Every claim links to a source. If we can't cite it, we don't say it.
The most common failure mode of an AI legal tool is making things up. We have a discipline against that, and this page is that discipline written down.
Four rules we follow, every brief, no exceptions.
Six kinds of source, ranked by reliability.
The brief will always prefer a higher-tier source when one is available for a claim.
Federal statutes and regulations
Acts of Congress, federal agency rules, presidential executive orders.
Example:No Surprises Act (42 U.S.C. \u00a7\u00a7 300gg-111 through 300gg-115) for emergency out-of-network bills.
State statutes and regulations
State legislatures, state agency rules, state attorneys general guidance.
Example:California Civil Code \u00a7 1942.5 (anti-retaliation protection for tenants who exercise rights).
Case law
State and federal court opinions, particularly state supreme court and federal circuit decisions.
Example:Henson v. Santander Consumer USA Inc., 137 S. Ct. 1718 (2017) for what is and isn\u2019t a \u2018debt collector\u2019 under the FDCPA.
Agency guidance documents
Letters, opinions, and consumer education from federal and state agencies.
Example:CFPB Bulletin 2013-07 (debt collection prohibitions under FDCPA \u00a7 808).
Market comps and pricing data
Public price databases, salary surveys, industry reports.
Example:BLS Occupational Employment Statistics for typical salary ranges by metro area and role.
Industry-published reports
Reports published by independent industry research organizations, with named methodologies.
Example:JD Power's annual customer retention surveys for telecom and insurance industries.
Specific things we won't use as sources.
- Wikipedia. Good for orientation, not a primary source. The brief never cites it. If a claim originates from a Wikipedia article, we trace it to the article's underlying source and cite that.
- Paywalled news articles. We can't ask a user to pay another company to verify our claim. We'll quote a free public source for the same fact or omit the claim.
- Reddit, Quora, and forum posts. Sometimes useful for tactics, never citable as authority. If a tactic came from a forum, the brief tags it “uncited insight” and you decide whether to use it.
- Other AI's output. We do not cite ChatGPT, Gemini, Claude (other than ours), or any other AI. AI outputs are not authoritative sources by definition.
- “Common knowledge.” If a claim requires verification (a price, a deadline, a statute, a rate), it gets a citation. “Everybody knows” isn't a defense.
- Hallucinated citations. If the AI produces a citation we can't verify against a real source, the citation is stripped from the brief before you see it. This is enforced by a verification step in the brief pipeline, not by trust.
We surface the disagreement, we don't pick a winner.
Sometimes two authoritative sources disagree. A federal law and a state law overlap in a way that's jurisdictionally contested. Two federal circuits have split. An agency's guidance contradicts the underlying statute. The hospital's claim about what a procedure should cost is different from the BLS average for the same code.
When this happens, the brief does three things:
- Names the conflict explicitly. “The California statute and the federal regulation disagree on X.”
- Cites both sources so you can read them.
- Defers to the more specific, more recent, more authoritative source where one exists. If neither is clearly more authoritative, the brief stays neutral and tells you to call a professional.
What the brief does not do is pretend the conflict doesn't exist. We'd rather give you an uncomfortable truth than a clean lie.
The “uncited insight” tag.
A lot of negotiation knowledge doesn't live in a statute or a case. It lives in practitioner playbooks, industry conventions, sales-script training documents, and the accumulated experience of people who've done a thousand of these conversations.
That knowledge is real. It's often the most actionable part of a brief. But it doesn't have a single citable source. When the brief uses it, we tag it explicitly:
These insights are weighted lower in the brief and you see them flagged. You can choose to act on them or skip them. We're not hiding the difference between “the law says” and “practitioners report.”
Tell us. We'll fix it.
If a brief cites a law that doesn't apply to your situation, a case that doesn't say what we claim, or a price that's wildly off, email us. We take these reports seriously and we feed them back into the verification pipeline.
cite-error@renegotiateanything.comFor the full set of disclaimers, see /disclaimers. For the outcome statistics that this citation discipline supports, see /track-record.