A solo lawyer billing $250 an hour needs an AI tool to give back about 14 minutes a month to pay for a $20 general-purpose plan — and about 3½ hours to pay for a legal-AI seat quoted at $300. The tool is rarely the problem. The billing model is: under the ABA's guidance, time the AI saves on an hourly matter is time you can no longer bill, so it only pays if it turns into other paid work.
The answerBreak-even is minutes for some tools, hours for others
The question "is legal AI worth it?" has a number behind it: how many hours a month the tool must give back before it costs nothing. That depends on three things you know better than any vendor — your rate, how much of the saved time survives your own review, and whether freed time becomes paid time.
Claude Pro or ChatGPT Plus ($20)
- Hourly practice
- 14 min
- Flat-fee practice
- 7 min
Claude Max 5x or ChatGPT Pro (from $100)
- Hourly practice
- 68 min
- Flat-fee practice
- 34 min
Legal-AI seat, if quoted at $300
- Hourly practice
- 3 h 26 min
- Flat-fee practice
- 1 h 43 min
Legal-AI seat, if quoted at $600
- Hourly practice
- 6 h 52 min
- Flat-fee practice
- 3 h 26 min
| Tool (monthly) | Hourly practice | Flat-fee practice |
|---|---|---|
| Claude Pro or ChatGPT Plus ($20) | 14 min | 7 min |
| Claude Max 5x or ChatGPT Pro (from $100) | 68 min | 34 min |
| Legal-AI seat, if quoted at $300 | 3 h 26 min | 1 h 43 min |
| Legal-AI seat, if quoted at $600 | 6 h 52 min | 3 h 26 min |
Source: CalculatorAI · calculatorai.app · CalculatorAI calculation · claude.com/pricing · openai.com/chatgpt/pricing · legal-AI seats are illustrative quotes
The minutes are raw time saved, before the 30% review allowance is taken off. A general assistant clears its bar in a single afternoon of drafting a month. A dedicated platform has to become part of how the practice runs.
The price listWhat the tools actually cost
Only the general-purpose assistants publish prices. As of October 1, 2026, Claude's pricing page lists Pro at $20 a month ($200 billed annually), Max from $100 a month, and Team Standard seats at $25 a month or $20 billed annually. ChatGPT's pricing page lists Go at $8, Plus at $20 and Pro from $100 a month. We compare those plans in detail in ChatGPT Plus vs Pro vs API and Claude Pro vs Max vs API.
The legal-specific platforms are different:
- CoCounsel and Westlaw (Thomson Reuters) — plans and prices appear only after you enter details about your firm on Thomson Reuters' plans pages.
- Lexis+ AI (LexisNexis) — the product page offers a trial and a sales contact, no list price.
- Spellbook — pricing "is determined by the number of team members on a license," with a 7-day free trial and a demo for a quote.
- Harvey — enterprise-only; no public pricing.
What separates the legal platforms from a general chatbot is not the model alone. CoCounsel, for example, says its research is grounded in Westlaw content and Practical Law guidance with linked citations, and its tabular analysis can review up to 10,000 documents against 100 questions. Citations you can check, and sources a court will accept, are what you are paying for. If you are shortlisting platforms, AI for Legal Research publishes head-to-head comparisons of the main ones — Harvey vs CoCounsel, Lexis AI vs Westlaw Precision, Spellbook vs Claude and more — plus a set of free legal-AI utilities to try before you book a single demo.
The billing trapWhy saved time is not automatically money
This is where most ROI pitches go wrong. The ABA's Formal Opinion 512 (July 29, 2024) says a lawyer billing by the hour charges for the time actually spent — not the time the task would have taken without AI. It also says you generally cannot bill clients for learning a tool, and that a general-purpose AI cost is treated as overhead unless the fee arrangement says otherwise and the client has agreed.
So on an hourly matter, an hour the AI saves is an hour you can no longer bill on that matter. It pays only if you fill it with other paid work, take on more matters, or spend it on work you would otherwise have written off.
Value = time you refill
Saved hours shrink the invoice. The tool pays when the freed time goes into other billable work. Our model assumes half of it does.
Value = the whole saving
The client pays for the outcome, so every hour the AI saves is margin. This is where legal AI pays fastest.
If you are moving any work to flat fees because of it, what to charge as a freelancer walks through pricing a fixed-fee job from a target hourly rate. The arithmetic carries over to legal services unchanged.
Three scenariosWhat a month actually returns
Here is the same lawyer in three months: a light one with 2 hours saved, a typical one with 6, and a heavy drafting month with 12. Each figure is after the 30% review allowance and the tool's monthly cost.
