publishing the same idea in between. Treat the grace period as an emergency exit, not a strategy.

Novelty vs. obviousness, in plain English

Examiners apply prior art through two distinct tests:

A worked miniature: suppose you claim a water bottle with (a) a double-walled vacuum body, (b) a ceramic inner coating, and (c) a lid-integrated UV sterilizer. One reference showing all three anticipates the claim and kills novelty. If instead one reference shows (a) and (b) while another shows (c) on a similar bottle, novelty survives but the obviousness fight begins: would combining them have been routine?

Novelty is mechanical: element-by-element matching against one document. Obviousness is a judgment call about whether a person of ordinary skill in the art, with all the prior art on the table, would have found your combination an obvious step. The examiner needs a reason the references would be combined, but "predictable use of known parts for their known functions" is often reason enough. This is why most rejections cite two or three references together, and why a search that only looks for exact matches badly underestimates risk.

The practical consequence for inventors: do not stop searching when nothing identical turns up. The question is whether your specific differences over the closest art would strike a skilled person as inventive or as routine. A good patentability report maps each element of your invention to the closest reference showing it, precisely because that is the shape of the examiner's analysis.

Knockout searches vs. exhaustive searches

Searches come in grades, and it helps to know which one you are buying (or running):

• Knockout search: a quick pass over the patent corpus looking for a single killing reference. Hours of effort, or minutes with AI search. Answers "is this obviously dead?"
• Patentability search: the standard professional product, $1,000-$3,000 from a firm (see costs): a structured hunt across US and major foreign patent collections, often with some NPL, delivering the ten to twenty closest references.
• Exhaustive searches (validity, FTO): litigation-grade efforts that chase foreign-language art, obscure NPL, and product evidence. Priced in the many thousands and up because searcher-hours scale with coverage.

What every search misses

Honesty about limits is part of doing this properly:

1. The 18-month blind spot. Patent applications generally publish 18 months after filing. An application filed last month is invisible to every search on earth, and it may still be prior art against you under § 102(a)(2) when it eventually publishes.
2. Language and indexing gaps. Machine translation has narrowed but not closed the gap on Japanese, Korean, and Chinese art; older documents and poorly-OCR'd PDFs index badly everywhere.
3. Non-patent everything. Product manuals, service bulletins, and defunct websites are prior art but barely searchable.
4. Judgment variance. Two competent searchers with identical tools return overlapping but different reference sets. Examiners, with years inside one art unit, routinely find things nobody else did.

This is why every credible search deliverable, human or automated, states its scope and cites its sources. Patent77's reports link every reference to its source document and say explicitly what was not covered; verify anything that matters, and treat any clean result as risk reduction rather than proof.

Sources and further reading

• USPTO Patent Public Search
• USPTO patent basics
• MPEP (Manual of Patent Examining Procedure)
• Google Patents
• Espacenet (EPO worldwide patent search)
• WIPO PATENTSCOPE