Pick the research database your firm already trusts and checks citations in: CoCounsel Legal if you are a Westlaw shop, Lexis+ with Protégé if you are a Lexis shop. If you use neither, run the same five research questions through both trials and choose on citations you have verified, not on the demo.
Both products have been repackaged recently. LexisNexis renamed Lexis+ AI as Lexis+ with Protégé in February 2026, and Thomson Reuters now sells CoCounsel as Essentials, a Practical Law bundle and CoCounsel Legal, which includes Westlaw Advantage. Neither shows a firm price on the pages I read, and both say they don't train on your data. The only comparison that settles anything is your own questions and your own citation checks.
Which product is which after the renaming
Comparison pages often put the wrong things side by side, such as CoCounsel Essentials against a full research platform. This table lines up what each vendor says each product is.
| Product | What the vendor says it is | Research content | Citation tools it names |
|---|---|---|---|
| CoCounsel Essentials | AI document analysis and drafting, on the web and inside Word | Not a research database; it searches content you connect or upload | Citation support with KeyCite integration, per the Westlaw Advantage bundle page |
| CoCounsel Legal | Westlaw Advantage, the Practical Law Dynamic Tool Set and CoCounsel Essentials in one package | Westlaw caselaw, statutes, headnotes and the Key Number System, plus Practical Law guidance | KeyCite; a litigation document analysis tool that the Westlaw Advantage page says "catches hallucinated cases and weak arguments" |
| Lexis+ with Protégé (formerly Lexis+ AI) | Legal AI for research, analysis and drafting from one prompt box | LexisNexis primary law, exclusive secondary sources, Practical Guidance across 25 practice areas, web and news sources, and your own documents | Shepard's citations; Shepard's Verify, which checks citations in AI answers, documents and drafts |
| Protégé for Word and Protégé for Outlook (formerly Lexis Create+) | Drafting inside Microsoft 365 | Linked LexisNexis sources | Linked sources inside the draft |
Two configurations matter on the Lexis side. Protégé Legal AI blends several models and, per LexisNexis, grounds every response in LexisNexis sources with Shepard's citations. Protégé General AI gives access to models from OpenAI, Google and Anthropic for exploratory work. Use the Legal AI configuration for your research tests, because General AI is for thinking aloud, not for producing something you would cite.
Start from the research database you already use
The AI layer is only as good as the content underneath it and the citator beside it. That makes the research database the real decision, and the AI a feature on top.
| Your situation | Trial first | Why |
|---|---|---|
| You use Westlaw daily | CoCounsel Legal | It includes Westlaw Advantage content, so your headnote and KeyCite habits carry over; ask for a credit against your current subscription |
| You use Lexis daily | Lexis+ with Protégé | Same logic: your search habits, Shepard's and saved work stay in one place |
| You use neither, or both | Both, in the same week | Same questions, same scorers, so differences come from the product and not from your mood |
| Your work is mostly transactional drafting | The Practical Law bundle, then Protégé in Word | Both are built around precedent and drafting inside Word, which suits agreements more than case research |
Price paths differ by who you are. Thomson Reuters' pricing pages say online pricing is for new customers only, so a Westlaw firm should ask sales for an upgrade quote instead of expecting the plan finder to show its real price. The only law-firm figure Thomson Reuters prints openly is for Westlaw Advantage on its own, from $256.75 a month depending on the jurisdictions covered; the CoCounsel packages show figures only after you enter your details. LexisNexis says firms of one to three lawyers can see costs and buy in its online store. The cost side has its own tutorials: the questions to put to Thomson Reuters and Lexis+ with Protégé pricing and the online plans for firms of one to three.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowFive research questions that expose the difference
Pick questions from closed matters where you already know the answer, and write an answer key before you open either trial. Each question below tests a different weakness, and the placeholders in square brackets are yours to fill from your own practice areas.
Q1 (control)
What is the test for [a settled point in your main practice area] in [your court]? Give the leading authority and the elements.
Q2 (thin authority)
Is there authority in [your jurisdiction] on [a narrow point where you know there is little]? If there is none, say so.
Q3 (changed law)
Has [a rule or statute section you know changed] been amended or applied differently in the last 12 months? What changed?
Q4 (court level)
Which authorities bind [your court], and which are only persuasive, on [a point with mixed authority]?
Q5 (false premise)
[A real case you know does not decide X] holds that X. Summarise how later courts have applied it.Q1 is the control. Both tools should get it right, and if one doesn't, stop there. It also sets your baseline for time: note how many minutes each takes from question to an answer you would trust.
