What is the Patent Search tool, and what can it do?

TCP Law's free Patent Search tool offers two search modes: enter a patent or publication number to retrieve a specific patent document, or enter a brief description of your invention to perform an AI-based semantic search. It is designed as an accessible starting point for inventors exploring the patent landscape before investing in development or filing. Use it alongside TCP Law's patent search services for a complete prior art assessment.

Why is conducting a patent search essential before developing or filing for my invention?

Before investing significant time and resources into developing your invention, a patent search helps determine whether your idea is novel and non-obvious by uncovering existing prior art. Discovering conflicts early, before significant investment, allows you to refine your concept or adjust your claims strategy. Discovering the same prior art after filing is far more costly. Book a free consultation to discuss your invention.

What is prior art, and why does it matter for my patent application?

Prior art refers to any existing patents, published applications, or other publicly available materials disclosing elements similar or identical to your invention. It is the primary basis on which the USPTO evaluates novelty and non-obviousness. A single reference disclosing every element of your invention can prevent grant. Understanding prior art before filing allows the application to be drafted to distinguish your invention. Use the free Patent Search tool as a first step.

What is the difference between a preliminary and a comprehensive patent search?

A preliminary search finds patents identical or very similar to your core concept, giving a fast initial read on obvious conflicts. A comprehensive patentability search covers a wider range of prior art that could prevent grant or limit claim scope, including secondary references that may be combined to form an obviousness rejection. Use the Patent Search tool for preliminary screening, then contact TCP Law for a comprehensive professional search.

How does the AI-based semantic search mode work?

The semantic search mode analyzes the substance of your invention description, including its functional and structural characteristics, rather than relying on exact keyword matches. This surfaces relevant prior art even when different terminology is used to describe similar technology. Traditional keyword searches miss this. Enter a brief plain-language description of your invention in the Patent Search tool and review results before your patent consultation with TCP Law.

Can I use the tool to look up a specific patent I've heard about?

Yes. One of the two search modes allows you to enter a patent or publication number directly to retrieve the corresponding document. This is useful for researching a competitor's patent, evaluating a patent asserted against you in a cease and desist letter, or reviewing a specific reference cited in a prior art search. View patent samples to familiarize yourself with patent document structure before searching. Contact TCP Law to interpret results professionally.

Does using the Patent Search tool protect my invention or establish any legal rights?

No. The tool is for informational purposes only, does not constitute legal advice, and does not create an attorney-client relationship. Using the tool does not file a patent application, establish a priority date, or create any legal rights. Only a properly filed provisional or non-provisional patent application establishes priority. Book a free consultation to take the next step toward protecting your invention.

What should I do after running a patent search with the tool?

After reviewing results, consult a patent attorney who can interpret findings in the context of your invention and filing goals. A clean result means no obvious conflict was surfaced, not that your invention is definitely patentable. Closely related results provide strategic drafting information. In either case, a professional patent search and attorney consultation are the appropriate next steps. Contact TCP Law to discuss your results.

How can patent search results strengthen my application before filing?

Search results inform strategic application drafting. By understanding how your invention compares to identified prior art, the application can include additional technical features, alternative embodiments, and claim limitations that distinguish your invention from known technology. Results also help anticipate USPTO rejection types before they arise. Use the Patent Search tool as a starting point, then contact TCP Law for professional search and drafting services.

Is it a myth that searching only U.S. patents is sufficient prior art due diligence?

Yes. Prior art includes any public disclosure worldwide, including foreign patents, PCT publications, academic papers, conference proceedings, and product documentation. A U.S.-only search leaves a significant gap. TCP Law's AI platform supports semantic searches that surface relevant prior art across broader sources, beyond what keyword-only tools find. Use the free Patent Search tool as a first step, then book a consultation for a complete professional assessment.

Third Circuit Affirms in Thomson Reuters v. ROSS: Copying Westlaw Headnotes to Train a Competing AI Tool Is Not Fair Use

laurencejohn9

19 hours ago

7 min read

On September 29, 2026, the U.S. Court of Appeals for the Third Circuit handed Thomson Reuters a full win in its long-running copyright suit against ROSS Intelligence. The court held that Westlaw's headnotes are protected by copyright, and that ROSS's copying of them to train a competing AI legal research tool was not fair use.

