We make the world's laws free and understandable.
Explore the risks and possibilities with a prompt for ChatGPT, Claude, or your agent.
Casetext began in 2013 as an attempt to make American law free and intelligible, first through crowdsourced annotations and later through machine-assisted research. Over ten years it moved from a “Wikipedia meets Reddit” legal library to CARA, neural search, automated brief drafting, and finally CoCounsel, a GPT-4-powered assistant for research, document review, deposition preparation, and contract analysis.[1]
This was not a failed startup. It was a patient technology bet that became strategically valuable when large language models crossed a capability threshold. The same history also exposes the limit of an independent legal-research challenger: Casetext could build the interface and AI, but not cheaply reproduce Westlaw's authoritative content, citator, and installed distribution. Thomson Reuters paid $650 million in cash in 2023, then absorbed the product into CoCounsel and its broader professional platform.[2]
Jake Heller combined an unusual pair of skills. He was a lifelong coder and a Stanford Law graduate who had served as president of the Stanford Law Review, clerked for First Circuit Judge Michael Boudin, and worked as a litigator at Ropes & Gray.[3] Joanna Huey, the other co-founder named in Y Combinator's 2013 launch post, had studied physics, led the Harvard Law Review, and clerked alongside Heller for Judge Boudin.[4] The gap between consumer software and the expensive, closed tools used by lawyers bothered Heller for years. He quit his firm, applied to Y Combinator, and coded the first version himself.[5]
The first product let users search federal cases and add public annotations. Heller described the thesis plainly: “Casetext uses crowdsourcing and data science to make the law free and understandable.”[5] Laura Safdie, a former Simpson Thacher litigator who had known Heller since high school, later joined as co-founder, COO, and general counsel. Pablo Arredondo, a former Quinn Emanuel and Kirkland & Ellis litigator and Stanford CodeX fellow, joined with a parallel conviction that lawyers deserved better tools. Arredondo recalled comparing the technology used by clients such as Apple with the software lawyers used to represent them: “I hated basically all of the technology we used to litigate.”[6] Sources differ on the legal founder roster over time; they agree that Heller, Huey, Safdie, and Arredondo each shaped the company during its founding period.
The founding vision contained both the company's mission and its first wrong turn. Lawyers wanted affordable access and better context, but most would not volunteer enough original commentary to annotate the common law. Arredondo said the team tried a specialized writing platform and still failed to generate the needed contribution volume. It found a workaround by importing client alerts that law firms already published for marketing, partnering with hundreds of firms and nonprofits, and mining those documents for case references.[6]
That pivot mattered. Casetext stopped waiting for a community to create a proprietary knowledge layer and began extracting structure from documents lawyers already produced. The same logic led to CARA. Arredondo described the insight as mining the information encoded in a brief or pleading to create a new form of legal research. Friends built a crude proof of concept within weeks; years of work by lawyers, data scientists, designers, and engineers turned it into a product.[6]
Casetext's product changed shape several times, but each version tried to reduce the cost of finding and applying law. The original service offered free case law and invited lawyers to publish annotations. When organic contributions proved insufficient, the company indexed law-firm client alerts and extracted their cited cases. It also harvested more than two million “explanatory parentheticals,” short judicial descriptions of holdings, directly from opinions.[6]
CARA changed the input. Instead of asking a lawyer to guess the right keywords, it let the lawyer upload a brief, complaint, or pleading. The system read the document's facts, issues, and citations, then returned relevant authorities that the document did not cite. This fit the actual workflow: a lawyer usually starts with a draft or litigation record, not an abstract search string. By 2018, roughly 300 law firms subscribed, including O'Melveny, Quinn Emanuel, DLA Piper, and Fenwick & West.[13]
Casetext kept adding structure. Holdings surfaced concise case summaries, Black Letter Law located foundational propositions, SmartCite competed with Shepard's and KeyCite, and Compose drafted portions of litigation briefs. AllSearch used neural retrieval to search private litigation documents without exact keyword matches. These were not disconnected features. They trained the company to turn a lawyer's messy work product into a bounded task with inspectable sources.
CoCounsel generalized that pattern. A lawyer could request a research memo, upload documents for review, prepare deposition questions, summarize records, draft correspondence, or analyze contracts. The product wrapped GPT-4 in task-specific instructions, validation, and a professional interface. Casetext had worked with OpenAI before the public GPT-4 release, and DLA Piper tested CoCounsel from September 2022.[9]
The important product decision was not merely adding chat. Casetext defined discrete legal jobs, tested them with lawyers, and connected outputs to source material. That workflow discipline survived the acquisition. Thomson Reuters later integrated CoCounsel with Westlaw, Practical Law, Microsoft 365, and document-management systems, turning a startup application into an interface across its professional content estate.[14]
Casetext deliberately spanned the market. Free primary law and basic search supported students, judges, and cost-sensitive lawyers. Paid CARA subscriptions served solo practitioners and small firms that could not justify incumbent prices, while enterprise sales targeted Am Law firms. In 2019, its small-law plan began at $65 per month for the first lawyer and included CARA, SmartCite, and federal and state materials.[15] CoCounsel moved upmarket and initially offered an unlimited individual subscription at $500 per month, according to a solo appellate lawyer who used it for three months.[16]
The original investor pitch estimated an $8 billion legal-research market dominated by Westlaw and LexisNexis.[17] That figure was promotional and should not be mistaken for audited revenue. The acquisition supplies the harder signal: Thomson Reuters paid $650 million for a company that Axios reported had raised roughly $68 million and was valued at $125 million after its early-2022 financing.[18] The premium priced speed, talent, and an early working product during the first enterprise rush into generative AI.
