Best AI Chatbot for Lawyers: General Models vs CoCounsel

For a small law firm, a legal-specific tool such as CoCounsel is the better choice when the work depends on citing real authority, and a business-plan general chatbot is enough for drafting, summarising and tidying your own documents. Often the answer is one of each, with a firm rule for which tasks go where.

The choice is rarely about which model is "smarter". It comes down to two practical questions: does the answer come with sources you can open in a click, and do the data terms let client facts in at all? A general chatbot can be excellent at the first draft and still be the wrong place to look up a case.

Follow me on Instagram@sagnikteaches

Whether a small firm needs a legal tool at all is a broader question, and the firm-level comparison of ChatGPT and a legal AI tool covers it. The decision here is narrower: a lawyer with budget for one or two seats who wants ChatGPT, Claude, Gemini and Microsoft Copilot measured against CoCounsel on sources, data terms and price before paying.

Connect on LinkedInSagnik Bhattacharya

The five tools, sorted by what they are built to do

ChatGPT, Claude, Gemini and Copilot are general assistants. They were trained on broad text, they answer from patterns in that training plus whatever you paste in or let them search, and none of them is tied to a legal database. CoCounsel Legal is a different kind of product: Thomson Reuters describes it as reasoning from Westlaw content, Practical Law guidance and your firm's own documents, with linked citations for verification. That tie to a curated source is the whole argument for it.

Subscribe on YouTube@sagnikteaches

Notice what the table does not say: that general tools cannot do legal work. They can draft an engagement letter, restructure a witness chronology or produce a plain-English version of a lease clause. The weakness is narrow and serious: they can produce a convincing reference to a case or statute that does not exist, or that exists and says something different. The wider routine for catching that is in how to stop AI inventing case law.

Price: why there is no price column for CoCounsel

CoCounsel has no list price to put beside the others. Its plans page for law firms asks new customers for their sector, number of lawyers, jurisdiction and preferred term (one, two or three years, with longer terms described as typically cheaper) before it shows pricing, and existing customers go through sales. So price the general side from the published rates and treat the legal tool as a written quote. On the general side, ChatGPT Business and Claude Team both list at about $20 per seat a month billed annually ($25 monthly), each with a two-seat minimum, and Microsoft 365 Copilot Business lists at $21 per user a month on annual billing. Gemini comes bundled into Google Workspace plans, so the cost is whatever your Workspace plan already is.

When you request the CoCounsel quote, put these in writing: the number of seats, which research content is included, the contract term, what the renewal uplift can be, and what happens to your uploaded documents if you leave. A quote that only names a monthly figure leaves out most of what you will actually argue about at renewal. Compare it against the cost of the time it would save, using the sum in the worked example further down, not against $25 a month.

Confidentiality terms: read these before any client fact goes in

Duty of confidence does not change because the software is clever. The practical test for any tool is the same four questions, and the answers differ sharply between free consumer accounts and business plans.

1. Is my content used to train the model? On the business plans (ChatGPT Business and Enterprise, Claude Team and Enterprise, Microsoft 365 Copilot, Gemini in Workspace) vendors state that business content is not used for training by default. On consumer plans (Free, Plus, Pro and similar) there is a privacy setting you switch off yourself, and a setting someone forgot to switch off is how client facts end up somewhere unintended. Thomson Reuters states that CoCounsel data is never used to train the underlying models and that the product is SOC 2 and ISO 42001 certified.
2. How long are my prompts and files kept? Retention differs by plan and by whether training is on. Anthropic, for example, says consumer Claude chats can be kept for up to five years when the Model Improvement setting is on. Ask each vendor for the retention period on the exact plan you are buying, in writing.
3. Who at the vendor can read my content? Look for the support-access and abuse-review terms. A vendor that can review flagged conversations is not the same as one that cannot.
4. Where is it processed, and can I choose? ChatGPT Business offers a storage-region choice; Gemini's in-region processing is restricted to higher Workspace tiers; Anthropic documents no regional processing option for its Claude apps. If your clients' contracts or your own regulator restrict where data sits, this decides the shortlist before features do.

