Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms

Pay per seat for people who'll use a legal AI tool most working days, and pay as you go for everyone else. To find the line, divide the seat price by the cost of one metered use. Most research-grade legal AI is sold per seat on annual terms, so pay-as-you-go usually means general AI billed by usage or monthly plans.

The mistake small firms make is buying seats for headcount rather than for use. In a five-lawyer firm it's common for one or two people to use a research tool daily and the rest to open it a few times a month. Seats for all five can cost more than double what the firm needs, and the vendor has little reason to point that out.

Follow me on Instagram@sagnikteaches

What each pricing model actually charges for

Few legal vendors publish prices. Harvey, Spellbook and Lexis+ with Protégé quote through sales. Thomson Reuters shows CoCounsel Legal pricing online to new customers once you enter your sector, number of lawyers, jurisdiction and term, with one-, two- and three-year plans and savings for longer terms. So the realistic comparison is often between a quote you have to request and general AI plans with public prices.

Connect on LinkedInSagnik Bhattacharya

Credit-based pricing needs its own sum, because the unit isn't obvious. Take an illustrative pack of 1,000 credits a month, where a short question uses 1 credit, a document review uses 1 credit per 10 pages, and unused credits expire at month end. A paralegal reviewing a 300-page disclosure bundle uses 30 credits, which sounds trivial. Run the same bundle through five different questions and it's 150, and a month with two large matters can empty the pack by the third week. During the trial, ask the vendor for the credit cost of your three most common tasks on your own documents, then multiply by last month's volume. If a normal month lands at 60 to 80% of the pack, it fits. Above that, check what top-up credits cost and whether they're priced above the pack rate, because a busy month will need them.

Subscribe on YouTube@codingliquids

Where a seat starts to beat paying per use

For any tool, the rule is: seat price per month ÷ cost of one use on the metered alternative = the number of uses a month at which the seat becomes cheaper.

Two published price lists make the arithmetic concrete. Paxton lists its individual plan at $499 per user a month on monthly billing, or $2,999 per user a year. The annual plan works out at about $250 a month, but only if you use it for more than six months: $2,999 ÷ $499 is just over six. A lawyer who needs a research tool for one four-month trial should pay monthly ($1,996) rather than annually ($2,999).

At the other end, metered API use is cheaper than most people expect. A 50-page document of around 25,000 words is roughly 33,000 tokens (a million tokens is about 750,000 words). On Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens, summarising it into 1,500 words costs about 9 cents. Two hundred such summaries a month is under $20. The catch is that the API is raw: someone has to put a tool, a confidentiality set-up and a checking routine around it, and it has no legal database behind it.

That's why price alone rarely settles it. Metered general AI is cheap for summarising and drafting from documents you supply; it's the wrong tool for finding authority. Which tasks need which kind of tool is covered in ChatGPT or a legal AI tool for a small firm.

Who in the firm actually needs a seat

Before you ask for a quote, log a month of real demand. Most vendors offer a trial of a week or two; use it with the whole team and count. Here's a filled-in log for an illustrative four-lawyer firm with two support staff:

The pattern is typical: research-grade demand is concentrated in the people who go to court or advise on contested points, while drafting and summarising are spread across everyone. Two research seats plus six general seats covers this firm; six research seats would be paying for access four people barely use.

Three firm shapes and their monthly bills

These budgets use published prices where they exist. Where a legal vendor only quotes, the figure is marked as a placeholder you replace with your own quote.

A solo wills and probate practice

Pay-as-you-go wins outright here. Business plans that don't train on content by default need two seats, so a solo lawyer either pays $40 a month on annual billing for one used seat or uses an individual plan with the training switch off and a strict rule about what goes in.

A four-lawyer litigation boutique

Mixed buying wins. Seats go to the two people whose log shows daily research; everyone else is on the general plan.