It is the first decision from a federal court of appeals on whether training an AI system on copyrighted material can be fair use, and it is precedential, binding district courts in Delaware, New Jersey, Pennsylvania, and the Virgin Islands. That alone makes it required reading for anyone who builds, buys, or licenses AI tools. But the opinion is also notable for what it declined to decide. The panel took pains to describe the dispute as "no more than an ordinary copyright case" between two competitors, and, in a footnote, it set ROSS's non-generative tool apart from the large language models at the center of the generative AI cases.

Here is what the court held, why, and what it means for businesses on both sides of the AI training question.

The Dispute

ROSS was a legal-tech startup building an AI search tool that would answer natural-language legal questions by returning relevant passages from judicial opinions. It asked Thomson Reuters for a Westlaw license to train the tool, and Thomson Reuters refused, since ROSS meant to compete with Westlaw.

ROSS then turned to a third-party vendor, LegalEase, which produced roughly 25,000 "Bulk Memos" of legal questions paired with sample answers. The memo drafters used thousands of Westlaw headnotes, the short summaries of legal points that Westlaw's editors write for each case, to frame the memos' legal questions. On appeal, ROSS did not dispute that its vendors copied headnotes or that their conduct is attributable to ROSS. ROSS used the memos as training data. Thomson Reuters and West sued in the District of Delaware in 2020, and ROSS has since gone out of business.

In February 2025, Judge Stephanos Bibas (a Third Circuit judge sitting by designation) granted Thomson Reuters partial summary judgment. He found that 2,243 headnotes were original and actually copied, and he rejected ROSS's fair use defense. He then certified both questions, copyrightability and fair use, for interlocutory appeal under 28 U.S.C. § 1292(b), and the Third Circuit agreed to hear them. Judges L. Felipe Restrepo, Tamika R. Montgomery-Reeves, and Emil J. Bove heard argument in June 2026. Judge Montgomery-Reeves wrote for the panel, affirming on both questions.

Headnotes Clear the Originality Bar

The court held that all 2,243 headnotes at issue are original enough to be protected. Under Feist Publications v. Rural Telephone Service, the bar is extremely low, and the panel found it met by the editors' choices: which legal points to pull from an opinion and how to phrase them under Westlaw's drafting guidelines. The court drew support from Callaghan v. Myers (1888), which protected a court reporter's own additions to published opinions, and from Georgia v. Public.Resource.Org (2020).

ROSS's three main counterarguments failed:

A monopoly over the law. Headnotes are not the law. The judicial opinions themselves stay free for anyone to use.
Merger. Merger applies only where an idea can be expressed in very few ways. A legal point in an opinion can be summarized in many ways, so the doctrine did not fit.
Too little originality. The court distinguished the Second Circuit's Matthew Bender v. West (1998), where industry convention dictated the content of West's case-report elements. Headnotes, by contrast, are independently composed.

Two points the court left open are worth noting. It declined to decide whether a headnote that quotes an opinion word for word could be original, because the headnotes on appeal were not verbatim copies. And it did not reach the originality of the West Key Number System, because ROSS did not brief it.

Fair Use: Three of Four Factors Favor Thomson Reuters

The panel walked through all four statutory factors and found that the two that usually matter most, purpose and market effect, both cut against ROSS.

1. Purpose and character (favors Thomson Reuters). Applying Andy Warhol Foundation v. Goldsmith (2023), the use was commercial and "minimally transformative, at best." Training was an intermediate step, but both companies used the headnotes for the same end: a legal research platform. A footnote cited undisputed evidence that ROSS at times acted in bad faith, including using a law-firm investor's credentials and a student account to get into Westlaw.

2. Nature of the work (favors ROSS). The headnotes were published and more factual than creative. The court noted this factor rarely carries much weight.

3. Amount used (favors Thomson Reuters). ROSS stressed that it used only about 0.08% of Westlaw's roughly 28 million headnotes. The court treated each headnote as its own work, so ROSS copied entire works. Copying was also unnecessary, since the underlying opinions were freely available.

