The way people plan trips is beginning to shift. Instead of opening a dozen tabs and clicking through page after page, more travelers now describe what they want to an AI assistant and get a full itinerary, or a direct answer to a question, in seconds. It is faster, it feels natural, and it is clearly part of where travel is heading. Airlines and travel companies are racing to add these tools, and the early experiences can feel genuinely impressive.
But a recent report from OAG, a company that tracks aviation data, makes a point that is easy to lose in the excitement. All of this cleverness rests on one thing: the data underneath. An AI assistant is not a source of truth in itself; it is a very fluent layer on top of whatever information it was given. Feed it good, current data and it can be genuinely useful. Feed it messy or out-of-date data and it will hand you a confident, well-worded answer that happens to be wrong.
To see why that matters, it helps to know how these systems actually work. Older software was predictable: ask the same question, get the same looked-up answer every time. Modern generative AI is different. It does not look a fact up; it produces its best guess based on patterns in its training data. When the data is strong, the guess is usually right. When the data is weak, the guess is wrong — but it arrives in exactly the same calm, fluent, certain voice as a correct one. There is nothing in the tone to warn you.
OAG points to a real and now well-known case. In 2024, an airline was taken to court after its own AI chatbot told a customer something about refund rules that simply was not true. The bot did not hesitate or hedge; it stated the wrong policy plainly, the customer acted on it, and the airline was held responsible. That is the quiet danger of a smooth answer: a human expert who is unsure will usually sound unsure, but an AI that is wrong sounds precisely as confident as one that is right. For a traveler, there is no built-in signal telling you which one you just received.
This is why Airpaz's Chief Technology Officer, Leonard Liu, says the company's first job is the data, not the demo.
"It is tempting to chase whatever AI feature looks impressive this month, because that is what gets attention," he says. "But we would rather get the unglamorous part right first — the prices, the schedules, the seats we show you should be real and current. A wrong answer delivered smoothly is worse than no answer at all, because you act on it. So most of our effort goes into making sure the information we hold across more than 450 airlines is accurate, not just fast or clever-sounding."
He is deliberately cautious about what he will claim.
"I would be wary of anyone who tells you their AI can predict the perfect moment to book, or never makes a mistake," he says. "We do not make that promise, because it would not be honest. What we can stand behind is the foundation — trustworthy data — since anything useful built on top is only ever as good as that."
The honest limit is that AI's error rate can be reduced but never driven to zero, and no booking site controls every airline's data at its original source. Some mistakes will always slip through, whoever built the tool. A careful traveler is still right to double-check the things that really matter — a tight connection, a refund rule, an unusual routing — rather than trusting any single confident answer, however polished.
What a company can control is how much care goes into the data underneath its tools. Good data does not make AI perfect; but bad data guarantees it will fail, quietly and at scale. In travel, where a wrong answer can mean a missed flight rather than a harmless slip, that foundation is worth far more than the cleverest feature sitting on top of it. The tools worth trusting will be the ones whose makers did the boring work first.