A twelve-person conveyancing and private client firm

This firm's work is document-heavy and precedent-driven, with little contested authority to research. Seats on a research tool would sit mostly unused.

For the practice-management add-on, ask for both prices and do the sum on actual users, not headcount. Suppose the quote comes back at an illustrative $35 per user a month, or $300 a month firm-wide. Per-user pricing for all twelve would be $420, so firm-wide wins if everyone uses it. If the trial shows only the six fee-earners opening the AI features, per-user seats for those six cost $210, and the firm-wide price pays $90 a month for access the support team won't touch. The break-even is $300 ÷ $35, a little under nine users. Recheck it when the firm hires, because the answer flips as the team grows.

Costs that never appear on the first quote

• Checking time. Every AI-assisted research answer needs its authorities opened and read. If a lawyer spends 15 minutes verifying each research session, that time belongs in the price. For the litigation associate in the log above, around 20 research sessions a month at 15 minutes each is five hours; at an internal cost of, say, $90 an hour, that's $450 a month on top of the seat. It isn't an argument against the seat, since research without it takes longer, but it's the figure to compare tools on: one whose citations need half the checking is worth more per seat. The routine in stopping AI inventing case law is the minimum.
• Underlying subscriptions. Some research AI sits on top of a legal database subscription. Ask whether the quote includes the content, or assumes the firm already subscribes to it.
• Minimum seats and bundles. A vendor may drop the per-seat price at five seats. That only helps if five people will use it.
• Named users. Seats are almost always tied to one person. Sharing a log-in usually breaches the terms and muddles the audit trail you may need later.
• Onboarding and integration fees. Connecting to a document management or practice-management system can carry a one-off charge.
• Renewal rises. Multi-year discounts can hide an uplift at renewal. Ask for the cap in writing.
• Getting your data out. Saved research, prompts and matter workspaces may not export cleanly if you leave.
• Hosting that outlives the matter. On review platforms charged per gigabyte per month, a closed matter keeps costing until someone deletes or archives it. A 40 GB matter left in place for 18 months after settlement is 720 gigabyte-months; at an illustrative $10 per gigabyte-month, that's $7,200 for data nobody opened. Put a close-out step in the matter checklist, and ask the vendor what cheaper archive tier exists.

Reading a quote before you sign

A general assistant is useful for pulling the commercial terms out of an order form so nothing is missed. Remove anything confidential first; order forms rarely contain client data, but check. A prompt like this works:

Below is a software order form. List, in a table: price per seat,
number of seats, billing frequency, contract term, auto-renewal (yes/no),
notice period to cancel, any price increase at renewal, minimum seats,
one-off fees, and what happens to our data when the contract ends.
For each item, quote the exact clause. If an item isn't in the text,
write "NOT IN THIS DOCUMENT" rather than guessing.

The answer that came back (illustrative) listed the seat price, a 24-month term, auto-renewal "yes", a 90-day notice period and, under price increase, "No price increase clause". That last line was the dangerous one. The order form said it was governed by the vendor's master terms, published separately, and those terms allowed an annual rise. The model had done exactly what it was asked with the text it had. The fix was to paste the master terms in as well and ask again, and to add one more line to the prompt: "List any other documents this one says it incorporates." For the clauses to look for, see auto-renewals, price rises and notice periods in AI software contracts.

Terms worth negotiating on a seat deal

An illustrative case shows why these matter. A six-lawyer firm accepts a vendor's offer of six annual seats at $250 each because the price per seat was lower at six than at two. That's $1,500 a month. After three months, two lawyers use it weekly and the other four have logged in twice. The cost per active user is $750 a month. Two seats at a higher small-quantity price of, say, $300 would have cost $600 in total.