Casetext first competed against Westlaw and LexisNexis on price and usability, then against newer research systems such as Fastcase, Ravel, ROSS, and Judicata on AI-assisted discovery. The incumbents possessed an advantage that interface design could not erase: licensed editorial content, citation histories, taxonomies, customer contracts, and decades of trust. A 2017 Hacker News discussion quoted research showing Westlaw and Lexis returned more relevant results than a cluster of newer providers, illustrating the content-and-structure gap even before generative AI.[19]
CoCounsel's direct rivals included Harvey and, soon after launch, generative products from Westlaw and LexisNexis themselves. Casetext's temporary edge came from early GPT-4 access plus a decade of legal workflow engineering. It did not own the foundation model, and it did not own the deepest legal corpus. That made it valuable to an incumbent and vulnerable as a standalone platform at the same time.
Casetext used freemium access as both mission and distribution. Primary law remained free, while the company charged for advanced tools that saved professional time. CARA and the full research suite supported individual and firm subscriptions; enterprise customers added sales and implementation work. CoCounsel raised willingness to pay because it addressed work that could consume hours or days, not just search queries.
Public revenue was not disclosed before the acquisition. A 2024 interview reported that the $650 million price was around twenty times revenue, implying annual revenue near $32.5 million, but that is a journalistic estimate rather than a filed number.[20] Axios reported approximately $68 million in total funding. Those figures suggest a capital-intensive decade followed by a high-multiple strategic exit, not a conventional SaaS sale based on mature earnings.
The model's tension was durable. Free law widened reach, but premium research had to fund content processing, legal experts, data science, model costs, security, and enterprise sales. Thomson Reuters could spread those costs across a large installed base and combine CoCounsel with products customers already bought.
The product produced credible adoption at each stage. In 2015 Casetext reported 250,000 monthly visitors, 35,000 users, and more than 100,000 posts.[7] CARA had about 300 subscribing law firms by 2018.[13] Thomson Reuters said the acquired business served more than 10,000 law firms and corporate legal departments in August 2023.[2]
Post-acquisition scale dwarfed the startup. In November 2023 Thomson Reuters said 1,400 customers had contributed more than 4,000 hours of CoCounsel testing and training.[21] By February 2026 it claimed one million professionals in 107 countries and territories had chosen CoCounsel, now spanning legal, tax, audit, accounting, compliance, and trade workflows.[12]
Casetext's first mechanism did not work. Lawyers would consume free law, but most would not create enough public annotations to form the expert layer the product needed. The team tried dedicated publishing tools, then changed the supply model by indexing analysis that firms had already written. That was the decisive early correction: preserve the mission while abandoning the behavior assumption.
The lesson is sharper than “listen to users.” Casetext recognized that professional reputation already motivated lawyers to publish, but on their own firms' sites. It moved distribution to the existing incentive instead of demanding a new one. That generated the structured material required for better research and freed the company to build CARA.
Heller said Casetext had explored language models for search from 2018, but “for a long time the tech was not where we needed. But we stuck it out.” When model capability improved abruptly, the company had legal data pipelines, task design, enterprise relationships, and lawyers who could test outputs.[20] Early GPT-4 access mattered because ten years of preparation let Casetext turn a research model into a product before the market caught up.
This argument has a counterfactual. Without the GPT-4 jump, Casetext might have remained a smaller alternative-research vendor after years of venture spending. Heller acknowledged timing and luck. The company “made our own luck,” he said, but benefited from managing partners suddenly demanding an AI strategy after ChatGPT.[20]
The acquisition interview contains the clearest diagnosis. Heller called content Casetext's “number one constraint” and said putting West's database behind CoCounsel would make legal research dramatically better.[3] This was not surrender. It was recognition that high-stakes professional AI requires both task intelligence and authoritative substrate.
The strongest alternative is that Casetext should have raised another round and remained independent. Heller said that path was plausible. Yet Thomson Reuters could reproduce a generic assistant more easily than Casetext could reproduce Westlaw, Practical Law, KeyCite, global sales, and existing workflow integrations. Selling converted a temporary model-access advantage into ownership of the combined platform.
Thomson Reuters did not preserve Casetext as a parallel research company. It migrated CoCounsel accounts to Thomson Reuters infrastructure, connected the assistant to Westlaw and Practical Law, and retired the original casetext.com experience.[11] The loss of chats and uploaded chat documents during migration shows the customer cost of platform consolidation, even as databases could move.
The outcome still validates the product thesis. CoCounsel became the name for Thomson Reuters' AI layer across several regulated professions. The startup's narrow assistant was absorbed into a much broader system with proprietary content, validation by thousands of subject-matter experts, and multi-model governance.[12] Casetext did not beat the incumbent. It supplied the interface through which the incumbent planned to defend itself.