LOIS citator [What It Changes]

TL;DR

• LOIS citator is Filevine's citation-graph and anti-hallucination check inside LOIS Legal Research: a lawyer highlights a passage, LOIS follows later opinions that address the same proposition even when they never name the original case, and returns a source-linked memo.

• As of September 1, 2026, Filevine's team ran LOIS against 68 sanctioned federal filings, checked 2,073 citations, verified 61.8%, flagged 38.2% for attorney review, and labelled 8.4% severe. Those detection figures are Filevine's evaluation, not a court study.

• Filevine also claims LOIS caught 14 silent negative treatments in one case set that Westlaw KeyCite and Lexis Shepard's missed — a vendor head-to-head. Filevine itself says no tool removes the duty to verify before filing.

• A 2-truck shop, a 10-person agency, or a clinic should care because this is “did a later document quietly kill the spec we are citing,” not because they will Shepardize a case.

Key Takeaways

• LOIS citator is not a replacement for Westlaw or Shepard's. It is Filevine's citator-plus-anti-hallucination layer on LOIS Legal Research.

• The mechanism is passage-level, not caption-level: later opinions that never cite the original case can still be surfaced if they address the same holding.

• Filevine's 68-filing test is a vendor evaluation against already-sanctioned federal briefs. It is not a false-negative rate for clean briefs.

• 8.4% of 2,073 citations labelled severe is 174 cites — the same 174 the September 1 press release states. That arithmetic is a check, not independent lab work.

• ABA Model Rule 1.1 still requires the lawyer to verify.

What LOIS citator actually is

LOIS citator is the Filevine tool, inside LOIS Legal Research, that checks whether a highlighted holding is still good law — including later opinions that never name the original case — and that verifies the citations, quotations, and claimed propositions in a draft. Filevine launched the citator and anti-hallucination capabilities on September 1, 2026, via a PR Newswire release and a long evaluation write-up.

A two-truck HVAC shop already lives this risk when it quotes last year's equipment spec in a bid and a later bulletin quietly withdrew the rating without citing the old PDF. A 10-person marketing agency lives it when a brand guideline URL still resolves but a later legal memo forbids the claim. A solo clinic lives it when a payer policy number is real and the quoted sentence is not what the current policy says. LOIS citator is Filevine's version of that check for case law: the caption can be real, the quote can be almost right, and the holding can still be inverted. Small operators will not buy LOIS. They should still steal the question: does our checker look only for a named citation, or does it look for later documents that kill the proposition?

Solos already weighing Clio alternatives are choosing a system of record. Filevine's pitch is that research, citator, and matter files share one environment. A family-law shop on MyCase versus Clio Manage, or a transactional shop on Smokeball versus Clio Manage, is in the same “where does verification live” decision. LOIS citator only helps if the draft and the opinion panel are in the same loop.

Filevine's sanctioned-filing test

Filevine's internal legal research team tested LOIS against briefs and pleadings that courts had already sanctioned or admonished. The evaluation post is the primary methods document.

The starting pool contained 1,314 potential cases. After limiting to attorney-filed federal matters with accessible records, 68 remained. The path there: an operative sample of 103 attorney-filed federal matters, of which 35 were excluded (12 sealed, redacted, or inaccessible; 9 where the court found no AI hallucination; 14 that could not be located). Each of the 68 was run through LOIS with the same prompt, against LOIS's case corpus only — no open-web search that might have read news about the sanctions.

According to Filevine's evaluation, LOIS audited 2,073 citations across those 68 filings, verified 61.8%, flagged 38.2% for a substantive issue requiring attorney review, and identified 8.4% as severe (mischaracterized holdings or citations to the wrong or non-existent case).

According to Filevine, every document in the 68 produced at least one issue for attorney review, the cleanest high-citation document still had roughly 9% of its citations flagged, and the most egregious had 76% flagged. Those are Filevine's labels, not a court's.

The press release restates the same study and adds a count the blog implies: Filevine's team “confirmed 174 severe errors the filing attorneys themselves had missed.” 8.4% of 2,073 is 174.132, which matches that 174.

Sources: Filevine evaluation; PR Newswire launch.

