Content & Accuracy

A Website That Quotes Dead Law

By ODBA Group · July 13, 2026 · 7 min read

In July 2025, H.R.1 rewrote large parts of how asylum works, a change documented in the Federal Register and reflected in USCIS guidance. Any immigration firm whose asylum page was written in 2024 is now, as of that month, describing a process that no longer runs the way the page says it does. The page did not change. The words still read as confident and correct. A frightened reader has no way to tell that the sentence she is trusting stopped being true in July, and the firm whose name sits at the top of that page is the one telling her the old rule as if it were the current one.

Immigration law does not hold still, so the content cannot either

This is the part of website content that a firm rarely prices in when it goes live: the law underneath it keeps moving. Statutes get rewritten, filing fees change, USCIS updates its Policy Manual, the Board of Immigration Appeals issues a precedent decision that shifts what a category actually requires. A firm's website is not a brochure that stays accurate because the firm's address stayed the same. It is a set of factual claims about a body of law that is edited constantly, and every edit to the law is a silent edit to whether the site is still telling the truth. A page that was accurate the day it launched becomes wrong not through any mistake by the firm, but simply because the ground under it moved and nobody told the page.

The problem is not just that a page went stale. It is that nobody can tell which one.

When the law moves, the firm faces a question that sounds simple and usually is not: which pages just became wrong? On most sites there is no way to answer it, because the content was written as prose with no record of where any single claim came from. The asylum page asserts a deadline; nobody can say, months later, which statute or regulation that deadline was drawn from, so nobody can check whether that source still says what the page says. The firm is left to reread the entire site by hand every time the law shifts, which means in practice it rereads none of it, and the wrong sentence stays live for a year. The failure runs deeper than a page aging. It is that an unsourced page cannot even be found and corrected when the thing it describes changes, because there is no thread tying the claim back to the document that would tell you it is now false.

Sourcing every claim is what makes a site fixable

The alternative is to build the content so that every factual claim carries, underneath it, the primary document it was drawn from: the specific section of the USCIS Policy Manual, the EOIR practice manual, the BIA decision, the line in the Federal Register. Written that way, the site stops being a wall of confident prose and becomes an auditable set of claims, each one tied to a source that can be rechecked. When H.R.1 rewrites asylum, the firm does not reread the whole site guessing; it looks at which pages were sourced to the provisions that changed, and those are exactly the pages that need rewriting. Sourcing is not academic decoration. It is the thread that lets a firm find the three pages that just died instead of hoping it remembers them, and it is the difference between a site that can be kept true and one that can only be trusted and hoped over.

Content with no source does not just age. Sometimes it was never true.

Staleness is the slower danger. The faster one is content that was wrong the day it was published, and that is exactly what a firm risks when it lets an AI builder generate a website, in the marketing phrase, in thirty seconds. A general-purpose language model does not retrieve the law; it predicts fluent text, and when it does not know a specific it produces a confident specific anyway. Lawyers already have a public warning about what that produces. In Mata v. Avianca, decided in the Southern District of New York in June 2023, two attorneys were sanctioned five thousand dollars after they filed a brief containing six case citations that did not exist, invented by ChatGPT and submitted as real law. Those were experienced lawyers checking a court filing. A website generated the same way, describing asylum or a U-visa to a scared reader who cannot check it, carries the identical failure mode with none of the review a court filing at least eventually gets.

The reader treats every line as the firm's word

Here is why the accuracy is not a technicality. To the person reading it, a law firm's website is not marketing; it is the closest thing to legal information she can get without paying for a consultation, and she reads it as the firm speaking. Nothing on that site is legal advice, and a well-built site says so plainly. But a false statement of what the law requires is still the firm putting its name behind a claim that is wrong, to exactly the audience least able to catch it, and in a profession where New York's own rules forbid misleading claims about a lawyer's services. The site does not get to be casually approximate because it is "just the website." It is the firm, in writing, to the public, and every line on it is the firm vouching.

The firm whose asylum page can be traced, claim by claim, to the sources it came from spent July rewriting the handful of pages H.R.1 actually touched, and left the rest alone because it could prove the rest was still current. The firm whose page was generated in thirty seconds from nothing in particular is still quoting the old rule, and will keep quoting it until a client, or opposing counsel, or a bar complaint, notices before the firm does. Both sites look equally confident tonight. Only one of them can tell you why it is still true.