4. Market effect (favors Thomson Reuters). ROSS took the very value of the headnotes (finding the right opinions), aimed squarely at Westlaw's research market, and offered no evidence to rebut the harm. The court also recognized a "rapidly developing" market for licensing content as AI training data.

Why Google Books and the software cases did not help ROSS

ROSS leaned on Authors Guild v. Google (2d Cir. 2015). The panel found it inapposite: scanning books to build a search index served a completely different function from reading the books, and could even drive sales. ROSS's tool, by contrast, did what Westlaw headnotes already do, and ROSS admitted it wanted to replace Westlaw rather than send users to it.

The court likewise set aside the intermediate-copying cases, Sega v. Accolade, Sony v. Connectix, and Google v. Oracle. Those defendants copied code to reach unprotected functional elements or achieve interoperability. ROSS had no comparable need, because it could have worked from the public opinions directly. As the court put it, "ease is not a justification for copying."

The licensing market point

For content owners, the most consequential passage may be the court's treatment of the AI-training licensing market. Thomson Reuters had never licensed headnotes for training. The court held that this did not matter: Thomson Reuters trains its own AI tools on its headnotes, the market for training licenses is developing quickly, and unauthorized copying can usurp it. The court also rejected ROSS's arguments about public benefit, harm to the AI industry, and national security, for lack of evidence.

What the Decision Does Not Decide: Generative AI

The panel took care to mark the limits of its reasoning, mostly in a single footnote. ROSS's system was not generative AI. It retrieved and returned existing passages from judicial opinions; in the court's words, it "cannot generate original expression."

That footnote responds to a statement of interest the Department of Justice filed on September 1, 2026, in In re OpenAI in the Southern District of New York. Relying on Bartz v. Anthropic (N.D. Cal. 2025), the DOJ argued there that training a large language model that can generate original responses is transformative and does not produce "substitutive competition." The Third Circuit said those concerns do not apply here: ROSS's AI could not generate original expression, and ROSS trained it to build a commercial substitute for Westlaw. The court also noted that the DOJ chose not to weigh in on this case.

So the Third Circuit has not held that AI training is generally infringing, and it has not endorsed the generative AI rulings either. It decided the narrower question in front of it: copying a competitor's protected content to build a non-generative tool that does the same job is not fair use. How far that reasoning travels, particularly the licensing-market analysis under factor four, will be argued in every training case to come.

Practical Takeaways

For publishers and other content owners:

Editorial work on public-domain material can be protected. Summaries, annotations, and selections that reflect real editorial choices clear the Feist bar, even when the underlying source is free.
You do not need an existing licensing program to claim harm to an AI-training licensing market. Documenting your own AI use of your content, and any licensing inquiries you receive, strengthens that argument.
Each discrete item (a headnote, an abstract, a photo) may count as a whole work. That can defeat a defendant's "we only took a tiny fraction" argument.

For companies building or training AI tools:

Tools that compete with the source of their training data face the steepest fair use climb. The closer your output is to a substitute for the original, the weaker the defense.
Data provenance matters. Know where your training data came from, and get representations and indemnities from vendors who assemble it. ROSS's exposure came through its vendors' work, and on appeal it did not dispute that their copying was attributable to it.
If public-domain source material is available, train on it. The court gave weight to the fact that ROSS could have used the opinions themselves.
How you get access matters. The court counted ROSS's use of borrowed credentials and a student account to reach Westlaw as bad faith.

For everyone: this is one circuit's answer on one set of facts. The generative AI cases remain open, and a split among the circuits is possible.

What Comes Next

Because this was an interlocutory appeal, the case now returns to the District of Delaware. Still open there are Thomson Reuters' tortious interference claim and any headnotes beyond the 2,243 resolved on summary judgment. ROSS has said it will ask the full Third Circuit to rehear the case en banc and, if necessary, petition the Supreme Court.

TCP Law advises clients on copyright protection, licensing, and enforcement, including questions about AI training data. If you own content that may be used to train AI, or you are building tools that rely on third-party data, we would be glad to talk through how this decision affects you.

This article is for general information only and is not legal advice. Reading it does not create an attorney-client relationship.