Papers AI
Built on legal reasoning.
Not legal-sounding output.
The gap between output that sounds right and reasoning that is right shows up at every stage of a legal case — evaluation, drafting, research, response. We built for the second.
The problem
Most people cannot tell when AI is wrong about the law.
That is not a weakness unique to AI. It is the nature of expertise. A trained attorney reading a petition draft sees things a first-time petitioner cannot — an argument that looks strong on paper but collapses under the adjudication standard, evidence that is persuasive in style but insufficient in substance, a framing that technically answers the question while missing what the adjudicator is actually asking.
Most AI tools answer that problem by generating output that looks professional — well-formatted, citation-rich, confident in tone. Looking professional and being professionally sound are not the same thing. The gap between them is where denials happen — not just in evaluation, but in every drafted section, every sourced claim, every RFE response that sounded thorough but addressed the wrong issue.
We approached the problem differently. Instead of teaching the AI what good legal filings look like, we built it to reason about whether a specific argument satisfies the applicable legal standard — and to say clearly when it does not. That discipline runs the full lifecycle: evaluation, petition drafting, evidence research, and response preparation.
Methodology
Five rules we enforce, not suggest.
They govern every stage — evaluation, petition drafting, evidence research, and RFE response preparation.
01
A legal burden, not a style check
Legal standards define a burden of proof, not a style of argument. We treat that burden as a hard constraint throughout the entire workflow — evaluation, drafting, evidence research, RFE response. If documented evidence does not satisfy it, the output says so. The analysis does not soften conclusions to be encouraging.
02
We reason from the adjudicator's perspective
Most tools are calibrated to help you build the strongest argument. We calibrated ours to ask what a neutral adjudicator would conclude given the record. An advocate asks what can be argued. An adjudicator asks whether the record supports the claim. We built for the second question — at every stage, from evaluation through petition drafting.
03
Documented, not assumed
Every claim in our analysis is tagged: Documented, Inferred, or Unsupported. Only Documented evidence counts. This discipline runs the full lifecycle — a petition section built on inferred claims is no stronger than an evaluation that fails on them. We tell you which is which, at every step.
04
Every rule cites its source
Every legal principle our AI applies cites the authority that establishes it — the statute, regulation, or administrative decision from which the rule derives. Not as decoration. As the basis on which the rule is enforced, so you or your attorney can verify it independently. A citation-free output is an assertion. A cited output is a claim you can check.
05
We don't fill gaps with guesses
When your record is silent on something material — a citation count, a deployment detail, an adopter's name — our AI says so. It does not generate plausible-sounding facts to complete an analysis. Fabricated information in a legal-analytical tool is not a rough draft. It is the kind of error that causes denials.
Open source
The reasoning is public.
Read how it works.
We open-source our evaluation skills as a Claude plugin — the same logic used in the hosted product, free to inspect, run, and critique. If you are relying on this AI to prepare a significant legal filing, you should be able to read exactly how it reaches its conclusions.
We also welcome methodology critiques from practicing attorneys and paralegals. The reasoning has been argued out once. It can be argued out again.
Responsible AI
Three things we will not say.
We will not tell you your case will be approved.
Immigration adjudication is discretionary. A case our AI rates as strong is not guaranteed approval. A case rated high risk is not guaranteed denial. Every filing is decided on its individual record, by a human adjudicator, at a specific point in time. What we give you is an honest assessment of where your record stands against the legal standard — and what would need to change.
We will not position AI as a replacement for attorney judgment.
An experienced immigration attorney brings something we have built with — judgment calibrated across real case outcomes, awareness of how adjudicators at specific service centers read evidence, and the ability to catch a problem in your specific filing that no general framework surfaces. AI prepares. Attorneys judge. That is not a limitation of the AI. It is the right architecture.
We are not a law firm.
The Papers Company does not provide legal advice, does not create attorney-client relationships, and does not operate a lawyer referral service. Our outputs are research-grade legal analysis intended to help you have a more informed conversation with a licensed immigration attorney — not to replace that conversation.