AI prior art search for patents
Find patents, papers and products that predate a patent's priority date, each mapped against the claim elements it discloses, plus §103 combinations.
A prior art search finds the disclosures that predate a patent and could invalidate or narrow its claims. One Transform searches patents, academic literature and historical product documentation at the same time, keeps only references that predate the priority date, and maps each one against the claim elements it discloses. Litigators preparing invalidity contentions, in-house teams answering an assertion and prosecutors checking a draft use it to get from a patent number to a ranked, element-mapped reference set.
What you get
- Three reference lists: patents, academic and other non-patent literature, and products. Every reference shows which claim elements it discloses, with the supporting passage, and a relevance score taken from that coverage.
- §103 combinations. References that each cover part of a claim are paired up, and each combination shows its combined coverage, the rationale for combining under KSR, the motivation to combine, and the analogous-art and skill-level facts behind it, laid out in an element grid with one column per reference.
- An invalidity report with an executive summary, a map of the strongest art for each claim element, how each reference qualifies as prior art and on what date, secondary considerations, and the gaps that remain with possible design-arounds.
- Deep dives on the strongest references.
- A USPTO record panel for the patent under attack: recorded owner, PTAB history, prosecution events and family.
- Charts and exports: an Insights tab, and Excel, PDF and Word exports with the same content.
How it works
- Claim analysis. The patent's claims are broken into elements and the technology area is classified. You can also start from pasted claims or a technical description instead of a patent number, for example to check an invention before filing.
- Discovery from several directions at once. The search starts with the references the examiner cited and the documents the patent record lists as similar. It then follows citation chains a level further, reads the citations of foreign family members (other examiners, other art), traces the inventors' earlier patents and papers, searches the patent's classifications, and generates queries in the vocabulary of the period. Non-patent literature comes from academic and conference sources; older products come from historical product documentation on the web. References already analysed in earlier searches are recalled by meaning as well as by keyword.
- Full text before scoring. Many cited references arrive as a title only. Patents are fetched in full and papers are enriched with their abstracts, so the scoring reads what the reference actually discloses.
- Date filtering. Anything published after the patent's priority date is removed. You can also exclude the patent's own family, the same owner, self-citations, or particular assignees and patent numbers.
- Scoring. Each reference is mapped against each claim element. Its relevance score is its element coverage, so a reference that discloses most of a claim outranks one that merely shares the topic. References the examiner cited always stay in the list, with their real coverage, because they matter even when they disclose little on their own.
- Closing gaps. The search checks which claim elements are still poorly covered and runs targeted queries for those elements before it stops.
- Combinations and report. Partial references are tested in combinations, and the report is written from the mapped evidence.
The element findings are model judgments against quoted passages; the dates, citation relationships and family membership come from the patent records.
When to use it
- Responding to an assertion or a demand letter, to build invalidity contentions from a ranked reference set.
- Before filing an inter partes review, to see whether the patents and printed publications available in that forum cover the claims; inter partes review basics explains the limits.
- Before filing your own application, to check novelty against an invention description.
- When an examiner's search looks thin, to find art the examiner did not consider.
If you need a full assessment of whether the patent survives (anticipation, obviousness, §112, §101, prosecution history and double patenting together), run a validity analysis. To find patents close to an idea without a validity question in mind, use similar patents.
Limits
- Coverage depends on what is published and reachable: patent records, academic indexes and web-archived product material. Internal documents, unpublished theses and many trade publications are out of reach.
- Dates for non-patent literature can be uncertain. Each reference shows its date basis; confirm public availability before relying on it, as non-patent literature as prior art explains.
- Element mappings are machine judgments. Read the passage behind every mapping you intend to use.
- Combinations show a rationale; they do not establish obviousness. That remains a legal judgment.
- The results support an attorney's analysis. They are not legal advice or an invalidity opinion.
Frequently asked questions
What counts as prior art in the results?
Patents, published applications, academic papers, conference material and product documentation published before the patent's priority date. Later material is filtered out, and each reference shows the date it was qualified on.
Why do examiner-cited references appear even with low coverage?
The examiner's citations are the starting point of any invalidity review. They are kept with their real coverage so you can see that they were checked, and they rank below references that disclose more of the claim.
Can I search without a patent number?
Yes. Paste claims, or describe the technology in plain language, and the search builds its claim elements from that text. This suits pending applications and pre-filing novelty checks.
Does it suggest obviousness combinations?
Yes. References that each disclose part of a claim are tested in combinations, and the strongest are listed with combined coverage, the rationale for combining and the supporting facts.
How is this different from a validity analysis?
A prior art search finds and maps references. A validity analysis builds its own reference record and then assesses every ground of invalidity, claim by claim. Both are included in every plan; see pricing.


