I Built 11 AI Tools for Law Firms — Pick These 5 First

Every law firm I have talked to in the last six months wants AI and is terrified of it. Both feelings are correct.

The want is rational — there is a real partner-hours problem at small firms, and AI is the only lever in a decade that moves the needle on document-heavy work without hiring associates. The terror is also rational — Mata v. Avianca ended careers, the state bars have been issuing opinions at a rate that makes compliance officers seasick, and "I asked ChatGPT" is a discoverable phrase in malpractice depositions.

I built 11 single-purpose AI tools for law firms over the last few months. No suite, no integration, no procurement. Each solves one task and produces an output your senior partner can sign off on. This post is the honest answer — if you can only stand up 5 at a small firm tomorrow, which 5.

Why every law-firm AI listicle is wrong

Search "best AI tools for law firms" and you get the same five posts. Casetext, Harvey, Spellbook, Lexis+ AI, Thomson Reuters CoCounsel. Each one written by an SEO contractor who has never sat in a conference room while a managing partner asks, politely, who is on the hook when the AI hallucinates a cite into a federal filing.

The problem is not that those tools are bad — Harvey is a serious product. The problem is none of them are deployable by a 5-person firm tomorrow. They are platforms. They want an integration project, a training rollout, a six-figure annual commitment, and a champion partner with a year of political capital. The small firm closes the tab and does nothing for another quarter.

The frame I keep coming back to — the AI tools that belong at a small firm right now are the ones that do not depend on the AI being smart. They codify procedures the firm already has, where the AI is checking work, not creating it.

The 5 picks

1. Citation Verifier — the one that addresses the elephant

In June 2023, Steven Schwartz of Levidow, Levidow & Oberman filed a brief in Mata v. Avianca citing six opinions. None existed. ChatGPT had invented them — case names, docket numbers, internal quotes. The court fined the firm $5,000. The lesson most lawyers took was "do not use AI." The correct lesson was "do not file AI work product without a verification pass."

The citation verifier is that verification pass. Paste a brief, motion, or memo. The tool extracts every cite, queries authoritative reporters and court records for each one, and flags any that do not resolve. Pinpoint cites get checked against the actual paragraph. Quoted language gets checked against the source text. Anything that looks AI-confabulated gets a red flag with a reason.

First pick because it is procedural. Opposing counsel is using AI somewhere in their stack. Your law clerk is too. The contract attorney you hired for a doc-review burst is too. You do not need to know which one. You need a checkpoint that catches the fabricated cite before the federal judge does. Under an hour to deploy, faster and cheaper than a paralegal cite-check.

2. Engagement Letter Generator — consistency without the partner-review tax

Every firm has engagement letters. Every firm has too many versions of their engagement letter. The 2014 template the senior partner uses, the 2019 template the litigation group uses, the 2022 version the associate cribbed from a former firm, the one-off the rainmaker drafted on a Saturday night. Half are missing the conflict waiver language the state bar updated in 2023. A quarter have fee provisions that do not match the firm's current rate sheet.

The engagement letter generator takes the matter type, fee structure, parties, and scope, and produces a clean letter that pulls from one canonical set of clauses. Scope, fee provision, retainer terms, conflict disclosure, termination rights, file-retention, governing law. Every letter is the same letter with the variables filled in correctly.

The value is not generation speed — a partner can dictate one in 15 minutes. The value is the firm stops having seven versions of the same document floating around with subtle differences that will be a problem in a fee dispute three years from now. The AI is being a very consistent paralegal that does not have a bad Friday. An afternoon to configure clauses, then any attorney uses it without partner review.

3. ABA 5.12 Audit — compliance is unsexy and required

ABA Model Rules 5.1 and 5.2 govern supervision — partners are responsible for the work of associates and non-lawyers under their authority, and the rule applies whether the "non-lawyer" is a paralegal or an LLM. The state bars have spent two years clarifying that AI use falls under 5.1/5.2. If your firm uses AI anywhere in the production stack, there is a defensible supervision posture you are supposed to have. Most firms do not have it written down.

The ABA 5.12 audit walks through your current practice — what tools, what tasks, who reviews what, what records you keep — and produces a written supervision posture mapped to the rule. Not legal advice. It is the artifact your malpractice carrier wants to see when they ask about your AI policy, and the one you want to hand a bar investigator if a grievance ever gets filed.

Compliance is unsexy. It is also the cheapest insurance policy in the firm. About 40 minutes of partner time. The output sits in the policy folder and gets updated when a tool or rule changes. Pair it with the state bar AI rules lookup for any jurisdictions where you take matters.

4. Discovery Analyzer — the biggest time sink at most firms

Ask any litigation associate where the hours go and the answer is doc review. Production sets of 20,000-200,000 documents that some human has to read, code, and surface. The traditional fix was a contract attorney rotation at $50-90/hr. The modern fix — Relativity, Everlaw, DISCO — costs about the same and needs a project manager.