Filevine told LOIS to look for three classes of error: fabricated case law; false or misattributed quotations; and misrepresented or mischaracterized authority, including inverted holdings and altered statutory text. The evaluation walks through examples Filevine selected: a reporter address occupied by a different case; a one-word substitution in a Second Circuit quote (“party” vs “deponent”); a California opinion cited for a duty the court actually rejected; and Nai v. National Asset Mortgage, LLC, where inserting the word “first” into 12 U.S.C. § 2614 changed an occurrence-based limitations trigger. Those examples are Filevine's illustrations, not a random sample of all US filings.

Filevine is explicit about limits. Courts' orders did not always list every bad cite, so the study measures LOIS against court-identified defects, not a comprehensive detection or false-negative rate. LOIS did not flag passages attributed generically to “the court” with no citation. Prompting changes results: a bad prompt can mark a cite “verified” while still asking the user to check a parenthetical. CourtListener currently indexes 8,300,000 precedential opinions — a reminder that Filevine's 68-filing set is a sanctioned-brief slice, not the corpus.

Semantic shepherdizing versus KeyCite and Shepard's

Traditional citators tell you whether a case is still good law by following later opinions that cite it. Filevine's product page says LOIS citator analysis “finds non-citing conflicts,” analyzes at the level of a specific holding or passage, and “returns a structured memo classifying how the holding has been treated.” The LOIS Legal Research page repeats that citator analysis as a named pillar, alongside automatic citation checking and an opinion panel that marks the exact passages LOIS relied on.

According to PR Newswire's Filevine release, in a head-to-head test LOIS caught all 14 “silent” negative treatments in a case set — holdings quietly overruled by opinions that never cite them — and Westlaw's KeyCite and Lexis' Shepard's caught none.

That 14–0 comparison is Filevine's test, not Thomson Reuters' or LexisNexis'. Westlaw's product page still sells KeyCite Overruling Risk and, separately, says Westlaw is built on 175 years of Thomson Reuters knowledge, with Practical Law maintained by 650+ attorney editors. Those Westlaw figures are Thomson Reuters' own. They do not rebut or confirm Filevine's 14-case set. Shepard's is named in Filevine's release as the Lexis citator; Filevine did not publish Lexis's methodology notes for that comparison.

LOIS's umbrella page publishes other Filevine marketing metrics — 15% increased output, 30% faster recall, 14%+ reduced drafting time, 24% efficiency gains. Those are not citator detection rates. Do not mix them into the 61.8/38.2/8.4 table.

Filevine scheduled a live demonstration, From Risk to Reliability: Check the Cite. See the Reasoning. Read the Opinion., for September 10 at 11 a.m. PT / 2 p.m. ET with CEO Ryan Anderson. A webinar date is not a validation study.

The mechanism, without the equations

A lawyer highlights a passage. LOIS follows the citation graph and also searches later opinions that address the same proposition even when they never name the original case. It returns a source-linked memo. Every citation LOIS generates, Filevine says, passes through a secondary verification layer; unverified cases are flagged; the opinion panel opens to the page with supporting paragraphs highlighted. The user can chat with that opinion alone.

That is why Filevine calls it more than Shepardizing. A citator that only walks citing references will miss a later court that kills the rule without mentioning the ancestor case. A checker that only asks “does this citation exist?” will miss a real case quoted for the opposite holding. Filevine's three error classes are the reason the product exists: fabrication, bad quotes, and inverted authority.

Teams already routing draft briefs through US Tech Automations can put a LOIS citator pass on the same step as “export to PDF”: highlight, run the memo, require a human to clear severe flags, then file. That is a model swap on the verify step, not a new matter system.

According to the ABA comment on Model Rule 1.1, Comment [8] says a lawyer should keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology. A citator that misses a silent overruling is a risk. So is a citator that, in Filevine's own sample, flags 38.2% of cites in already-bad briefs and tempts a lawyer to skip reading the opinion.

USTA analysis: 8.4% of 2,073

USTA analysis. Filevine's blog reports 8.4% severe on 2,073 citations. Filevine's press release reports 174 severe errors. 0.084 × 2,073 = 174.132, which rounds to the 174 in the release. The two Filevine documents agree.

The same inputs give 1,281 verified citations at 61.8% (0.618 × 2,073 = 1,281.114) and 792 flagged at 38.2% (0.382 × 2,073 = 791.886). Filevine did not publish those headcounts; they are derived. They still describe only the sanctioned-filing sample.

On the silent-treatment test, Filevine's score is 14 caught and KeyCite/Shepard's 0 in that set. A 14–0 gap is not a market-wide miss rate. It is Filevine's case set.

