Il prossimoProssima guida
Encoder incrociati vs Bi-Encoder
Intelligenza artificiale linguistica
GUIDA alle aziende
Lexis+ AI from LexisNexis and Westlaw's AI features from Thomson Reuters, including AI-Assisted Research and CoCounsel, add generative AI to the two largest US legal research databases.
Both use retrieval-augmented generation to base answers on their own cases, statutes and editorial content, with linked citations. The comparison matters because independent testing has found both tools can be wrong, so choosing between them depends on how each grounds its answers and how well it performs on your own test questions.
Lexis+ AI from LexisNexis and the AI features in Thomson Reuters' Westlaw add generative AI to the two dominant US legal research databases. LexisNexis launched Lexis+ AI commercially in late 2023 and later added a personalized assistant called Protégé. Thomson Reuters introduced AI-Assisted Research in Westlaw Precision in late 2023. The same year it acquired Casetext and its CoCounsel assistant, which it then folded into its product line. Both rely on retrieval-augmented generation (RAG). Instead of answering from what a language model memorized, the system searches the vendor's own collection of cases, statutes, regulations and secondary sources. It passes the passages it finds to the model and asks it to write an answer that cites them. Each vendor also draws on its own editorial assets: Westlaw: its headnotes, the West Key Number System and KeyCite, plus practice content in Practical Law; and Lexis: its headnotes and Shepard's, plus practice content in Lexis Practical Guidance. Citations link back to the source documents, and citator signals help show whether an authority is still good law. Grounding reduces made-up answers but does not eliminate them. A 2024 study by Stanford researchers found that both Lexis+ AI and Westlaw's AI-Assisted Research gave incorrect or misgrounded answers on a meaningful share of test questions. Westlaw's tool erred more often in that test, and the vendors disputed parts of the method. The subtler failure is misgrounding: citing a real case for a point it does not support. The practical conclusion is to choose between them based on your own testing, not marketing. Build a set of questions from your practice area whose answers you already know. Include: jurisdiction-specific questions; recent changes in the law; questions built on a false premise; and issues where the controlling authority is obscure. Then compare accuracy, whether each tool finds the controlling authority, how easily you can check its citations, and how it handles uncertainty.
Le roadmap dei fornitori influenzano le funzionalità che il tuo team può sviluppare successivamente.
I termini commerciali e le opzioni di implementazione influiscono sui costi e sui rischi a lungo termine.
Gli incentivi aziendali modellano le impostazioni predefinite dei prodotti, la postura di sicurezza e l’apertura.
Both companies are moving toward multi-step, agent-style research that plans queries, reads many documents and drafts memos. They are also linking research more closely with document drafting and firms' internal knowledge. These features make checking more important, not less, because longer outputs contain more claims to verify. Independent benchmarks for legal research AI are still developing, so firms will keep relying on their own test sets. Pricing and bundling are likely to shape adoption as much as quality does. The market may also shift as smaller legal AI companies and general-purpose assistants add research features.
A litigation group runs the same 30 jurisdiction-specific research questions, with answers the lawyers already know, through both platforms and scores accuracy and whether each tool found the controlling authority.
A lawyer asks each tool a question built on a false premise, such as a statute that was repealed, to see whether it corrects the premise or answers as if it were true.
An associate uses a generated research memo only as a starting point. She opens every cited case, reads the relevant passage and checks its citator flag before relying on it.
A law firm librarian compares how easily each platform shows the sources behind an answer, which affects how long lawyers spend checking the output.
Gli annunci di lancio potrebbero superare la stabilità nei flussi di lavoro di produzione reali.
I prezzi delle API o i cambiamenti politici possono infrangere le ipotesi da un giorno all’altro.
La dipendenza da un unico fornitore aumenta i costi di lock-in e di migrazione.
Valuta i fornitori utilizzando le tue attività e i tuoi set di dati.
Esamina la privacy, la sicurezza e i termini legali prima dell'integrazione.
Mantenere un piano di riserva tra modelli o fornitori.
Monitora le note di rilascio in modo che le modifiche alla roadmap non sorprendano i team.
Free newsletter
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Lexis+ AI from LexisNexis and Westlaw's AI features from Thomson Reuters, including AI-Assisted Research and CoCounsel, add generative AI to the two largest US legal research databases. Both use retrieval-augmented generation to base answers on their own cases, statutes and editorial content, with linked citations. The comparison matters because independent testing has found both tools can be wrong, so choosing between them depends on how each grounds its answers and how well it performs on your own test questions.
Entrambi i sistemi ricercano i propri casi, statuti e fonti secondarie, quindi fanno scrivere a un modello linguistico una risposta che citi ciò che è stato trovato.
Il fraintendimento è più sottile che inventare un caso. L'autorità è reale, ma l'affermazione ad essa allegata è sbagliata.
Lo studio ha riscontrato errori in entrambi i prodotti, più spesso per lo strumento di Westlaw in quel test. I venditori hanno contestato parti del metodo.
KeyCite e il Key Number System sono risorse di Thomson Reuters. Le note di testa di Shepard, Lexis e la Guida pratica appartengono a LexisNexis.
Un buon strumento di ricerca dovrebbe individuare e correggere una premessa falsa, come una legge abrogata, invece di costruire una risposta su di essa.
Continua a imparare
Altre guide selezionate per questo argomento
Il prossimoProssima guida
Encoder incrociati vs Bi-Encoder
Intelligenza artificiale linguistica