What AI Gets Wrong About Visa Requirements

AI chatbots are increasingly the first place people turn for visa information. The problem is that large language models hallucinate, and when they hallucinate about immigration requirements, the consequences can include denied boarding, refused entry, or months of wasted preparation.

The core problem

Large language models don't look up visa requirements in real time (unless they're using a live search tool, and even then they can misinterpret results). They generate plausible-sounding text based on patterns in their training data. Visa rules change constantly. Countries update fees, processing times, document requirements, and eligibility criteria multiple times per year. A model trained on data from 2023 will confidently state 2023 rules as current fact in 2026.

Outdated fees and processing times

Visa fees change regularly. The US B-1/B-2 visitor visa application fee increased from $160 to $185 in June 2023. The UK Standard Visitor visa fee has changed multiple times in recent years. AI models routinely cite old fee amounts because their training data predates the change.

Processing times are even more volatile. The US State Department publishes wait times by embassy that can swing from 2 weeks to 6 months depending on season and staffing. AI models tend to give a single number that may bear no relationship to current reality at any specific embassy.

Invented requirements

This is the most dangerous category. AI models sometimes generate requirements that don't exist, blending rules from different countries or visa categories into a single confident answer.

Examples that have been documented:

• Claiming that ESTA (the US Electronic System for Travel Authorization) requires proof of onward travel to be uploaded during the application. ESTA is a pre-screening form. Proof of onward travel may be requested at the border by CBP officers, but that's a separate process.
• Generating detailed "requirements" for visa categories that have been discontinued or restructured. Thailand's visa landscape has changed repeatedly, and AI models frequently describe arrangements that no longer exist.

Missing country-specific exceptions

Visa rules are full of bilateral exceptions, and AI models are bad at tracking them:

• Citizens of certain countries can enter Japan visa-free for 90 days, others for 30 days, others for 15 days, and some need a visa for any visit. AI models sometimes apply the wrong duration to the wrong nationality.
• The UK has different rules for EU nationals versus Commonwealth citizens versus everyone else, and these rules changed substantially after Brexit.
• New Zealand's NZeTA requirement applies to visa-waiver nationals, but not to Australian citizens or residents. AI models frequently omit this distinction.

Confused visa categories

AI models routinely conflate different visa types within the same country. The US alone has dozens of nonimmigrant visa categories (B, F, H, J, K, L, O, etc.), each with different requirements. Asking about "US work visa requirements" often produces an answer mixing H-1B, L-1, and O-1 rules into a composite that matches no actual visa category.

Why this matters more than other hallucinations

When AI hallucinates about a historical date, the consequence is minor embarrassment. When it hallucinates about visa requirements, the consequences include arriving at an airport without required documentation, submitting applications with incorrect supporting documents, missing deadlines, or overstaying because the stated visa-free period was wrong for your nationality.

How to actually get accurate visa information

The only reliable sources:

1. Official government immigration websites. The US State Department, UK Home Office, Immigration New Zealand, IRCC Canada.
2. The embassy or consulate of your destination country. Not a third-party visa service, not a travel blog, not an AI chatbot.
3. Your airline. Airlines use Timatic, a database maintained by IATA that is updated in near-real-time and used at check-in counters worldwide.

AI is useful for getting a general overview of what visa categories exist. It is not reliable for specific requirements, fees, timelines, or eligibility criteria.

tl;dr

AI chatbots regularly produce incorrect visa information: outdated fees, wrong processing times, invented requirements, missed bilateral exceptions, and confused visa categories. Unlike other AI errors, visa hallucinations can result in denied boarding, refused applications, or illegal overstays. Always verify against official government immigration websites, embassy contacts, or airline Timatic checks.