The discovery analyzer is the small-firm version. Upload a production set. The tool surfaces likely-privileged documents, hot documents matching your case theory, custodian patterns, and a deposition-prep summary by witness. It does not replace human review on high-stakes cases, but it cuts first-pass time 60-80% and produces a privilege log a senior litigator can sign off on after spot-checking. Honest caveat — this is the tool where AI does the most "thinking." The tool flags its own low-confidence calls, which is what a contract attorney never does.

5. Client AI-Use Consent — the conversation you legally must have

Several state bars — California, New York, Florida, Texas — have issued formal opinions in 2024-2025 that informed client consent is required when AI is materially involved in legal work product. Most firms are quietly doing AI work without the conversation, because the conversation feels weird and partners do not want to lose the client. That is a malpractice problem hiding behind an awkward-conversation problem.

The client AI-use consent tool generates three artifacts. A consent form tuned to the matter type and the firm's actual AI usage. A one-page plain-English explainer the client can read. A talking-points script for the kickoff meeting so the conversation is calibrated, not improvised. The signed copy in the matter file is the artifact that protects the firm if there is ever a complaint.

Fifth pick not tenth because the cost of not having it is asymmetric. Ten minutes of partner time per matter, versus a bar grievance and a malpractice claim if anything goes sideways. Most lopsided risk trade on the list, and the one most firms are getting wrong.

The other 6 — what they do and when to deploy them

The other six solve narrower problems or need more lawyer input to produce defensible output. Real tools that real firms use. Short version of each.

• AI policy generator — produces the firm-wide written AI use policy that pairs with the ABA 5.12 audit. Deploy when you do the audit; the two artifacts together are the policy posture.
• Case value tool — early-stage matter valuation for personal injury and contingency-fee intake. Useful at intake, less useful once a case is in active litigation. Plaintiff-side shops will use this. Defense shops will not.
• Demand letter drafter — fast first draft for PI and collections work. Real value at firms doing high-volume demand-letter work; a luxury at firms drafting one a quarter.
• Deposition prep tool — generates outlines, exhibit lists, and likely lines of questioning from a matter file. Pairs well with the discovery analyzer; partial overlap means you can defer this one until after discovery is deployed.
• Flat-fee converter — converts hourly matter histories into defensible flat-fee proposals. Real revenue lever for firms moving away from hourly; irrelevant if you are committed to hourly billing.
• State bar AI rules lookup — jurisdiction-by-jurisdiction lookup of current ethics opinions on AI use. Use once when you do the ABA 5.12 audit; revisit when you take a matter in a new state.

None of these are bad tools. They are second-wave tools — deploy them after the first five are running, when you know which problems your firm actually has, instead of which problems the listicle said you have.

Where I would start if I had one hour

If a managing partner gave me 60 minutes and said "pick one, deploy it, prove it works" — citation verifier. Every single time.

Minute 0-10. Pull three recent briefs the firm shipped in the last six months. One from a senior partner, one from a mid-level associate, one from a junior. Filings that went out the door, not internal drafts.

Minute 10-25. Run each through the citation verifier. The report flags any citation that did not resolve, any pinpoint pointing at the wrong paragraph, and any quoted language that does not match the source. Nine times out of ten the senior partner's brief comes back clean. Once in a while you find a transposed reporter number from a paralegal's typo. The tool catches it before the judge does.

Minute 25-40. The interesting one is the associate brief. Roughly half the time the tool flags either a citation that looks slightly off, or one it cannot resolve at all and flags as possible fabrication. If you get the latter, you have just caught the thing that ended Levidow, Levidow & Oberman's month. Have the conversation with the associate. Find out what tool they used. Fix the gap.

Minute 40-60. Make the rule. Every brief, motion, and memo runs through the citation verifier before it is filed. Verifier output stapled to the file copy. Email the managing partner — "deployed the citation verifier, caught X, setting it as a required pre-filing step, here is the policy memo." That is the proof. That is the conversation that gets you green-lit on the other four.

For firms worried about where the AI runs — what data it touches, whether any matter is hitting a hosted vendor — pair this stack with a local LLM setup. The tools above check work; the local LLM stack handles drafting and analysis on hardware that never leaves your office. Those two layers together are the answer to the "we want AI but cannot send client data to OpenAI" problem.

Full list of 11 tools plus the firm-specific deployment playbook on the law-firms vertical page. Each one is a hosted form — no install, no integration, no procurement. Pick one. Run a brief through it. Decide for yourself.

"The AI tools that belong at a small firm right now are the ones that do not depend on the AI being smart. They are the ones that check work, not the ones that create it. Pick those first. Everything else is the second move."