Google has pulled one of its newest artificial intelligence features from Google Earth, just weeks after launching it, following concerns that the tool was generating inaccurate and potentially misleading information about geographical locations and landmarks.
The feature, which used generative AI to produce descriptive summaries about places visible in Google Earth, was quietly removed after researchers and journalists flagged several instances where the tool had confidently stated things that were simply wrong. In some cases, it reportedly fabricated historical details about buildings and misidentified land use in rural areas.
A Google spokesperson acknowledged the withdrawal, saying the company had decided to “take additional time to improve accuracy and reliability” before any relaunch. It’s a familiar refrain from the tech giant, which has faced repeated criticism over AI tools that ship before they’re truly ready.
“The problem with layering generative AI onto something like Earth is that users inherently trust what they see,” said one digital misinformation researcher familiar with the incident. “A hallucinated fact about a restaurant is annoying. A hallucinated fact about a border or a heritage site is something else entirely.”
That distinction matters. Google Earth is used by journalists, academics, urban planners, and NGOs doing serious work. It’s not a toy. Slapping a generative AI summary layer on top of it without watertight fact-checking built in was always going to raise eyebrows.
This isn’t the first time Google has had to retreat on an AI launch. The company’s Gemini image generator drew significant backlash early in 2024 after it produced historically inaccurate images, and the rollout of AI Overviews in Search prompted widespread mockery when users screenshotted responses suggesting they eat glue to make cheese stick to pizza.
There’s a pattern forming, and Google knows it. The pressure to compete with OpenAI and Microsoft has visibly accelerated product releases in ways that occasionally skip the part where someone checks whether the thing actually works properly.
The Earth AI tool could yet return in a more reliable form, and Google has the resources to do that work properly. Whether it will slow down enough to get there before the next stumble is the more interesting question.
