AI Disclosure

Version 1.0 · Effective July 17, 2026.

This page states, plainly and specifically, which parts of Citation Safe™ use an AI (large language model) and which do not. We keep this page in sync with the code that actually runs — if a feature isn’t listed here, assume it is deterministic unless it is separately labeled “AI-assisted” in the product.

1. Layers 1 and 2 — deterministic, zero LLM

Existence check (Layer 1) and quote-match check (Layer 2) never call a language model. Existence is checked against public case-law and citation indexes (currently CourtListener / Free Law Project) via direct lookup. Quote matching compares the text you cited against the text actually in the source using string and similarity matching. Both layers run on every plan, including the free tier, and are what powers the published error-rate scorecard.

Nothing about Layer 1 or Layer 2 output is generated, phrased, or judged by an AI model. These results are computed, not composed.

2. Layer 3 — proposition support (AI-assisted, paid tiers only)

Layer 3 is the only step in Citation Safe™’s verification pipeline that uses an AI model. It runs only on citations that already passed Layer 1 (existence), only on paid plans, and only judges whether a quoted excerpt appears to support the proposition it was cited for. Layer 3 is powered by Anthropic Claude. By design, Layer 3 can never change or override a Layer 1 or Layer 2 result — it is additive evidence, not a veto.

Layer 3 output is always labeled “AI-assisted” in the product (shown as a proposition-support verdict with a confidence score, e.g. SUPPORTS_STRONGLY, SUPPORTS_PARTIALLY, DOES_NOT_SUPPORT, or AMBIGUOUS) and comes with its own explanation sentence. Free-tier accounts do not receive Layer 3 output at all — they see Layers 1–2 only, with an upgrade notice in place of a proposition verdict.

If the AI call fails, times out, or returns output we cannot validate against our schema, Layer 3 fails closed to AMBIGUOUS or UNAVAILABLE — it never guesses its way to a supports-verdict it did not actually produce.

3. What Layer 3 is not

Layer 3 does not draft legal arguments, does not write your brief, and does not decide whether your citation is “safe to file.” It answers one narrow question: does this quoted excerpt, read on its own, appear to support this specific proposition. You remain solely responsible for independently reviewing every authority, exactly as stated in our Terms of Service and Disclaimer.

4. Your submitted content and AI training

We do not use documents, citations, quotes, or propositions you submit to train or fine-tune any AI model — ours or a third party’s. Our AI sub-processor (Anthropic) is contractually barred from training on data submitted through their API. See Section 4a of our Terms of Service.

5. Other Citation Safe products

Tax Cite Safe, Med Cite Safe, and FAR Check follow the identical Layer 1/2/3 architecture described above — deterministic existence and quote checks, with the same narrowly scoped, clearly labeled AI-assisted proposition layer on paid tiers only.

6. Contact

Questions about this page or how a specific result was produced: support@citationsafe.com. Digital Empire Holdings LLC, 30 N Gould St Ste N, Sheridan, WY 82801.