The new discovery engine searches the open web, product libraries, citation records, and competitor data, and runs several frontier models together to find real products. Two checks sit on top: one confirms every result against a real source, and one removes anything that does not actually meet the claim. ClaimHit soft-launched at IPBC Global in San Diego, where more than 20 licensing executives ran their own patents through it. Many are now in trials.

CHEYENNE, Wyo., July 23, 2026 /PRNewswire/ — Most patent licensing programs begin with a statistical analysis on a portfolio. A team has hundreds of patents and enough budget to pursue a few of them, so it has to decide which ones are worth the effort before doing any real work.

The usual way to decide is to score the patents. Count the forward citations, measure the claim breadth, weigh the family size, and rank what comes out. It takes weeks, and at the end you still do not know the one thing that matters: whether anyone is actually selling a product that comes close to the patent.

A citation count cannot tell you that. Only looking can.

ClaimHit LLC today released V3, which is built to do the looking.

How the engine works

The workflow is straightforward: start with a patent, discover the products close to its claims, pick one, and build a claim chart against it.

Discovery is the part that is hard to do well. To find the products close to a claim, the engine looks in several places at once: the open web, product and specification libraries, citation records, and competitor data: and runs several frontier models together rather than one model working alone. That much is not unusual. What matters is the two checks that follow. The first confirms every candidate against a real, retrievable source, so nothing reaches the user because a model happened to assert it. The second drops any product that is not genuinely close to the requirements of the claim, not merely the general subject of the patent. What is left is a short list of real products, each with evidence a user can open and read.

From that list, the user selects a product and the engine builds the claim chart, mapping the claim against the product with the supporting evidence attached. A product uploaded by the user, such as a teardown report, can be charted the same way.

V3 runs this in two modes. One patent at a time, or a batch of them uploaded together.

The Hit Matrix

Batch mode has a feature of its own: the Hit Matrix. When a set of patents is uploaded together, the results come back as a single grid. Patents sit along the top. Companies run down the side. Where a patent has a product close to it, the cell fills in. The grid builds as the search runs, so a user watches it take shape.

It is easy to read. One company lights up across ten of thirteen patents, and that is a conversation worth having. One patent fills most of its row, and that patent is worth pursuing. Another patent stays empty, and no scoring model was needed to tell you so.

From any filled cell, a user can open that patent and build a full claim chart. The whole grid exports to Excel. Work that used to take weeks now takes minutes, and it produces real products instead of a ranked list of maybes.

What else is new in V3

Whole-portfolio analysis. Run an entire portfolio in one pass instead of one patent at a time.
Your own teardown reports. Upload a teardown or technical report and the engine uses it to strengthen the claim chart.
A record of what was considered. See every product the engine looked at, which ones passed, which ones were dropped, and the reason in each case.
Version control on products. Move a candidate to the latest model, or to a specific model you already know is relevant, so the chart lines up with the right product.
Targeting by company size. Choose whether to surface products from large, mid-size, small, or niche companies.
Export and sharing. Export the full Hit Matrix to Excel, or send results to a colleague from inside the platform.

The questions in San Diego

ClaimHit chose a hard place for its first soft launch. At IPBC Global in San Diego, the industry’s main licensing conference, it did not run a scripted demo. It let people type in their own patents. More than twenty patent owners and licensing executives did, using patents they had known for years, and watched the search run.

The questions were direct. The first one, nearly every time, was whether this is just an AI wrapper around ChatGPT or Claude.

“It is a fair question,” said Bikram Singh, founder of ClaimHit. “A wrapper hands your question to a chatbot and gives you back whatever it says. If you ask a general model to find products close to a claim, it will answer with confidence, and some of what it gives you will not be real. Citations that do not exist. Products that were never made. And it cannot tell you which parts it found and which parts it made up. The worst, the list is not comprehensive.

We built ours to work the opposite way. It finds real products first through six different channels and LLMs is just one, checks how closely each one lines up with the actual claim, and only then puts it in front of you, with the evidence attached. The model is one part of a larger system that checks its work.”

One executive had watched a different tool invent a citation the week before. He asked how he was supposed to trust what he saw on the screen.

The answer was to show him what most tools keep hidden. ClaimHit’s new record of the search lists every product the engine considered, the ones it kept, the ones it dropped, and why in each case.

“The models are good. We use them,” Singh said. “But a raw model gives you an answer that looks right. We wanted an answer you can check, down to the products we rejected and the reason we rejected them. In a negotiation or in court, an invented citation is not a small mistake. It costs you your credibility. We built the whole system so that cannot happen.”

What happened when people ran their own patents

Two things kept happening. Some people saw the same companies their own lawyers had already found, but in minutes rather than weeks. Others saw a company they had never connected to that patent before.

Many of the executives who tried it in San Diego are now in trials. For a tool being judged by people who know their patents better than anyone, that is the verdict that counts.

“There is nowhere to hide in a live demo,” Singh said. “Someone types in a patent they have lived with for years, and they know in seconds whether what comes back is real. Seeing that happen more than twenty times on patents I had never seen, told me more than any test we could run ourselves.”

Who it is for

Because the work now takes minutes instead of weeks, V3 is useful to more than just large licensing operations.

Companies with large portfolios, who need to know which patents have real traction before spending on enforcement, and which ones they might sell.
Brokers, funders, and advisors, who assess portfolios for a living.
Small and mid-size companies, who own good patents but cannot spare the time or budget for a full screening.
Individual inventors, who until now could rarely afford even to find out whether anyone was using their invention.

A large portfolio, a thin budget, a short timeline. ClaimHit is meant to handle all three.

Availability

V3 is open by invitation, in a limited beta. Apply here

About ClaimHit

ClaimHit LLC is a patent licensing intelligence platform. It finds the products and companies that come close to a patent, across single patents and whole portfolios, and builds claim charts against those products and against technical standards. Its results are research meant to support the work of patent owners, licensing teams, and their counsel, not to replace professional legal judgment. ClaimHit LLC is registered in Wyoming, USA.

Media contact: [email protected]
Web: claimhit.com
Address: 1023 E Lincolnway, Cheyenne, WY, 82001