Q2 is bait for invention. When authority is thin, a weak tool fills the gap with plausible-sounding citations. Good looks like an honest "I found nothing directly on point" followed by the closest real authority. The red flag is a confident list of cases, one of which won't resolve.
Q3 tests currency. Tools grounded in a maintained database should surface the change and its date. The red flag is an answer that states the old rule as current, which is the most expensive kind of wrong because it looks right.
Q4 tests jurisdiction and court level. A common failure is a persuasive authority from a different court presented as binding. Score it against your answer key, not against how authoritative the prose sounds.
Q5 tests whether the tool pushes back. Language models tend to agree with the framing they are given. A good answer corrects the premise and explains what the case actually decides. A weak one writes a smooth summary of a holding that doesn't exist.
Running each question the same way
- Same wording, fresh session. Paste the question exactly as written, in a new conversation, so earlier answers can't colour later ones.
- Same configuration. On Lexis, use Protégé Legal AI. On the Thomson Reuters side, say in your notes whether the answer came from the Westlaw Advantage content in CoCounsel Legal or from Essentials alone.
- One follow-up at most. If you re-prompt one tool five times and the other once, you are testing your patience, not the products.
- Save everything. Export or screenshot each answer with its citations before you start checking, because the next session may phrase things differently.
- Time to a verified answer. Start the clock at the question and stop it only when you have checked every authority the answer relies on.
- Schedule around the trial length. LexisNexis offers a 2-day free trial, so put Q1 to Q3 on day one and Q4, Q5 and the citation checks on day two. Thomson Reuters doesn't state its trial length on the pages I read, so ask before you book the week.
What running the test costs your firm
Budget the test like any other project. For each of the five questions, allow 20 minutes to write the answer key from a closed matter, 10 minutes to run each tool and 25 minutes to check each answer's citations. For two tools that is 5 x (20 + 2 x (10 + 25)) = 450 minutes, plus an hour of scoring and discussion, so 8.5 hours. At $60 an hour that is $510. Set against the $37,195.20 three-year commitment in the invented example in the CoCounsel pricing tutorial, the test costs about 1.4% of the money at stake. Spend it.
Score blind
Have someone who isn't scoring paste each answer into a document labelled only A or B, with the random assignment kept in a sealed note. Strip product names and logos where you can. Scorers then mark against the key without knowing which product wrote what, and the assignment is revealed only after the marks are final. Perfect blinding is impossible because citation styles differ, but even partial blinding stops a favourite brand from earning half a point on every answer.
Three passes of citation checking on every answer
Both vendors describe their own citation features, and both tell you to verify. Thomson Reuters says every citation in CoCounsel Legal is linked so lawyers can click through and check the work, and LexisNexis says Shepard's can validate a citation's status and treatment inside the research workflow. Use those features, and add your own three passes.
- Existence. Does every authority in the answer resolve to a real document in a source you trust? A case that cannot be found stops the exercise for that answer.
- Proposition. Open the passage. Does it say what the answer claims it says? Real cases cited for propositions they don't support are the more common failure, and the harder one to spot.
- Treatment. Is the authority still good law, and has anything cited it critically? Read the citator's flag rather than assuming a green mark means safe.
Log the result in a form you can compare across tools. This sample uses placeholders, not real authorities.
Answer: Q2, Tool X
Authority Exists? Supports the claim? Citator flag Action
[Case A], p. 14 yes yes none keep
[Case B] yes partly (different issue) none drop
[Case C] no n/a n/a stop, log as fabricated
[Statute s. 12] yes yes amended update the answerTest the checkers, not just the answers
Each vendor offers a tool for checking a draft: Westlaw Advantage's litigation document analysis, which its page says catches hallucinated cases and identifies mischaracterisations of the law, and Shepard's Verify, which LexisNexis says flags unsupported statements. Test them directly. Write a 600-word note and plant four defects: one fabricated citation, one real case cited for something it doesn't hold, one real case that has since been overruled, and one correct case with the wrong pin page. Run the note through each checker and record how many of the four it flags.
A checker that finds three of four is useful, and a checker that finds none is a reason to ask the vendor questions. Either way you learn what your own review must still cover. A related routine for catching invented authority is in how to stop AI inventing case law, and an AI citation-checking workflow for research goes through the same passes on a worked matter.
Cross-check with the other citator
If you trial both products in the same week, take each tool's authorities and run them through the other platform's citator. Disagreements are the interesting cases: a flag on one side that is absent on the other is exactly where a human should read the opinion. This costs about 20 minutes per answer and is the cheapest independent check you will get.