Two further points sit outside the vendor's terms. First, the engagement letter: check whether it says anything about using AI tools on client matters, and add a short clause if not. Second, free accounts and personal logins: staff who feel the firm's tool is slow will paste into their own. A one-page acceptable-use policy that names the approved tool and bans the rest closes that gap better than a warning email.

For the confidentiality side in more depth, see whether lawyers can use ChatGPT without breaching confidentiality. None of this is legal advice: your professional conduct rules and your regulator's guidance decide what is permitted, so check them or ask an adviser.

Citation checking: where the two types actually differ

This is the dividing line, so it deserves precision. With a general chatbot, any citation in the answer is a claim you must verify from scratch: find the case in a proper database, confirm the court and year, read the passage, check it is cited for the proposition the chatbot gave and confirm it has not been overruled. With CoCounsel, the answer is built from Westlaw content and links back to it, so the first two steps collapse into a click. Reading the passage and checking it is still good law remain your job.

That asymmetry changes the economics. A general chatbot's research answer can look quicker and then take longer, because every line needs verification from nothing. A cited answer may need fifteen minutes of reading rather than forty minutes of hunting. Here is a routine that works for either tool.

1. Copy every authority named in the answer into a list, one line each, with the proposition it is supposed to support.
2. Open each one in your own research database, not through the chatbot's link alone.
3. Read the passage that is supposed to support the proposition. If you cannot find it, mark the line "unverified" and do not use it.
4. Check the status of the authority: later treatment, amendments, repeal.
5. Record the check, with initials and date, in the matter file.

The same routine is spelled out for a research task in a citation-checking workflow for legal research.

A realistic miss, so you know what it looks like

An illustrative case. A paralegal asks a general chatbot for authority that a landlord must mitigate loss after a tenant abandons a lease. The answer returns three case names with years and a neat summary. Two are real and say roughly that. The third has a plausible name, a real-sounding court and a quotation, but no such decision can be found. Nothing in the tone separated the real ones from the invented one. The giveaway was only that the quotation did not appear in the cited report. The fix is not a better prompt; it is the rule that no authority from a general chatbot goes into a document until a person has opened it in a proper source.

Sums for a five-lawyer firm trialling two seats

This is an illustration with made-up numbers. Take a firm of five fee-earners handling a mix of property, employment and small commercial matters. In a typical week, the firm identifies four recurring task groups. For each, the team estimates hours across the firm and which tool type would serve:

Say the firm believes a good tool trims 25% off the research and checking hours (14 a week, so 3.5 hours) and 30% off drafting and summarising (18 a week, so 5.4 hours). That is roughly nine hours a week recovered across five people, about 36 hours a month. At an internal cost of, say, $60 an hour, that is around $2,160 a month of fee-earner time. Those percentages are guesses, so the trial in the next section exists to replace them with measured numbers. If your measured saving on research is 5% rather than 25%, the legal tool's quote has to be much lower to make sense, and you will know that before signing.

On the cost side, two general seats for the pilot are about $40 to $50 a month at the listed Business or Team rates, which is small enough to be a rounding error. The legal tool is the real commitment, which is why its trial should be scored against the same ten tasks, below.

Scoring sheet: seven criteria that decide it

Print this, give each tool a score from 0 to 3 on each line and multiply by the weight. The weights below suit a firm doing mostly advisory work; a litigation-heavy firm might double the weight on authority.

A maximum score is 42. General assistants often win on drafting and admin, and the legal tool tends to win on the two heaviest-weighted lines. If a tool scores a 0 on data terms, it is out whatever its total.

A two-week trial with ten real tasks

Pick two or three people, one general assistant on a business plan and one CoCounsel trial. Use matters that are real but low-risk, and anonymise facts for the general tool until you have confirmed its settings. Give both tools the same ten tasks:

1. Research a narrow point in an area you know well, so you can judge the answer.
2. Research a point in an area you know poorly.
3. Draft a client letter explaining an outcome in plain language.
4. Summarise a 40-page contract into one page of key terms.
5. List the inconsistencies between two versions of an agreement.
6. Produce a chronology from a long email chain.
7. Draft a first-pass clause from an instruction.
8. Check a draft against a precedent you supply.
9. Answer a question that has no good answer, to see whether the tool admits it.
10. Ask for authority on a point where you already know the leading case.

