Google Earth AI Image Tool Pulled After One Day of Deepfake Fears

A feature that lasted about a day

On July 30, 2026, Google added a "create image" button to Google Earth on the web. Users could zoom in on any spot, type a prompt, and have Nano Banana 2, the image model built on Gemini, generate a scene on top of the company's own satellite, aerial and 3D imagery.2426 The launch post from product manager Bryan Horowitz invited people to "transform any place."26 The feature was global from day one, with no waitlist and no check on who was using it.27

It was gone the next day. On July 31, Google added a note to its launch post saying it had seen screenshots of generated imagery that appeared to break its policies. It said it was rolling the feature back to build "stronger guardrails" and noted that it trusted people "uniquely trust Google Earth for a reliable view of the world."2226

The outlets that covered the episode agree on what happened. Henk van Ess, a researcher who writes the Digital Digging newsletter, used ordinary prompts to place refugees near the US-Mexico border, a nuclear plant in Iran, a fatal crash on an Amsterdam street and a bomb crater beside a Gaza hospital.2223 He wrote that "nothing was refused, nothing was softened."22 BBC Verify ran its own tests and got a collapsed Eiffel Tower, a sinkhole swallowing the Great Pyramid and Russian tanks in Kyiv.24 Social media users reportedly made disaster and military scenes, including a recreation of the September 11 attacks.10

Where Google and its critics disagree

The reporting splits on whether Google's safeguards meant anything. Google's position, repeated in its statements, was that generated images never showed up in the shared Google Earth view and carried watermarks marking them as AI-made.2224 The company told people to check suspect images with Lens or the Gemini app.24

Critics said those controls were aimed at the wrong problem. Van Ess reported that a video made in Google Earth fooled Hive's AI-detection tool.22 One account of the testing also found that Gemini could be talked into calling the fake images real.23 Van Ess's main point was that a forged satellite image doesn't need to be convincing by itself, because "it inherits the credibility of the map it was born on."24 Keeping outputs inside the app doesn't matter much when a screenshot can travel anywhere.27

The critics have the stronger case. Google's own statement concedes that people trust Google Earth for a reliable view of the world, and that trust is exactly what made the outputs dangerous. A watermark that only a forensic tool can read does nothing for someone scrolling past a screenshot with a confident caption. The Hindu put it bluntly: the rollback shows the company "is still not ready" to deal with these basic risks.26

The trust damage in a war

The timing made things worse. Writing for Rest of World, Mahsa Alimardani notes that since Israel and the US began strikes on Iran in February, analysts and journalists have leaned on commercial satellite imagery, Google Earth included, as the closest thing to ground truth.1 Investigators trained for conflict work in Syria, Yemen and Sudan learned on the platform.1 Her argument is that Google pulled the feature quickly but cannot pull back the doubt it created. Fake satellite frames are still being screenshotted into new claims, and real imagery now carries a question mark it hasn't earned.1

Bellingcat's Jake Godin made a similar point. Fake satellite imagery already existed, but the feature was "just streamlining the process," which would help it spread.27 The episode also follows other quick retreats. Meta pulled its Muse Image tool three days after a July 7 launch over privacy and copyright complaints.24 OpenAI shut down its Sora video app in March after months of criticism.1 Alimardani describes a familiar cycle: the product ships, the harm shows up, the company withdraws it and promises guardrails it could have built in advance.1

The same push into search

The Google Earth episode is more than a one-off product mistake. Fortune placed it in a longer effort to weave generative AI through Google's products, from Gemini in Docs to AI Overviews in search to Nano Banana in Earth.29 One commentary argued that the company had put Gemini "in every corner" of its software and that in some places it "feels forced."23

Search is where that push has its biggest economic effect, and the parallel with Google Earth is close. Both products work as reference layers that people trust by default, and in both Google has put AI-generated output on top of material it didn't create. In Earth, that material was satellite imagery. In search, it is publishers' journalism, and the cost falls on the outlets that do the verification work the Earth episode showed is now essential.

