Gemini Is Quietly Eating ChatGPT's Lunch

The Plot Twist Nobody Saw Coming

For most of 2024 and 2025, the AI conversation began and ended with OpenAI. ChatGPT was the category, full stop. So it is worth pausing on a fresh batch of SimilarWeb data that shows Gemini's generative AI traffic share climbing from 5.4% to 18.2% over the past several months, while ChatGPT's share slid from 87.2% to 68%. Gemini visits are up 49% over the past twelve weeks. OpenAI is down 22%. Those are not rounding errors.

This is the most concrete signal yet that the AI assistant market is actually competitive, and that being first is not the same as being best, or even being preferred. For enterprise buyers, investors, and anyone trying to figure out where to build, this reshuffling matters enormously.

Why the Gap Is Widening

The irony here is rich. Google spent years looking like the company most threatened by the generative AI wave, and it has quietly turned its distribution advantages into a real weapon. Gemini lives inside Gmail, Search, Android, and Chrome. OpenAI, for all its brand recognition, is still largely a destination product. People have to choose to go there. Gemini increasingly just shows up.

At the same time, OpenAI is reportedly exploring an acquisition of Pinterest, a move that would extend its reach into social search and visual discovery. Pinterest shares rose about 3% on the news. It is an interesting signal about where OpenAI thinks its gaps are, and visual data is clearly part of the answer.

The Web's Old Bargain Is Breaking

While the traffic numbers shift, a quieter structural fight is playing out. CloudFlare's CEO made pointed remarks late last year accusing Google of abusing its monopoly position in search by merging its search and AI crawlers into a single system. The practical effect: websites must choose between blocking AI scraping to protect their intellectual property, or allowing crawling in exchange for search referral traffic that may or may not materialize.

For nearly a quarter century, the web ran on an implicit deal. You let the crawler in, you got the traffic. That deal is fraying. CloudFlare has also flagged aggressive crawling by Anthropic and OpenAI, with very little referral traffic returned. Regulators, the CEO believes, should take a hard look.

Copyright Risk Is No Longer Theoretical

Fresh research adds another layer of pressure on the frontier labs. A study tested four major models, including Gemini 2.5 Pro, Claude 3.7 Sonnet, GPT-4, and one other, for their ability to reproduce copyrighted text. The results were striking. Gemini returned copyrighted material at a roughly 76% extraction rate. After jailbreaking, Claude came in near 96%. GPT-4 was significantly more resistant, refusing after a small number of attempts.

The findings do not settle the legal debate, but they give plaintiffs in ongoing copyright lawsuits something tangible to work with. The question of whether AI models have memorized and can reproduce protected text is no longer just theoretical.

The Fed, the Economy, and the Bigger Picture

Away from AI, the Federal Reserve is navigating its own turbulence. The DOJ launched a probe into the Fed's $2.5 billion headquarters renovation, which Fed Chair Jerome Powell did not treat as a building dispute. He said plainly that the threat of criminal charges is a consequence of the Federal Reserve setting interest rates based on its best assessment of the public interest rather than presidential preferences. Former Fed chairs from both parties and multiple former Treasury secretaries have rallied behind Powell. Markets wobbled on the news.

On the jobs front, December added just 50,000 positions against a forecast of 73,000, with the November figure also revised downward. The unemployment rate came in slightly better than expected at 4.4%, but the overall picture is one of a softening labor market heading into 2026.

What It Means Going Forward

The Gemini numbers should reframe how the industry talks about AI dominance. Market leadership in a platform era is not about who had the best product in year one. It is about distribution, defaults, and data flywheels. Google has all three. The coming months will show whether OpenAI's moves into visual data and social discovery are enough to stabilize its position, or whether the gap continues to widen.

Meanwhile, the legal and regulatory environment around AI is tightening from multiple directions at once: copyright, crawler behavior, and chip export controls. DeepSeek's growing footprint in developing markets adds a geopolitical dimension that is only going to intensify. The race is far from over, but it looks very different today than it did twelve months ago.