To avoid that position, ask for:

1. A paid pilot on two or three seats for one to three months, with the pilot price credited against an annual deal.
2. A ramp: start with the seats the usage log supports and add seats at the same unit price later in the term.
3. Seat reassignment when someone leaves or changes role, without a new licence.
4. Monthly billing for the first quarter if the vendor won't pilot.
5. A renewal cap on any price rise, and a notice period you can actually meet.
6. Written confirmation that the firm's content isn't used to train models and is deleted on exit, with a date.

Send the requests in writing before the sales call, so they're on the table from the start rather than raised at the end. A filled-in version for the four-lawyer boutique (illustrative):

Thanks for the quote for six seats. Our trial log shows two people using research daily, so we'd like to start with two seats on a three-month paid pilot, with the pilot fees credited against an annual agreement if we go ahead. Seats added later in the term should be at the same unit price. Please also confirm in writing that seats can be reassigned when someone leaves, the maximum rise at renewal, the notice period to cancel, and that our content isn't used for training and is deleted within a stated period after the contract ends.

Every answer to a letter like that is a written commitment you can point to at renewal, which a friendly assurance on a call is not.

If a vendor refuses all six, that tells you how the relationship will run at renewal. For a wider view of the trade-off outside legal software, per-seat vs usage-based AI pricing sets out the same sum for general business tools.

Checking the seats earn their keep at 90 days

Most seat-based tools give an administrator some kind of usage view, such as log-ins, queries or documents per user; if you can't find it, ask the vendor for it during the trial. Pull it at 90 days and set it next to the trial log you started with. In the boutique's case, an illustrative month-three report might show the associate on 19 working days and the partner on only 4, because she moved onto a long trial where the research was already done.

Four days of use a month on a seat costing about $250 is roughly $62 a day of use. That alone isn't a reason to drop it, since trials end and research picks up again, but it's a prompt to look at the pattern across the whole quarter. If her use stays light, the published prices give a clear alternative: two months a year on Paxton's monthly plan at $499 is $998, against $2,999 for the annual seat. For a light user whose research comes in bursts, buying the busy months and using the general assistant the rest of the time saves about $2,000 a year. Note the date the annual term renews and the notice period, and make the decision a month before it, not the week after.

Further reads

• AI Implementation Plan for a Small Law Firm: The First 90 Days — Where buying fits in a firm's first 90 days with AI.
• What an AI Consultant Does for a Law Firm, and What It Costs — What outside help costs, if you want the choice made with you.
• Can Solicitors Use ChatGPT Without Breaching Confidentiality? — The confidentiality question behind the cheaper options.
• Standard vs Premium AI Seats: Who Needs the Bigger Plan? — Deciding who needs the bigger plan on general AI tools.
• Outcome-Based AI Pricing: Paying per Resolution, Task or Result — The newer pricing models some vendors are moving to.
• AI Contract Review for Small Firms: What It Catches and Misses — What contract-review tools catch, before you price one.
• How Small Law Firms Use AI to Draft Letters and Routine Documents — A precedent-first way to draft routine legal letters with AI: risk bands, a cleaned precedent pack, a fill-and-flag prompt and a fee-earner checklist.
• AI Time Capture for Solicitors: Recover Hours You Never Billed — How passive AI time capture works for solicitors, the three ways a small firm can start, a daily review routine and honest arithmetic on what it recovers.
• How Conveyancers Use AI to Cut Admin on Each Transaction — Stage-by-stage admin savings for a conveyancing file, a title-summary prompt with sample output, chaser wording and the tasks AI must never touch.
• AI for Small Law Firms: What to Automate First — A scoring method and a firm-tested order for what a small law firm automates first, with the legal work that should wait.
• Which Legal Tasks Should a Small Firm Never Hand to AI? — A four-question test for legal AI risks, the seven jobs that stay with a named lawyer, and wording to put the line in your firm's AI policy.
• 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 plans and pricing page; Paxton pricing page; OpenAI ChatGPT Business and Anthropic Claude Team pricing pages; Anthropic API pricing; vendor websites for Harvey, Spellbook and Lexis+ with Protégé (no published prices).