Inputs: Filevine evaluation; PR Newswire. Arithmetic is USTA's.

US Tech Automations should store the citator memo next to the filed PDF so a later reviewer can see which flags were cleared, not just that a tool ran.

What no citator removes

The press release is blunt: “No technology removes a lawyer's professional obligation to verify every citation before it's filed.” Filevine says courts are sanctioning attorneys over AI-generated errors, “costing careers, licenses, and in at least one case, an entire trial.” That sentence is Filevine's, not a docket search we independently ran. The evaluation post says LOIS can help a practitioner avoid embarrassment and, in some cases, sanctions or fines, and that every document filed or received from opposing counsel should be run through an anti-hallucination tool. That is a vendor recommendation.

According to PR Newswire, Filevine's team checked 2,073 citations in 68 sanctioned federal filings — the same study — and the company will demonstrate the product on September 10. Two Filevine-origin documents, two publishers: the blog and the wire.

LOIS citator does not establish that KeyCite and Shepard's failed outside Filevine's 14-case set. It does not establish a price; Filevine's legal-research FAQ on the product page says pricing varies and to contact them. It does not establish that state-court filings, which Filevine excluded for access reasons, would score the same. It does not establish that a clean brief would see a 38.2% flag rate; the sample was already in trouble.

The Model Rules table of contents still governs. Thomson Reuters' CoCounsel page and Westlaw Edge's plans and pricing are the incumbent research stack many firms will keep even if they add LOIS. Filevine's About copy in the release says the company is headquartered in Salt Lake City and was honored in 2025 by LegalTech Breakthrough, SaaS Awards, Deloitte Technology Fast 500, and a Business Intelligence AI Excellence Award — awards, not citator accuracy.

A buyer's evaluation sequence

A team that already uses US Tech Automations to block “file” until a checklist is green can add “citator memo attached, severe flags = 0 or waived with initials” as the gate. That is the only automation that matters. The model is not the signer.

Signal vs Speculation

Demonstrated signal: as of September 1, 2026, Filevine launched a citator and anti-hallucination check inside LOIS Legal Research; Filevine's evaluation reports 2,073 citations across 68 sanctioned federal filings, 61.8% verified, 38.2% flagged, 8.4% severe (174); Filevine claims 14 silent negative treatments that KeyCite and Shepard's missed in one set; Filevine states the lawyer still must verify; the product page describes non-citing conflict detection and a source-linked memo; a September 10 webinar is on the calendar.

Our read: over 12 months, expect every legal-AI vendor to ship a “did this quote actually appear, and does the holding still stand” pass, because sanctions on fabricated and inverted cites are the failure mode buyers can see. Filevine's 68-filing set will be treated as marketing until a court, a bar, or a third-party lab reruns it.

Our read: over 24–36 months, if passage-level, non-citing conflict detection works outside Filevine's sample, citators that only follow citing references will look incomplete. That is the bet. It only holds if false flags do not bury the severe 8.4%, and if firms still read the opinion. A clinic, an agency, and an HVAC shop will meet the same pattern in policies and specs long before they meet LOIS.

Frequently asked questions

What is LOIS citator?

LOIS citator is Filevine's holding-level citator and anti-hallucination check inside LOIS Legal Research: highlight a passage, follow later opinions that may never name the original case, and get a source-linked memo plus citation verification.

When did it launch?

Filevine announced the citator and anti-hallucination capabilities on September 1, 2026.

Are the 61.8% / 38.2% / 8.4% figures independent?

No. They are Filevine's evaluation of 2,073 citations in 68 already-sanctioned federal filings. Filevine says the study is not a comprehensive detection rate.

Did LOIS beat KeyCite and Shepard's?

Filevine says LOIS caught 14 silent negative treatments in one case set that KeyCite and Shepard's missed. That is a vendor head-to-head, not a market-wide study.

Does LOIS remove the duty to check cites before filing?

No. Filevine's own release says no technology removes that obligation.

What errors does the anti-hallucination pass look for?

Filevine instructed LOIS to look for fabricated case law, false or misattributed quotations, and misrepresented or inverted authority, including altered statutory text.

Can a non-law business use LOIS citator?

LOIS is built for legal research on case law. The operational lesson — check later documents that never cite the original — applies to specs, policies, and guidelines; the product itself is for legal teams.

If your firm already blocks filing until a human has cleared citation flags, put the LOIS citator memo on that same approval step instead of treating the memo as the signature.