General-purpose models are trained on everything, not just primary law. That includes every legal authority you can come up with, but it also includes social media posts, as well as mainstream and niche media. When a legal subject becomes a public controversy, a flood of non-lawyer commentary can overwhelm the actual holdings.
Probability map. Many people are under the mistaken impression that an LLM is some sort of hypercomplex thinking machine. Really it’s just an incredibly huge probability map of words, patterns, and phrases followed by other words, patterns, and phrases. Unless it’s attached to a database of authoritative sources, nothing connects what it generates back to reality.
Completion pressure. Every model runs under a standing directive that strongly favors providing an answer rather than tell you it can’t. Hand it a gap where a real authority should be, and it fills the gap instead of flagging it.
Latent contradiction. Your LLM prompt can look fine on its face, but against the backdrop of law, non-obvious contradictions can be hiding in wait. That conflict is invisible to you, and sometimes even to the model itself. Something has to give, and completion pressure decides what: it honors the request and manufactures the rest.
An LLM model built on the open web absorbs far more journalism, advocacy, and commentary about the law than it does actual opinions and statutes. For some legal issue that only lawyers care about, never hits mainstream and social media, that rarely matters. For a subject that becomes a headline, the commentary can vastly outnumber the primary sources — and volume is persuasive to a model.
The result is that the model can let popular characterizations or legal mythology bleed in, treat a contested or even fantastical position as established because the internet treated it that way. Secondary framing — how a decision was described — crowds out the decision itself.
It gets worse when a rogue trial court adopts rhetoric over precedent. That one opinion is legally real, and it’s legally authoritative. So the model may over-weight it and repeat its framing as the law, even while it’s an outlier on a crash course with reversal. A citation can be genuine and still be a poor guide to what the law actually is.
The same cause, a few ways it turns up in ordinary practice. Each is routine work you’d never flag as risky — which is exactly how the cause slips in unnoticed.
This applies to everything with mass appeal, whether it’s hot-button political controversy on constitutional law, or mythology around traffic laws and police encounters.
The viral order of the moment — a trial court judge’s headline-grabbing controversial ruling, later reversed on appeal, that the model treats as controlling authority because it was so widely discussed.
Was stop and frisk actually held unconstitutional? Was abortion banned by Dobbs? The model can repeat the popular gloss, even when it’s not what the opinion actually said.
Adding a citation directive (“only cite real cases”) mathematically cannot ever fix the problem. Humans read a citation directive and think the LLM must comply. But every LLM has completion pressure built into its architecture, a mathematical property of the model itself, that will readily override such a directive.
Verbatim checks the words, not the reputation. It confirms whether the case is real and whether the language your brief quotes actually appears in the opinion — so a citation carrying a popular gloss the opinion never contains surfaces as a mismatch. What a case is famous for stops mattering; what it says is what gets verified.
It draws on primary authorities — the opinions and codes themselves — rather than the commentary layered on top of them. The report points you back to the source text at the pin cite, where the actual holding lives, not to the version that trended.
Verbatim reads a finished brief and reports, for every authority it cites, whether the cite is real and whether the quoted language actually appears at the pin cite — so a fabrication surfaces on your screen, not in a show-cause order. Bring a brief and we’ll walk you through the report.