Task 9 is the most revealing. A tool that confidently invents an answer to an unanswerable question is showing you how it will behave when you are tired and busy. Task 10 is a planted check: if the tool misses a case you know is central, you have learned how much to trust its coverage.

Score each output with the sheet above and time each task including your checking time. The checking time is the number most buyers forget to record.

The same research request in a general model and in CoCounsel

Here is an illustrative prompt for a general assistant on a business plan, with client facts removed:

Role: assistant to a lawyer. I need a first-pass outline, not advice.
Issue: whether a commercial landlord must try to re-let premises after
a tenant abandons a fixed-term lease.
Jurisdiction: [state your jurisdiction].
Give me: (1) the general rule in plain English, (2) the three questions
a court would ask, (3) a list of the kinds of authority I should search
for, described by type. Do NOT cite specific cases or statutes. If you
are unsure, say so.

Notice the last instruction. Rather than asking a general model for citations it cannot reliably give, the prompt asks it to map the problem and describe what to search for. A sample output, again illustrative, would give a short statement of the rule, the three questions (was the abandonment accepted, were reasonable steps taken to re-let, what did the lease say about mitigation) and a search list: statutory provisions on landlord and tenant, appellate decisions on mitigation, practice notes on lease termination. You would fix one thing: the "general rule" paragraph needs checking against your jurisdiction's actual position, because the model blends rules from different places.

The same task in a legal tool is different in kind. You describe the issue, the tool proposes a research plan, runs it against its own database and returns a draft with linked authorities. Your work becomes reading those authorities and confirming the status of each. You keep the general assistant for the part after research: turning your notes into a letter, or a one-page summary for the client.

Which firm should buy what

These are rules of thumb, not guarantees.

• Solo practitioner doing routine conveyancing, wills or employment letters. Start with a general assistant on a business or team plan, plus your existing research subscription, and write down your verification routine. A legal AI tool becomes worth a quote only once you do research-heavy work most weeks.
• Three-to-ten lawyer litigation firm. The legal tool is easier to justify. Research and brief preparation are where cited, source-linked answers pay back, and an invented case in a filed document is the costliest error in the profession.
• A firm that already lives in Microsoft 365 or Google Workspace. Copilot or Gemini may cover the tidy-up work at little extra cost, since both read the files and mail already there. Check admin settings so they only see what each person is allowed to see.
• A firm with a strict client-consent regime, such as regulated clients or in-house legal departments. Let the data terms and regional processing drive the shortlist before you look at features.
• A firm wanting the AI inside its case-management system. That is a different purchase, covered in Clio, Smokeball and Lawmatics compared.

What to ask on a CoCounsel sales call

You will get more from the demo if you arrive with a list, because demos are built around the product's best day.

• Which jurisdictions and practice areas are fully covered, and which are thin?
• Show me an answer where you cannot find authority. What does the tool say?
• Can I export the research trail and attach it to a matter file?
• What exactly does "never used to train" cover: prompts, uploaded documents, outputs, feedback ratings?
• How are documents I upload stored, for how long, and how do I delete them?
• What does the contract say about price increases at renewal?
• Can I add and remove seats mid-term?
• Are features released during the term included? The law-firm plans page says new features come without an upgrade during the initial contract term, so ask what changes at renewal.

Ask the same questions of a general vendor, adjusting for the product. A general vendor's answer to "show me where you cannot find authority" is itself informative.

Mistakes that show up after the purchase

Three recur. The first is buying the legal tool and then using it as a general chatbot for tasks that a $25 seat would handle, so the quote never earns its keep. The second is the reverse: leaving research on a general chatbot because "everyone already has it", with no verification routine. The third is silent rollout. If the firm has not agreed which tasks go where, you get a mix where one lawyer verifies everything and another pastes a general chatbot's reference straight into a letter.

A cheap fix for all three is a single table in the staff handbook, four columns wide: task type, approved tool, what may be pasted in, who checks the output. Here is an illustrative version for the five-lawyer firm above:

Notice that the last column names a person in a role, never "the AI". If you would rather build that with someone, that is what my AI implementation consultation is for, though a firm can do it alone in an afternoon.