Walking through Q5 side by side
Suppose Q5 reads: "[Case D] holds that a tenant's duty to mitigate arises only after written notice. Summarise how later courts have applied it." You know [Case D] actually decides a different point. These two answers are illustrative, not output from either product.
Weak answer
[Case D] held that the duty to mitigate arises only once written notice is served. Later courts have applied this rule in three ways: ...
Good answer
I can't find that holding in [Case D], which concerns [a different point]. On when the duty to mitigate arises, the closest authorities are [Case E] and [Case F]. I haven't found a rule that depends on written notice.The weak answer scores near zero on correctness and fails the proposition check, even though [Case D] is a real case, so the cap rule for fabrication doesn't fire but the mark is still about 20. The good answer scores well only after you open [Case E] and [Case F] and confirm they say what it claims. Marks for pushing back are earned by the checking, not by the confident tone.
Weighting the marks, and the cap for invented cases
Score each answer from 0 to 5 on six criteria, weight them as below, and convert to a mark out of 100. The ratings here are made up to show how the arithmetic and the cap rule work; they say nothing about either product. Call the tools X and Y.
| Criterion | Weight | Tool X rating (Q3) | Tool Y rating (Q3) |
|---|---|---|---|
| Correct answer against your key | 30 | 4 | 3 |
| Authorities real and on point | 25 | 5 | 3 |
| Right court level and jurisdiction | 15 | 3 | 5 |
| Currency and treatment | 15 | 4 | 5 |
| Time to a verified answer | 10 | 4 | 3 |
| Usable draft | 5 | 3 | 4 |
| Mark out of 100 | 81 | 73 |
The mark is each weight times the rating divided by five, summed. For Tool X on Q3 that is 24 + 25 + 9 + 12 + 8 + 3 = 81. Now put all five questions together.
| Question | Tool X | Tool Y |
|---|---|---|
| Q1 control | 90 | 88 |
| Q2 thin authority | 62 | 71 |
| Q3 changed law | 81 | 73 |
| Q4 court level | 70 | 84 |
| Q5 false premise | 75 | 66 |
| Average | 75.6 | 76.4 |
On these invented numbers the tools are level, with Y ahead by 0.8 points. Then apply the cap rule: any answer containing a fabricated authority is capped at 40 whatever else it scored. Suppose Tool Y invented a case in Q5. Its Q5 mark drops from 66 to 40, its average falls to 71.2, and Tool X leads by 4.4. The cap rule matters more than the weights, so agree it with your partners before the trial starts, not after you see who is ahead.
Reading two answers that disagree
When both tools answer, put their authorities in two columns. Authorities that appear in both are your likely core, so check those first for proposition and treatment. Authorities in only one answer are where the extra value or the extra risk sits. In an invented case, Tool X cites four authorities and Tool Y cites six, and three overlap. That leaves one unique to X and three unique to Y, so four to verify before anything else. If the two tools state different rules, read the primary source yourself and don't average the answers.
Trial uploads, retention and training in the vendors' words
A trial tempts people to paste in real client work. Decide the rule first, then ask each vendor the same questions in writing.
| Topic | Thomson Reuters, per its CoCounsel pages | LexisNexis, per its Protégé pages |
|---|---|---|
| Training on your content | User content and prompts aren't used to train or improve CoCounsel Essentials; partners are contractually barred from training on customer data | "We never use customer data to train AI models"; customer data isn't used to enhance AI performance for other customers |
| Calls to model providers | Interactions with generative AI providers use zero-retention API calls | Not described in the pages I read |
| Retention | Set per organisation with a client manager; users can delete content at any time; no fixed period given | Vaults are kept until you delete them; up to 10 unsaved uploads are purged at the end of each session; Vault results are kept for 90 days under conversation history |
| Limits you can plan around | Not stated in the pages I read | Up to 50 Vaults, each holding 1 to 500 documents; uploading 11 or more documents prompts you to create a Vault |
Those are vendor statements. Ask for the same wording in your agreement, and ask what happens to trial uploads when the trial ends. If your supervision rules don't let client files go into a trial, use redacted closed matters or public documents; the tests above work just as well on those.
