AI patent infringement search

Find products that may practice a patent's claims. Each candidate is charted element by element, with quoted evidence and a score based on coverage.

An infringement search starts from a patent's claims and looks for commercial products that may practice them. One Transform splits each claim into its limitations, searches the open web for products whose documentation appears to meet them, and charts every candidate element by element with the evidence it found. Licensing teams, patent owners and litigation counsel use it to build a first list of targets and to see how strong each read is before spending attorney time on it.

What you get

  • A ranked list of accused products. Each entry names the product and the company, links the product page that was checked, and carries a score from 0 to 10.
  • A claim chart for every product. Each limitation is marked present, partially present or not found, next to the passage of product documentation that supports the finding.
  • Top matches and key gaps. For each product you see which limitations carry the read and which ones are missing.
  • Deeper analysis of the strongest products. It covers literal infringement, the doctrine of equivalents and possible design-arounds, means-plus-function limitations and claim construction, induced and contributory infringement, willfulness, and damages questions such as the royalty base and apportionment.
  • An expert report across the top accused products, with a portfolio ranking, consolidated claim construction, reports on the main defendants, damages, indirect infringement and design-around options.
  • A USPTO record panel for the patent: the recorded owner, any PTAB trials, the number of prosecution events and the US continuity family.
  • Charts and exports. An Insights tab shows coverage across products and the spread of scores. The full result exports to Excel, PDF and Word with the same data in each, and any single product can be exported on its own.

How it works

  1. Claim parsing. The patent is fetched and its claims are broken into labelled elements. US patents granted too recently for general patent databases are read from the USPTO's own full text. You can also paste claim text instead of, or alongside, a patent number.
  2. Search angles. The engine writes a large set of search angles: by claim element, by market and by likely implementer. Each angle is rated for promise, and the most promising run first.
  3. Discovery and evaluation. Each angle looks for products and evaluates them against every element in the same pass. Product links are checked and duplicates merged. When the first angles stop turning up new, well-covered products, the search widens to adjacent markets, component and module suppliers, regional vendors and recent entrants, and then stops.
  4. Scoring. A product's score is its claim coverage and nothing else. Present elements count fully, partial matches count for less, and any element the evaluation did not address counts as not found. A product without an element chart is dropped rather than scored. A product that reads on eight of ten limitations outranks one that only sounds similar.
  5. Deep dives and report. The top products get the deeper analysis, and the report is written from the charts and evidence already collected.

Element findings are model judgments made against quoted evidence; the score is arithmetic on those findings. That split is deliberate: you can recompute any score from its chart, and you can overrule a single row without distrusting the rest.

If you add several patents, each runs as its own search, so the same market can be compared across a family. Filters let you name target companies, countries or regions, industries and technical features, exclude companies or products, and add free-text criteria.

When to use it

  • Before a licensing campaign, to find who sells products that read on a patent and rank them by coverage.
  • Before asserting a patent, to test whether the strongest read holds element by element.
  • When valuing a patent you own or are thinking of buying, to see how much of the market it reaches.
  • When a family has continuations, to see which member reads most strongly on the same products.

Other search types answer neighbouring questions. To check whether your own product is exposed to someone else's patents, run a freedom to operate search. To test whether the patent itself would survive a challenge, run a validity analysis or a prior art search.

Limits

  • The evidence is public: product pages, datasheets, manuals and technical articles. Features that are never described publicly, such as internal firmware or server-side processing, show as partial or not found even when they exist.
  • "Not found" means the search found no evidence, not that the product lacks the feature.
  • Element findings are machine judgments. Read the quoted evidence and the linked page before relying on any row; how to read an element-by-element claim chart covers what to check.
  • Claim construction affects every row. The analysis reads the claim language as written and raises construction questions in the deep dives; it does not resolve them.
  • The results are research to support an attorney's work. They are not legal advice and not an infringement opinion.

Frequently asked questions

How is the infringement score calculated?

The score is the share of the claim's elements that a product covers, scaled from 0 to 10. Partial matches count for less than full matches, and elements without evidence count as not found. It is never an overall impression of similarity, so every score can be checked against its chart.

Can I run the search on claim text instead of a patent number?

Yes. Paste one or more claims and they are parsed the same way as a granted patent's claims. This suits pending applications or amended claims.

Does it find indirect infringement?

The deep dives on the top products assess induced and contributory infringement and willfulness from public evidence. Knowledge and intent usually depend on facts outside public documentation, so treat those sections as a list of what to establish rather than a finding. The doctrine of equivalents is covered in the same deep dives; the doctrine of equivalents, explained sets out the tests.

Can I steer the search toward particular companies or markets?

Yes. Name target companies, countries or regions, industries and technical features, or exclude companies and products you have already cleared or licensed. Free-text criteria steer discovery as well.

What does a search cost?

Each patent you search counts as one search on your plan. The standard plan includes 40 searches a month after a 7-day trial with 5 searches, and the bring-your-own-AI-keys plan has unlimited searches. Current plans are on the pricing page.

Try it on one of your patents

The fastest way to evaluate One Transform is to run a search on a patent you already know well.

7-day trial with 5 searches · No credit card · Work email required