AI Legal Research Assistants Compared: CoCounsel vs Harvey vs Lexis+ AI (2026)
Legal research used to mean hours in Westlaw or LexisNexis, running Boolean searches and reading headnotes. AI research assistants promise to compress that into minutes by understanding natural language questions and returning relevant case law with analysis. But which one actually finds the cases you need?
I tested CoCounsel (Thomson Reuters), Harvey AI, and Lexis+ AI (LexisNexis) on the same research question to compare their output directly.
The Test: Real Legal Research on a Real Problem
I gave each tool the same research prompt:
“Find cases in the Second Circuit from the last 5 years addressing whether a software license constitutes a ‘good’ under the UCC or a ‘service’ under common law, specifically in the context of breach of warranty claims.”
This is a genuinely difficult question with split authority, nuanced holdings, and recent developments. Perfect for testing AI research capabilities.
CoCounsel (Thomson Reuters)
CoCounsel is Thomson Reuters’ AI legal assistant, built on GPT-4 and integrated with Westlaw’s database of case law, statutes, and secondary sources.
How It Works
You type a research question in plain English. CoCounsel interprets the question, searches Westlaw’s database, and returns a memo with case citations, holdings, and analysis. It also provides the full text of cited cases so you can verify the citations.
Results on Our Test
CoCounsel returned 12 relevant cases within 3 minutes. It correctly identified the key split between the “predominant purpose” test and the “gravamen of the action” test. It cited Advent Systems Ltd. v. Unisys Corp. (3d Cir. 1991) as the foundational case and found 4 Second Circuit cases from 2021-2025 that addressed the issue.
Accuracy: 10 of 12 citations were directly on point. The other 2 were relevant but less directly applicable. No hallucinated citations.
Analysis quality: The memo correctly identified the trend toward treating software licenses as goods when the software is “off the shelf” versus services when customization is involved. It noted a 2023 SDNY decision that both sides would want to cite.
Pricing
Included with Westlaw Edge subscriptions. Standalone pricing starts at $100/month per user for firms already on Westlaw. If you’re not on Westlaw, this isn’t a standalone option.
Strengths
- Access to Westlaw’s complete database, which is the gold standard for case law
- Citations are real and verifiable (because they come from Westlaw’s index)
- Integrates into existing research workflow
- Good at finding recent cases that haven’t made it into treatises yet
Weaknesses
- Requires a Westlaw subscription, which most firms already have but is expensive
- Analysis is solid but sometimes misses the practical implications courts are wrestling with
- Slower than Harvey for complex, multi-issue research questions
Harvey AI
Harvey is the AI platform built specifically for lawyers, trained on legal data and designed to handle research, drafting, and analysis.
How It Works
Harvey uses a chat interface where you ask research questions. It searches across case law, statutes, and secondary sources, then returns a structured analysis with citations. Harvey also lets you upload case files and ask questions about specific documents.
Results on Our Test
Harvey returned 15 cases in about 2 minutes. It correctly identified the same split in authority as CoCounsel and cited many of the same foundational cases. However, Harvey also pulled in 3 cases from outside the Second Circuit that were persuasive authority, which CoCounsel missed.
Accuracy: 11 of 15 citations were directly on point. Two citations were to cases that existed but didn’t address the UCC question as directly as Harvey suggested. Two citations were hallucinated: real case names that, when I checked, didn’t contain the holdings Harvey attributed to them.
Analysis quality: Harvey’s analysis was more detailed than CoCounsel’s and included a section on “practical implications for litigation strategy” that was genuinely useful. But the hallucinated citations are a serious problem for research that will end up in a brief.
Pricing
Custom pricing, typically $50-100/user/month for firms. Volume discounts available for larger teams.
Strengths
- Fast, detailed analysis with practical litigation angles
- Good at synthesizing multiple cases into a coherent narrative
- Handles complex, multi-issue questions better than CoCounsel
- Chat interface makes it easy to ask follow-up questions
Weaknesses
- Hallucinated citations are a real risk. Every citation needs verification.
- Not integrated with Westlaw or LexisNexis databases, so the case law index may be less complete
- Newer platform, still building out features
Lexis+ AI (LexisNexis)
Lexis+ AI is LexisNexis’ answer to CoCounsel, integrating AI capabilities into their existing research platform.
How It Works
Similar to CoCounsel: you ask a research question and get back a memo with citations and analysis. The difference is the underlying database: Lexis has its own case law index, Shepard’s Citations for validation, and different secondary sources than Westlaw.
Results on Our Test
Lexis+ AI returned 11 cases in about 4 minutes. The citations were accurate and well-organized. It correctly identified the key legal tests and found the same foundational cases as the other tools.