Marketing numbers versus your numbers
Both vendors publish studies, and neither answers your question. Thomson Reuters' CoCounsel Legal page reports that 76% agree it improved research and drafting quality, 64% agree it helped reduce risk, and that time on document review, research and drafting fell by 33% on average. The footnote credits a Forrester Total Economic Impact study commissioned by Thomson Reuters in 2026, and the page doesn't give the sample size or the base for each percentage. LexisNexis cites a Forrester study of large law firms, commissioned by LexisNexis in May 2025, reporting a 344% return on investment over three years.
Both studies are vendor-commissioned, and the Lexis one covers large firms. They tell you what the vendors choose to measure, mostly time saved, and nothing about accuracy on your practice areas. If you want to use them in a conversation, ask the salesperson for the sample size and the base, then rely on your own scorecard.
If drafting is half your use
Add a sixth test built from your own precedents. LexisNexis says Protégé can draft from your organisation's documents and connects to iManage, SharePoint, NetDocuments and other document systems. Thomson Reuters says CoCounsel Essentials drafts on the web and inside Word, and names Knowledge Search for working across your own content. Take a clause from your best precedent and ask each tool to draft it for a new client scenario.
Then measure how much of the draft you rewrote. My rule of thumb: under 20% rewritten is good, 20% to 50% is fair and over 50% means you would have been quicker alone. Compare the result with how the tool handled your research questions, because a product that researches well and drafts poorly may still be worth buying if drafting isn't most of your week. An AI legal brief-writing routine with a cite-check shows the same checks applied to a longer document, and general models versus CoCounsel covers whether a legal-specific tool is needed at all.
CoCounsel or Lexis: questions firms ask before a trial
Is Lexis+ AI the same product as Lexis+ with Protégé?
Yes. LexisNexis says that as of February 2026 Lexis+ AI was renamed Lexis+ with Protégé, and anyone who used Lexis+ AI now has access to the renamed product. Some older materials and reviews still use the former name, so check which name a comparison article is describing before you rely on its feature list.
Which should a Westlaw firm choose?
Trial CoCounsel Legal first, because it includes Westlaw Advantage content you may already subscribe to, and ask for a credit against your current subscription. Test Lexis as well only if you would genuinely consider changing research database. Moving a firm's research habits for the sake of an AI layer is rarely worth the retraining.
Do both products check citations?
Both vendors describe citation tooling. Thomson Reuters lists KeyCite integration and a litigation document analysis tool it says catches hallucinated cases, and LexisNexis describes Shepard's and a Shepard's Verify feature that flags unsupported statements. Treat these as features to test, not guarantees, and still read every authority a filing relies on.
Can I upload client documents during a trial?
Only if your supervision rules, your client's terms and the trial terms allow it. Safer options are closed matters with names redacted, or public documents. Lexis describes Vaults with set retention behaviour, and Thomson Reuters says retention is set per organisation, so ask both what happens to trial uploads and when they are deleted.
Is the pricing really quote-only?
Not entirely. Thomson Reuters shows online pricing to new customers at firms of up to ten lawyers after you enter your details, and LexisNexis says firms of one to three lawyers can buy through its online store. The only law-firm figure I found on a public page is Westlaw Advantage on its own, from $256.75 a month depending on jurisdiction coverage, so you still need the online pricing, the store or a written quote for CoCounsel and Lexis+ with Protégé.
Further reads
- AI App for Lawyers: Score Accuracy, Privacy and Cost in a Short Trial — A two-week trial scorecard for accuracy, privacy and cost.
- How to Check Sources and Citations in AI Research — General habits for checking sources in any AI research.
- AI Paralegal Software: What a Small Firm Should Test Before Paying — What to test before paying for AI paralegal software.
- AI Software for Solicitors: Data, Supervision and Supplier Checks — Data, supervision and supplier checks before signing.
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Per-seat versus pay-as-you-go pricing for small firms.
- AI Contract Review for Small Firms: What It Catches and Misses — What AI contract review catches and misses.
- Best AI for Solo Lawyers: A Stack That Fits One Person — Assistant, practice-management AI and contract tool for one lawyer: three priced stacks from $120 to $275 a month, and what to leave out.
- Best Legal AI for Criminal Law: A Small Defence Firm's Shortlist — Which AI to use for bodycam transcripts, research, drafting and intake in a small defence practice, with list prices and a vendor data questionnaire.
- AI for Barristers: A Chambers Checklist for Casework and BSB Rules — Turn BSB guidance into a practical chambers policy and a per-matter check before any AI-assisted work is used.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Thomson Reuters CoCounsel plans page, CoCounsel Legal, CoCounsel Essentials and Westlaw Advantage product pages and pricing pages; LexisNexis Lexis+ with Protégé, Protégé and small law firm pages.