How Google leapfrogged ChatGPT – and why the AI race may already be over

When Google unveiled its latest artificial intelligence model in November – three years after the launch of rival ChatGPT – the tech giant described it as a “new era of intelligence”. Gemini 3 was faster, better at reasoning, and achieved a record score in Humanity’s Last Exam – a test designed by AI safety researchers to identify artificial intelligence that can meet or surpass human intelligence.

Google’s announcement contained the same kind of bombast that has become common with the launch of new models from major AI firms – but this time it seemed different.

Early users were quick to spot that the new AI model was not just an iterative update, but a whole new way of using the technology. Marc Benioff, the chief executive of tech company Salesforce, described the leap in “reasoning, speed, images, video… everything” as “insane”.

In a post to X, he wrote: “I’ve used ChatGPT every day for three years. Just spent two hours on Gemini 3. I’m not going back.”

He was not alone. Artificial intelligence pioneer Geoffrey Hinton, who is sometimes referred to as the Godfather of AI, said in an interview that Google was now “beginning to overtake” ChatGPT, adding, “my guess is Google will win.”

Another analyst declared Gemini 3 as “the best model ever” after it beat every other model in 19 out of 20 industry benchmark tests that companies use to measure an AI’s capabilities. The only one it took second place in involved coding, which Anthropic’s Claude model came top.

Its success led ChatGPT creator OpenAI, to declare a “code red”, three years after Google declared its own code red in response to the launch of ChatGPT.

The pace of advances left some questioning how Google achieved it in such a short space of time, with the standalone Gemini app launching just last year. Those working for the tech giant have explained the “secret” to Gemini’s success by saying that they simply improved everything. But the real reason may be more insidious.

For more than two decades, Google has been the gateway to the internet. Its search engine is the way most people access and find information on the web, with publishers and content creators relying on it to send traffic their way.

Websites give Google’s search crawler bots special access in order to appear on the results page, with every click allowing the sites to gain visitors and monetise them through online ads, subscriptions and sales.

But as the internet now shifts away from search to AI – with so-called zero-click searches now accounting for more than 60 per cent of all queries – Google has taken its privileged position as the dominant search engine to gain significantly more access to the web than anyone else. It is now using that dominance to train its AI tools.

As the so-called “middle-man of the internet”, security firm Cloudflare acts as a protective barrier against cyber attacks for more than a fifth of all websites. This gives it a unique perspective into how companies are using AI bots to crawl websites in order to train their models.

“Cloudflare sits in front of a substantial portion, more than 20 per cent of the web, and so we have a representative sample to see how much more of the web Google’s bots have access to versus other AI companies that are out there, and the answer is astonishing,” Cloudflare CEO Matthew Prince tells The Independent.

Figures from Cloudflare, shared with The Independent, show that Google has a massive advantage over its rivals when it comes to training its AI.

open image in gallery

Google’s AI crawlers, which have the same access as the ones it uses for its search engine, see 322 per cent more of the web than OpenAI. It also sees 478 per cent more than Meta, 484 per cent more than Anthropic, and 437 per cent more than Microsoft.

“If you believe that whoever has access to the most data will win, then Google will always have an advantage in the market, which no one will be able to overcome. That seems pathologically unfair,” Prince says.

“There is a massive structural advantage to being Google, and if you want to ask, ‘why did Gemini just leapfrog OpenAI and everybody else in the space’, it’s not because of the chips, it's not because of the researchers, it's not because they're smarter. It's because Google has access to more data, and it's giving an unfair advantage. And Google only has access to that data because it is leveraging its monopoly in search.”

Gemini 3 still has fewer users than ChatGPT – around 650 million users compared to 800m – but it is growing at a faster rate. Following the launch of the Nano Banana image generator in August, which is integrated into Gemini, the number of users shot up by 200 million.

Gemini 3 is also built into other Google products like Search, Gmail, and Drive, meaning it reaches billions more people worldwide. Together with Android, Chrome, YouTube and Maps, Google has seven products that have more than 2 billion users.

Gemini 3 is still far from perfect. It still suffers from hallucinations, can fail at technical tasks like debugging a large codebase, and has what some claim to be overly restrictive safety features that limit its performance.

But despite the flaws it is ahead of the competition, and may well stay there if things do not change. One way to make things more fair, according to Prince, is to split Google’s AI crawler from its search crawler.

He has already met with regulators, and is pushing the UK’s Competition and Markets Authority (CMA) to enforce stricter rules on Google to allow other startups to compete in the AI race.

“From a pure fairness perspective, I believe that Google should have to play by the same rules as everyone else in this space,” he says. “And the easiest way to do that, and this is one of the remedies that the CMA is currently considering, is to split the AI and search crawlers up. That way they have to start from scratch with AI, the same way that all the other AI companies are starting from scratch.”

open image in gallery

In October, the CMA designated Google Search with Strategic Market Status under the UK’s new digital markets regime, which will allow it to impose new rules and regulations on how it works.

Google responded by claiming that “unduly onerous regulations” would stifle innovation and growth, as well as slow down product launches. A spokesperson also told The Independent that it already has tools in place for publishers to control how Google accesses their content.

One such tool is Google-Extended, which allows site owners to control whether their content is used for AI training, which Google claims does not impact their inclusion on its search pages. Google currently operates an opt-out model for AI training, meaning websites have to manually add specific instructions to their code to block it.

Prince warns that if new rules are not introduced, then the AI race may already be over, and Google could monopolise artificial intelligence in the same way it did with online search.

“Google is refusing to play by fair rules. And that's the sort of place where it's right for competition authorities to step in and say, ‘this is a failure in the market. You can't leverage your monopoly from search in order to gain a monopoly in AI’,” he says.

“Google has set the world up in such a way that it has such privileged access that I worry that the market for AI will never catch up. And that seems just radically unfair.”