2 h
- $20 plan, hourly
- +$155
- $300 seat, hourly
- −$125
- $300 seat, flat-fee
- +$50
6 h
- $20 plan, hourly
- +$505
- $300 seat, hourly
- +$225
- $300 seat, flat-fee
- +$750
12 h
- $20 plan, hourly
- +$1,030
- $300 seat, hourly
- +$750
- $300 seat, flat-fee
- +$1,800
| Hours saved | $20 plan, hourly | $300 seat, hourly | $300 seat, flat-fee |
|---|---|---|---|
| 2 h | +$155 | −$125 | +$50 |
| 6 h | +$505 | +$225 | +$750 |
| 12 h | +$1,030 | +$750 | +$1,800 |
Source: CalculatorAI · calculatorai.app · CalculatorAI calculation (drafts/legal-ai-roi-solo-lawyer-numbers.mjs)
Two patterns stand out. A cheap general assistant is positive in every scenario, because its bar is so low. A specialised platform is a bet on volume: at 2 hours a month it loses money on hourly work, and at 6 hours or more it earns its seat several times over. Our AI ROI calculator runs the same math with your own hours, rate and tool cost, and shows the payback period.
Where it saves timeAnd where review eats the saving
The tasks with the best return share one feature: checking the output takes much less time than producing it.
First drafts of routine documents
Engagement letters, demand letters, standard clauses. Reviewing a draft is faster than writing one, so most of the saving survives.
Summaries of long records
Deposition transcripts, discovery sets, contracts you need the gist of. Spot-checking a summary is quick.
Issue spotting and checklists
Lists of what to look for in a lease or an NDA. Easy to verify against your own judgment.
Research with checkable citations
Tools grounded in a legal database return sources you can open and confirm. A general chatbot's citations need every one checked.
Before you startConfidentiality and consent
Formal Opinion 512 also asks you to understand how a tool uses the data you put into it, to keep client confidences secure, and — where client information goes into a generative AI tool — to obtain the client's informed consent. The opinion says boilerplate consent in an engagement letter is not enough. In practice:
- Read the tool's data terms before the first prompt: is input used for training, how long is it kept, who can access it?
- Prefer business or enterprise plans that exclude your data from training, and confirm that in writing.
- Keep identifying details out of prompts to general assistants unless the client has agreed.
- Record how the tool was used on the matter, in case the fee or the work is ever questioned.
Measure itTrack the hours before you buy
Every number in this article depends on hours saved, and very few lawyers know theirs. Spend two weeks timing the tasks you would hand to AI — drafting, summarising, research — in a simple log. Our free Timesheet Template does that, with totals worked out automatically. Then run the same tasks with a trial tool and compare.
If you bill those hours, the Invoice Generator turns the log into an invoice, and how to invoice a client as a freelancer covers what a clean invoice should include. To check whether a flat subscription or pay-per-use API pricing is cheaper at your volume, use the AI Subscription vs API calculator.
MethodWhere these numbers come from
- Prices were read from the vendors' own pages on October 1, 2026: claude.com/pricing (Pro $20/month or $200/year, Max from $100/month, Team Standard $25/$20 per seat), openai.com/chatgpt/pricing (Go $8, Plus $20, Pro from $100). Thomson Reuters (CoCounsel, Westlaw), LexisNexis (Lexis+ AI), Spellbook and Harvey do not publish list prices; their rows use illustrative $300 and $600 placeholders.
- The lawyer is illustrative: $250 an hour, 30% of raw saved time spent reviewing output, and — for hourly practice — half of freed time turning into other billable work. Flat-fee practice keeps all of it.
- Formula: net value = hours saved × (1 − 0.30) × $250 × refill share − monthly cost. Break-even raw time = cost ÷ ($250 × refill share) ÷ 0.70.
- Bias: a 30% review allowance is generous for drafting and too low for unverified research. A refill share of 0.5 is a guess; a busy practice with a waiting list refills more, a quiet one less. The calculation ignores onboarding time, which Formal Opinion 512 says generally cannot be billed. It therefore flatters the first month.
- Ethics guidance is summarised from ABA Formal Opinion 512 (July 29, 2024). This article is not legal or ethics advice; check your state bar's own guidance, which can be stricter.
FAQLegal AI costs and ROI
How much does legal AI cost for a solo lawyer? General assistants publish their prices: Claude Pro and ChatGPT Plus are $20 a month, and higher tiers start around $100. Legal-specific platforms — CoCounsel, Westlaw, Lexis+ AI, Spellbook, Harvey — price by quote, often by seat and firm size, so ask for a written quote and run the break-even on it.
How many hours does AI need to save to be worth it? At $250 an hour with a 30% review allowance, about 14 minutes a month pays for a $20 plan on hourly work. A $300-a-month seat needs about 3½ hours. On flat-fee work both figures halve.
Can I bill clients for the time AI saves me? No. Under ABA Formal Opinion 512, hourly billing covers the time actually spent. You can build the value into flat fees, and you can pass a tool's cost through only with a clear prior agreement and the client's informed consent.
Is ChatGPT or Claude good enough for legal research? For drafting, summarising and issue spotting, often yes, with careful review. For research you will rely on, a tool grounded in a legal database with checkable citations saves more, because verifying a general chatbot's citations can cost as much time as it saved.
Do I need client consent to use AI on a matter? Formal Opinion 512 says you should obtain informed consent before putting client confidential information into a generative AI tool, and that boilerplate engagement-letter consent is not enough. Your state bar may add its own requirements.
