AI Tools for Lawyers (2026)

Five Jobs, Not One Subscription

Most firms do not have an AI problem. They have five different jobs and one subscription pretending to cover all of them. Split the work first and the tooling decision gets simpler, cheaper, and much easier to defend.

Research triage, document review, drafting, know-how, and filing carry different risk. The first four can be assisted. The fifth is where a human signs their name, and nothing on this page changes that.

The Shortlist, With Status

Each row links to a tracked status page carrying the last automated check, a confidence score, and any death signals we picked up. Never adopt from a review site alone - open the status page and look at the date.

For the industry-level view - regulation, procurement, and firm governance - read the companion AI tools for the legal industry collection. To see which categories are losing tools fastest, check the shutdown leaderboard.

Which of These Could Disappear

Shutdown risk in legal AI is not really about product quality. It is about whether the vendor owns the corpus, and whether the model layer it rents is being commoditised underneath it. Point solutions that only summarise are the most exposed, because general models now do that adequately for free.

The Four-Check Citation Gate

Every other AI risk in a firm is commercial. This one is professional. Courts have sanctioned filings built on citations that did not exist, and the reasoning was never that AI had been used - it was that nobody checked. The duty of competence does not soften because a machine wrote the paragraph.

The control is boring and it works. Make it a step in the drafting template, not a cultural expectation, so skipping it is visible.

Pick a Tool in Sixty Seconds

If you only have a minute, answer three questions in order: what is the job, does it touch client material, and would you cite the output. That sequence resolves almost every choice a practitioner faces.

Confidentiality: Read Your Own Plan

Confidentiality failures are quieter than citation failures and much harder to unwind. The question is never whether a vendor seems trustworthy. It is whether your specific plan excludes client material from training and retention by default.

Consumer tiers usually do not. Business and enterprise tiers usually do, but the exclusion often lives in a sub-processor annex rather than the marketing page. Get it in writing, note the effective date, and re-check after any acquisition.

In the EU, the AI Act layers transparency and documentation duties on top of existing conduct rules. It does not replace them, and it creates no safe harbour for unverified output.

Seven-Day Evacuation Plan

Assume one tool in your stack will change hands or shut down within eighteen months. This is the week that makes that a chore instead of a crisis.

1. Inventory. List every AI tool in use, including personal cards, with owner, plan tier, cost, and which of the five stages it touches.
2. Terms. For each tool, paste the training-exclusion clause and its date into the same sheet. No clause means no client material, starting today.
3. Status. Open each tracked status page and note the last check date and any death signals. Flag anything acquired or dormant.
4. Consolidate. One tool per stage. Cancel duplicates - most firms are paying twice for the same summarisation.
5. Export. Pull full exports of precedents, prompt libraries, and matter notes into your document management system.
6. Gate. Add the four-check citation step to your drafting template so the log line cannot vanish quietly.
7. Write it down. One page: approved tools, approved uses, prohibited uses, verification requirement, review date.

Five Mistakes Firms Keep Making

1. Treating a chatbot as a research database. A general model has no citator and no duty of accuracy. It will produce a plausible citation to a case that never existed.
2. Buying at firm level before testing at matter level. Run one real matter through the tool. Pilots on sample documents flatter every vendor.
3. Assuming consumer terms match enterprise terms. They rarely do, and the difference is exactly the clause that protects your client.
4. Leaving know-how in the vendor. Precedent banks built inside a tool are hostages. Export monthly or accept the loss.
5. Never re-checking after an acquisition. Roadmaps, pricing, and data terms all reset when a publisher buys a product. See what happened to Harvey AI and the Casetext status page for how quickly the ground moves.

What This Actually Costs

General models sit at roughly twenty to thirty dollars per seat per month. Legal-specific platforms are negotiated annually and usually land an order of magnitude higher per seat, because you are buying a licensed corpus and an indemnity posture rather than a chat box.

The real cost, though, is rework. One unverified citation reaching a filing can consume more partner time than a year of subscriptions, and it arrives with no warning. Judge spend against that number, not against the free tier.

Price the exit too. Exporting a precedent bank is an afternoon if you have done it before and a fortnight if you have not. Knowing which it is for your firm is most of the value of this page.

Keep Exploring the Graveyard

Professional-services tools fail for the same reasons everywhere: rented models, thin moats, and buyers who consolidate. If you are auditing a wider stack, see AI tools for writing, AI tools for transcription, AI tools for marketers, and AI tools for presentations.

Want the underlying data? Browse every tracked tool from the homepage, compare survivors on the leaderboard, read the latest postmortems, or report a tool you believe is winding down.

Bottom line: choose per job, insist on a licensed corpus before you cite anything, get training exclusion in writing, and keep an export you control. Do that and no acquisition or shutdown can take your practice's work with it.