Keeping the decision honest at the six-month point

Diary a review six months after the choice. Pull the actual usage report, ask each user for their two best and two worst experiences, and recompute the saving from real hours. Tools in this area change quickly, and a vendor's pricing, content coverage and data terms can all move in that time, so re-read the privacy and training pages at the review rather than trusting what you read at purchase. If a tool has dropped into occasional use, cut the seats at the next renewal date instead of letting them roll over. If the general assistant is carrying the work well with a sound verification habit, the legal tool may not need to renew. If research has become the bottleneck, that is the signal to widen the legal tool's seats.

For the hub of everything AI-related on the site, see the AI guide. CoCounsel's own product page, with its data and security statements, is on the Thomson Reuters site.

Questions small firms ask before choosing

Does CoCounsel remove the need to check citations?

No. CoCounsel links its answers to Westlaw and Practical Law material so checking is faster, but a lawyer still reads the authority, confirms it says what the answer claims and confirms it is still good law. The tool shortens the check; it never replaces it.

Is a free chatbot account ever acceptable for client work?

Only for work containing no client facts, such as explaining a legal concept in general terms or rewording your own marketing text. Anything that identifies a client or matter belongs on a business plan or legal tool whose terms you have read and whose training setting you have confirmed.

Can I use both a general model and CoCounsel?

Many small firms do. The legal tool handles research and authority-backed drafting. The general assistant handles tidy-up jobs: summarising your own notes, rewording letters, building checklists. The rule is that each task type goes to one tool and everyone knows which.

How do I get CoCounsel pricing for a small firm?

Thomson Reuters publishes no list price. Its law-firm plans page asks for your sector, number of lawyers, jurisdiction and a one-, two- or three-year term before showing pricing, and it offers a free trial and a demo. Get the figure in writing, with seats, research content and renewal terms spelled out.

Should the whole firm switch on the same day?

Better to pilot with two or three people on real but low-risk tasks for a fortnight, score the results, then decide. A firm-wide rollout before you know where the tool fails tends to produce one bad incident and a ban.

Further reads

• Can Lawyers Use AI Note Takers in Client Meetings? — Recording and transcribing client meetings has its own confidentiality rules.
• Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Per-seat against pay-as-you-go pricing models for legal AI.
• AI for Small Law Firms: What to Automate First — Which jobs to hand to whichever tool you pick, in order.
• AI Contract Review Software for Lawyers: Spellbook vs Gavel Exec — Compare Spellbook, Gavel Exec and Copilot in Word on pricing, playbooks and data handling, then test all three on one planted-error contract.
• Best AI for Australian Lawyers: Tools, Privilege and Practice Rules — Legal AI against general assistants for an Australian firm, with a privilege test, a court-by-court map of AI rules and a supervision routine.
• Best AI for Immigration Lawyers: Intake to Filing, Stage by Stage — A nine-stage map of an immigration case with the AI tool type, the data check and the manual step for each, plus a costed worked example.
• 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.
• CoCounsel Pricing: What a Small Firm Should Ask Before Buying — What Thomson Reuters publishes about CoCounsel pricing for small firms, what it doesn't, and eleven questions plus an email to get a real quote.
• CoCounsel or Lexis+ With Protégé? Legal Research AI for a Small Firm — Run the same five research questions through CoCounsel and Lexis+ with Protégé, check every citation, and score the answers before you pay.
• AI for Immigration Lawyers: Docketwise or ChatGPT for Forms? — Run one immigration form through Docketwise and ChatGPT: what each catches, the cost per filing, the privacy settings and the steps that stay manual.
• Lexis+ With Protégé Pricing: Online Plans for Firms of One to Three — What a one-to-three-lawyer firm can learn about Lexis+ with Protégé pricing online, plus the renewal terms and extra charges to check.
• AI Content for Law Firms: Website Articles That Pass Advertising Rules — A topic brief, drafting prompt, twelve advertising-rule checks and a before-and-after rewrite for law firm website articles.
• AI Legal Brief Writing: A Draft and Cite-Check Routine — A routine for small litigation firms: build an authority table, draft only from it, cite-check each source line by line, and meet court AI-disclosure rules.
• 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 Legal product, security and plans pages; OpenAI, Anthropic, Google and Microsoft business-plan pricing and data-use pages.