How it works
Most "AI legal translation" is a model guessing. This isn't that. PlainLanguage runs your text through three plain steps. The AI step is optional, and it belongs to you.
Identify
We read your text and find every legal term we know: single words like plaintiff and multi-word phrases like burden of proof. The match is exact wording at an exact position. If a term isn't in our glossary, we say nothing about it.
We never guess, and we never invent a match.
Explain
Each identified term gets its plain-language meaning from our curated glossary of 7,257 legal terms. Every entry is distilled from multiple sources and verified against corroborating definitions (here's how the glossary is made), and carries context: what to watch for, examples, related terms. The same input gives the same output, every time.
That's what we mean by auditable: every substitution traces back to a specific glossary entry you can inspect. Nothing is generated on the fly.
Refine
A glossary can explain the terms; rewriting whole sentences is a judgment call. If you want that, we prepare a prompt that packages your text with its glossary and strict rewriting rules: preserve legal effect, honor the watch-outs, never guess. Then you run it through the AI you already use and trust.
We never send your text to a model. Not ours, not anyone's.
Why this order matters
Deterministic first, AI last (and optional) means the trustworthy part does the heavy lifting. For a government agency writing under the Plain Writing Act, a legal-aid office, or anyone who has to stand behind a document, "the computer guessed" is not an acceptable answer. "Every term came from a reviewed glossary" is.
- Repeatable: the same text always gets the same suggestions.
- Inspectable: every highlight maps to a glossary entry with a full detail card.
- Contained: your text is processed and returned. It is never stored, never used for training, never shared.