The engine
How it works
A professional prior-art search is a method, not a secret database. Patent77 runs that method as four passes and shows its work at each one. Here is each pass, and — just as important — what the engine cannot do.
Pass 1 — Plan
The engine reads your plain-English description and extracts claim-style elements: the discrete technical pieces an examiner would compare against the record. A self-watering planter becomes a reservoir, a moisture sensor, a pump, a self-calibrating controller — not one fuzzy product description.
For each element it builds the synonym table a professional searcher keeps in their head: pump → fluid transfer device, peristaltic pump, wicking element. Patent drafters write in deliberately broad vocabulary; searching only your own words is the classic DIY failure.
Pass 2 — Search
It then drafts and executes corpus queries against the live Google Patents record: single elements, element pairs, expanded vocabulary, classification-informed variants. Queries iterate — if a query surfaces a close document, its language and classification codes seed the next round.
Every executed query is returned with the result. If you would have phrased one differently, you can see that, run your own, and compare. The search is auditable by design.
Pass 3 — Read
Candidate documents — often hundreds — are read against your elements and reranked. Each surviving reference gets a relevance score from 0 to 1, a one-line statement of what it teaches that overlaps, and the list of your elements it appears to read on.
Titles and abstracts are marketing; claims and descriptions are law. The reading pass works from the document text, and the result records how many candidates were considered so the winnowing is visible.
Pass 4 — Map
The paid report maps every element against every close reference: reads on, partial, or not shown, with a note on what the reference actually discloses. One reference reading on every element is a novelty problem; a combination covering them all is an obviousness question.
Around the matrix the report writes the examiner-style narrative — the argument a rejection or an allowance would make — a headline verdict, and the limitations section stating what this search could not see.
The honesty section
Limits, stated up front
Search products fail people by implying completeness. These are the boundaries of what a Patent77 run can tell you.
01
What we search
The live patent record — patents and published applications in the Google Patents corpus, US and major international collections — queried at search time, not from a stale snapshot.
02
What we can't see
Non-patent literature (papers, manuals, products), applications still inside the 18-month publication window, and collections beyond the corpus. Examiners can cite those; no patent-record search can find them. Every report says so.
03
Why every reference is linked
A language model can misread a document, so we never ask you to take a reference on faith: publication number and canonical source link, every time. If a reference matters to a real decision, read it — or have your attorney read it.
For the deliverable itself, see the example patentability report; for the method by hand, the DIY search guide.
No card · cancel anytime
Watch it run on your idea.
Three free searches a month. The queries, scores and sources are all on the page.