Best legal-ops tools

Two different checks are sold under one word. “Cite-checking” means either does this authority exist and is it still good law — a citator’s job, solved since 1873 — or does this authority actually support the sentence it is attached to. The second check is the one that fails, and the one the sanctions are about.

Sort every tool below into one of those two buckets before you buy. A citator returning a green signal on a real, still-good case tells you nothing about whether the pincite supports the proposition next to it.

What changed the buying question

A database tracking AI-fabricated citations in court filings passed 1,590 documented matters worldwide by mid-2026, up from roughly 640 in late 2024. ABA Formal Opinion 512 treats verifying that a case supports its cited proposition as a professional duty rather than an optional QA step.

The sanction that moved firm risk tolerance was the Sixth Circuit’s early-2026 decision in Whiting v. City of Athens: two Tennessee attorneys, more than two dozen fake or misrepresented citations, $15,000 in punitive fines each plus joint responsibility for appellees’ fees and double costs. Part of what drew the sanction was real cases cited for propositions they did not support — the failure mode no citator catches.

Machine detection of that second category is still weak. The LePhantomCite benchmark (1,300 brief excerpts with injected errors, published May 2026) put GPT-5 in an agentic verification loop at 84.4% recall and a 55.0% F1, averaging 15.3 verification steps per excerpt. Those two figures imply precision near 41% — about three of every five flags is a false positive. Budget associate or paralegal time to triage the flag list; no tool on this page files for you.

1. Clearbrief — the record-support checker

Clearbrief is the one built around the second question rather than the first. It runs inside Microsoft Word and ties each sentence to the source it cites, flagging where the cited material does not support the assertion, where a quote is misstated, and where a citation resolves to nothing.

The audit artifact: a hyperlinked report mapping assertion to source, plus generated chronologies from the record. That export is the thing you hand a supervising partner or attach to a certification — the reason to buy Clearbrief over a citator is that it produces evidence of the check, not just a signal.

Pricing: $150 per user per month, or $300 per brief for contingency-fee matters where the cost gets recovered. Free for law students and professors.

Where it is weak: it verifies against the record and the cited sources you give it. It is not a substitute for a treatment check on whether the case is still good law.

Full Clearbrief review →

2. Thomson Reuters CoCounsel — citation integrity with editorial backing

CoCounsel sits on KeyCite and Westlaw’s editorially reviewed treatment data. For federal practice, KeyCite’s depth-of-treatment bars remain the fastest at-a-glance triage signal, and the editorial layer is the reason the output survives an argument about provenance.

The audit artifact: KeyCite treatment history with editorially assigned depth and quotation markers — defensible because a human editor classified the treatment, not a model.

Where it is weak: treatment analysis answers “is this still good law,” not “does this passage support my sentence.” Quick Check narrows the gap on a filed brief but does not close it.

Full CoCounsel review →

3. LexisNexis Protégé — Shepard’s, plus an existence marker

Protégé pairs Shepard’s with Shepard’s Verify, a trust marker confirming a citation resolves to a real case. For state appellate history, Shepard’s signals are more granular than KeyCite’s.

The audit artifact: Shepard’s report plus the Verify marker on each citation — the cleanest paper trail for the existence question specifically.

Where it is weak: same ceiling as CoCounsel. Verify confirms the case is real; it does not read your sentence.

Full Protégé review →

4. Paxton AI — the standalone AI citator

Paxton ships an AI Citator the vendor benchmarks at 86% accuracy. It is the credible answer for firms that will not buy a Westlaw or Lexis seat but need treatment analysis.

Pricing: $499 per user per month, or $2,999 per user per year — the annual rate works out near $250 per user per month, roughly half the monthly list.

Where it is weak: 86% vendor-reported accuracy on treatment classification is a screening signal, not a filing clearance. Treat a Paxton green as “no further review scheduled,” not “verified.”

Full Paxton AI review →

5. midpage — the explainable citator at solo price

midpage classifies every citing reference as Negative, Caution, or Neutral and shows the source quote that drove the classification. Surfacing the underlying quote is what makes the output checkable by the associate reviewing it.

Pricing: $100 per month for one seat, plus metered AI cost, so the real bill moves with volume. $25 per month for students and faculty.

Where it is weak: the plus-AI-costs model makes budgeting harder than a flat seat, and coverage is US case law, statutes, and regulations only.

Full midpage review →

6. vLex Vincent AI — breadth across jurisdictions

Vincent runs on a library of more than one billion editorially enriched documents across 110 jurisdictions. vLex was acquired by Clio for $1B in 2025 and Vincent continues to ship as part of Clio. The reason to pick it is multi-jurisdiction work, not citation verification depth.

Where it is weak: the ownership change concentrates your research and practice-management vendors in one company. Price that concentration into a multi-year commitment.

Full Vincent AI review →

Also worth a look: BriefCatch RealityCheck

BriefCatch launched RealityCheck in March 2026 as a verification layer rather than a research platform. It runs two passes: deterministic validation of case and statute citations against authoritative databases, then a model pass checking that the quoted language appears in the opinion and that the authority supports the proposition. It catches fabricated quotations, misstated holdings, reversed holdings, bad pincites, and misused signals. Applied to the brief in Fletcher v. Experian Information Solutions, Inc., the vendor reports it found every authority error the Fifth Circuit later identified plus seven the court did not mention — a vendor-run test on a known-bad document, so read it as a capability demonstration rather than a measured hit rate. Pricing is not published; there is a free trial.

The verdict

If you can only buy one thing, buy Clearbrief. The existence question is already covered by whatever research seat you own; the record-support question is not covered by anything else, and it is the one generating sanctions. At $150 per user per month it costs less than a single hour of partner time spent reconstructing why a citation went out wrong.

Match rules:

• Federal litigation on an existing Westlaw seat — CoCounsel for treatment, Clearbrief layered on top for support. Do not buy a third citator.
• State appellate practice on Lexis — Protégé with Shepard’s Verify, same Clearbrief layer.
• No Westlaw or Lexis seat, under 10 attorneys — midpage at $100 per month for the citator, Clearbrief per-brief at $300 on filings that matter.
• Contingency-fee shop — Clearbrief’s per-brief billing is the only model here that maps to cost recovery.
• Multi-jurisdiction or cross-border — Vincent AI, and accept that verification depth is the trade.

If none of these fit, the fallback is a written verification protocol with named owners: every citation in a filing gets opened, the pincite gets read against the sentence, and the reviewer signs the checklist. That is what the tools automate, and a protocol enforced by a person beats a tool nobody triages. Buy the tool when the filing volume makes the manual pass the bottleneck — not before.