What the traffic data shows

The evidence on AI Overviews has become much harder to dismiss over the past year. The strongest piece is a randomized field experiment by Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon. When an AI Overview appeared, outbound organic clicks fell 39.8% and searches ending without any click rose 34.5%.1520 Clicks on sponsored results didn't change, which suggests the summaries replace visits to websites rather than just moving attention around the page.20 Across all searches, including those without an overview, the estimated drop in outbound organic clicks was about 18.5%.18

Citations inside the summaries recover little of that traffic. Only 7.1% of outbound clicks on pages with an overview came from links in the AI answer.18 Pew Research found that users clicked a traditional result on 8% of visits when an AI summary appeared, against 15% when none did, and clicked links inside the summary just 1% of the time.15 Ahrefs estimated that an overview cuts clicks to the top-ranked result by 58%.15 That is close to double the 34.5% drop it measured in April 2025.17

The field study also cuts against Google's main defense. The company has long said AI Overviews raise search usage and send "higher quality" clicks, without publishing data that defines what that means.11 The experiment found no change in user satisfaction, perceived search quality or ease of finding information when an overview appeared. Time spent on outside sites stayed about the same.18 On that evidence, readers aren't getting a better product in exchange for the clicks publishers lose.

How big the losses are

The figures for publisher traffic vary, and the differences are worth noting. Chartbeat data on about 2,500 publisher sites, cited by the Reuters Institute, showed Google Search referrals down 33% globally and 38% in the US between November 2024 and November 2025.1215 Chartbeat's 2026 Publisher Playbook later put the year-over-year drop to July 2026 at 40.2%. In that time Google Search fell from 9% of pageviews in July 2024 to 5% two years later.19

Digital Content Next, a trade group for publishers, adds a caveat: AI Overviews can't explain the whole drop, since algorithm changes, publisher strategies and audience habits all play a part.15 That caution is fair, but it doesn't undo the causal findings at the level of individual searches. Smaller outlets are hit hardest. Chartbeat's breakdown shows search referrals down 60% for small publishers, 47% for mid-sized ones and 22% for large ones.12

The argument that AI chatbots will make up the difference also looks weak. Some marketing analysts describe a "split outcome," with AI engines sending smaller but more valuable visitors to sites they cite.16 Chartbeat's newsroom data shows AI referrals making up about 0.01% of publisher traffic, and that share slipped during 2026.19 Digital Content Next concludes that referrals from AI platforms are growing but remain far too small to replace what search used to send.15

Fights in court and with regulators

Publishers have moved from complaints to lawsuits. Penske Media, which owns Rolling Stone and Variety among other titles, argued in a February 2026 filing that Google is "cannibalizing" its traffic.14 Chegg has sued on antitrust grounds, citing a 49% collapse in non-subscriber traffic.12 The European Publishers Council has filed a complaint with the European Commission. A third of publishers surveyed say they would block AI Overviews if they could.17 Google says it is "exploring" a way to opt out of AI Overviews without losing regular search rankings, under pressure from the UK's Competition and Markets Authority. As of early 2026, no such control was widely available.14 In May, Google added a "Further Exploration" links section and other features meant to send more clicks to publishers.17

What the two stories share

In both cases, Google offered small fixes like watermarks and extra links while the core design stayed the same. Google may well bring the Earth feature back. Its statement describes a pause, not a cancellation.27 If it returns with a stricter prompt filter and the same invisible watermark, that will show the company took the wrong lesson from the episode.27

The bigger worry is what happens when both trends continue. The Google Earth episode showed how much journalists and fact-checkers are needed to sort real images from fakes, and how wide the public's digital literacy gap is.23 The search data shows the referral traffic that pays for that work shrinking. Google said it pulled the Earth tool because people rely on it for a reliable view of the world.22 The same reasoning applies to the newsrooms whose search traffic AI Overviews are steadily taking.