It’s never been easier to get information about immigration law. But that means it’s actually harder to know if the information you’re getting is actually good. Especially for people who are trying to get advice on their own immigration cases, reliability matters.
New technologies have meant there are more resources, like Documented, dedicated to delivering immigration news and providing context. And immigration lawyers can now reach millions on social media every day with videos explaining new developments. But it’s also become easier than ever for bad actors to deliberately spread misinformation or carelessly misuse information for their own purposes: For every immigration lawyer with a huge TikTok following, there’s an army of imposter bots.
The integration of AI into the ways people typically seek out information has made this all the harder. A June Pew Research study found that 42% of Americans now use AI chatbots like ChatGPT, Claude or Google’s Gemini to search for information, but even more may be running into AI answers that they may not realize are AI. Google, for example, has made a point of feeding you AI overviews in response to your web searches. For every legitimate news site or immigration-lawyer blog, there are pages filled with AI-generated answers to common search questions.
This presents new challenges particularly for immigration law. A recent report from the startup Vals AI, which evaluates and sets benchmarks for AI tools, assessed how well the most popular AI models performed when asked questions about scenarios involving immigration laws. It checked the answers against “rubrics” provided by experts in immigration law that laid out what a fully correct response would need to include.
According to the report, even the best-performing model (a particularly expensive version of Anthropic’s Claude) still provided incomplete answers in the vast majority of situations. Other models did far worse.
Overall, the report concluded that AI tools “should not be relied on to independently advise clients on their specific situation. The gaps in applying the law to a client’s facts are particularly concerning.”
The bottom line: Don’t trust a non-professional to do a professional’s job, especially not now.
There are a few reasons why AI is a particularly bad fit for answering questions about immigration cases. First of all, AI models are limited by the text they’re trained on — which poses a big problem when it comes to an area where things are changing all the time. A model that was trained on a dataset that ended in 2024, for example, wouldn’t know anything about the second Trump administration. (Some models incorporate active web search to try to mitigate this problem.)
Secondly, AI models don’t know what you should be asking, only what you do ask. If you leave out a detail that turns out to be relevant to your immigration case – for example, a potential asylum applicant might not know that applications are generally denied if the applicant’s been in the U.S. for over a year before applying – a chatbot will have no idea. But the confident, authoritative tone used by most AI bots wouldn’t give you any hint of uncertainty or ambiguity.
That’s not to say that all AI models are created equal, or that they always get things wrong. But it’s worth knowing their limitations.
AI tools are particularly bad at navigating recent changes in immigration policy — the exact sort of thing you might expect people to need help navigating. When asked about new regulations and policies from 2025 and 2026 even models that could search the web for recent content were able to provide less than a third of the information the experts were looking for, according to Vals AI’s report.
The finding that jumped out to me, though, is that the category of questions where AI models were least likely to give complete and correct answers was “temporary or fragile status,” including Temporary Protected Status (TPS) and Deferred Action for Childhood Arrivals (DACA). That means that those who are the most vulnerable to losing their status and getting deported are also the most likely to get incorrect and unreliable answers from an AI model.
We don’t know how many people are actually relying on AI tools to give them legal advice, to be clear. And there’s some indication that people generally trust these tools to give them information, but don’t rely on them for high-stakes legal issues. A 2025 survey by Rev — a company that offers AI tools for legal analysis — found that while 43% of Americans were comfortable asking an AI model for help clarifying legal terms, only 6% said they’d be comfortable relying on it for their immigration case.
This was the case even before the rise of AI tools. A 2021 University of Denver study that focused on people who had experienced legal issues found that out of 175 people whose most serious issue had been immigration-related, only 33 reported using the internet as a resource — and even in that case, many of them were going directly to agency websites.
I still worry, for a few reasons. One of them is that AI has become so fully integrated into the internet that people may not know the text they’re reading is AI-generated. Google, for example, has made a point of feeding you AI overviews in response to your web searches (and for what it’s worth, the Google AI model that was tested in the Vals study performed worse overall than every other model tested). Similarly, immigration law websites could be filling their pages with AI-generated responses to common search questions. It’s hard to know who to trust!
Even good information can turn bad when it’s taken out of context. One example from my colleague’s recent feature on TikTok lawyers (which, seriously, you need to read): a video about applying for asylum got reposted by a copycat account months later, at a time when U.S. Citizenship and Immigration Services simply wasn’t processing asylum applications. (Asylum applicants should be especially cautious, given that this administration is making it ever easier for them to get referred for deportation proceedings instead of having USCIS adjudicate their application.)
While the ideal situation would be that everyone who needs a lawyer gets one, there are simply not enough immigration lawyers to go around. As of August 2026, only a quarter of closed immigration court cases had had legal representation. Since immigrants without representation are more likely to lose their cases quickly, this is somewhat of an undercount, but it’s undeniable that the number of immigrants with lawyers is trending downward, and has always represented only a fraction of immigrants facing a deportation order from a judge. For many people, the internet, and by extension AI tools, may be one of the only resources available. But it is not reliable enough to trust on its own.
The best advice I can give you — other than to, seriously, seek out legal advice from actual immigration lawyers if at all possible — is to find authoritative, expert sources you trust and rely on them. If there’s an organization in your area that offers free legal clinics, follow their social media accounts. Follow news outlets that care about informing you accurately about immigration issues (ahem, Documented, ahem ahem). If your friend shares a story or video that turns out to be accurate and useful, maybe listen to them a little more than the friend who keeps reposting things that turn out to be outdated or wrong.
Given the consequences of listening to bad legal advice, it’s worth doing a little extra work to make sure you’re not making a terrible mistake.