Accuracy: 10 of 11 citations were directly on point. One was relevant but slightly off-topic. No hallucinated citations. This is the most reliable citation accuracy of the three tools.
Analysis quality: Solid but conservative. Lexis+ AI provided accurate case summaries but was less willing to synthesize a narrative or predict how a court might rule. It presented the law as it is without much analysis of where it’s going.
Pricing
Included with Lexis+ subscriptions. Standalone AI access starts at $65/month per user. Like Westlaw, you need the underlying Lexis subscription.
Strengths
- Most accurate citations of the three tools (benefiting from Lexis’s Shepard’s Citations system)
- No hallucinated citations in my testing
- Integrates with existing Lexis research workflow
- Good at finding cases across jurisdictions
Weaknesses
- Analysis is less detailed and less willing to make predictive judgments
- Slower than Harvey and comparable to CoCounsel
- Requires Lexis subscription, which competes with Westlaw (most firms pick one)
Head-to-Head Comparison
| Feature | CoCounsel | Harvey AI | Lexis+ AI |
|---|---|---|---|
| Citation accuracy | Very good | Good (verify everything) | Excellent |
| Analysis depth | Good | Excellent | Good |
| Speed | 3 min | 2 min | 4 min |
| Hallucination risk | Low | Medium | Very low |
| Database access | Westlaw | Independent | Lexis |
| Price | Westlaw subscription | $50-100/mo | Lexis subscription |
| Best for | Westlaw users | Complex analysis | Citation-critical work |
Which Tool Should You Use?
Use CoCounsel if you’re already on Westlaw and want AI integrated into your existing workflow. The citation accuracy is good, and the analysis is reliable enough for most research tasks.
Use Harvey if you need deep analysis and are willing to verify every citation. Harvey’s synthesis is the best of the three, but the hallucination risk means you can’t trust the citations without checking them. Good for strategy memos and internal research, risky for court filings without verification.
Use Lexis+ AI if citation accuracy is paramount. For briefs, motions, and anything filed with a court, Lexis+ AI’s citation accuracy is the most reliable. The analysis is less detailed, but you won’t embarrass yourself with a hallucinated case citation.
Use multiple tools if the research question is high-stakes. For a case worth $500,000 or more, run the question through both CoCounsel and Lexis+ AI. The overlap confirms the key cases, and the unique finds from each tool fill in gaps the other missed.
What AI Legal Research Still Can’t Do
- Predict judicial outcomes: AI can find cases and summarize holdings, but predicting how a specific judge will rule on a novel issue requires courtroom experience and local knowledge
- Replace oral argument prep: AI doesn’t know your judge’s pet peeves or the opposing counsel’s tendencies
- Handle truly novel issues: When there’s no on-point case law, AI research tools struggle. They’re great at finding existing authority, less useful for arguing for new law
- Verify its own citations: Always check. Every time. No exceptions.
Related reading: AI Legal Research for Lawyers · ChatGPT Prompts for Lawyers · Westlaw vs LexisNexis vs CoCounsel · Best AI for Case Law Research 2026 · Local AI for Lawyers: Private, Free · AI Legal Research Prompts for Lawyers
Not a lawyer? See AI tools for teachers or best AI tools by profession.
FAQ
Can I cite AI-generated research in a court filing?
You can cite the cases AI finds, but you must verify every citation independently. Multiple courts have sanctioned attorneys for filing briefs with hallucinated case citations found by AI. The AI is a research tool, not a citation source. Pull up each case, confirm it exists, confirm the holding matches what AI reported, and Shepardize or KeyCite it before citing.
Is AI legal research ethical under bar association rules?
Most state bar associations have issued guidance saying AI use is permitted but lawyers must supervise the output and verify accuracy. The ABA’s 2024 formal opinion states that lawyers may use AI tools but must maintain competence over the work product and cannot delegate professional judgment to technology.
How much time does AI legal research actually save?
For straightforward research questions (find cases supporting a known legal position), AI saves 1-2 hours per question. For complex research (multi-jurisdictional, novel issues), the time savings are smaller because verification takes longer. Firms using AI research tools report saving 5-8 hours per attorney per week on research tasks.
Do I need both Westlaw and LexisNexis if I use AI research tools?
Most firms subscribe to one platform. AI tools reduce the advantage of having both because they find the same core cases. However, for high-stakes research, running questions through both CoCounsel (Westlaw) and Lexis+ AI (LexisNexis) catches cases the other’s database might miss. For routine research, one platform plus AI is sufficient.
What’s the biggest risk of using AI for legal research?
Hallucinated citations. AI tools can generate case names, citations, and holdings that look real but don’t exist. This has led to court sanctions, malpractice claims, and public embarrassment. The mitigation is simple: verify every citation. But it requires discipline, especially when the AI output looks confident and authoritative.