Google’s Gemini Officially Beat ChatGPT

In partnership with

Hey there,

A study just revealed that 700 million people use ChatGPT every week. That's 2.5 billion messages per day & about 29,000 messages every single second.

But contrary to this, 90% of AI projects fail within their first year. 

So, we have millions of people using AI on one side, and so many AI businesses failing on the other. 

What's going on?

How People Are Actually Using AI

Before we get into why so many AI businesses fail, let's look at how people are actually using AI. OpenAI just released research analyzing over 1.1 million real ChatGPT conversations, and the results are fascinating.

The research shows people use ChatGPT in three main ways:

• Asking (49%): Getting information, advice, or different perspectives

• Doing (40%): Creating content, writing drafts, or generating outputs

• Expressing (11%): Reflecting on ideas or just playing around

Here’s the interesting part: "Asking" is growing faster than "Doing." 

People use ChatGPT more as a thinking partner than just making it do stuff.

What are they actually talking about? Three categories dominate nearly 80% of all conversations:

1. Practical Guidance (how-to questions, tutoring, brainstorming ideas)

2. Finding Information (facts, research, product comparisons)

3. Writing (drafting and editing text)

But here's what surprised me: Only 4.2% of messages are about programming, and just 1.9% are about relationships or personal reflection. Despite all the hype about "AI coding" and "AI therapy," most people are using it for everyday thinking and decision-making.

The App Store Battle Shows What's Really Happening

While we're talking about how people use AI, something big just happened: Google's Gemini replaced ChatGPT from the #1 spot on the App Store for the first time.

Why? One new feature: Nano Banana, their AI image editing model. There’s a new trend going around Nano Banana that gave Gemini a huge boost.

The numbers are crazy:

• Gemini gained 12.6 million downloads in September (up from 8.7 million in August)

• It became a top 5 app in 108 countries globally

• Generated $1.6 million in August alone (up 1,291% from January's $115,000)

• 23 million first-time users tried the new image editing features

Users are flocking to Gemini because they can now do complex image edits and create realistic images right inside the main app, instead of using separate tools.

Why Most AI Startups Are Building the Wrong Thing

Now, coming to why even after such high AI adoption rates, most AI projects are failing. That’s because most AI entrepreneurs are still stuck on building solutions for problems that either don't exist or won't exist in 6 months. 

All those AI content generators online? ChatGPT, Gemini, Claude, etc., are constantly improving themselves and can do all that for users. 

In fact, most users don't even bother searching for specialized tools anymore – they go straight to ChatGPT.

And those AI image generators and editors? Just like Gemini's Nano Banana, every major platform is building these features directly into their main apps. 

So if you’re still focusing on “AI content generators”, know that you're basically building features that will become free tomorrow, not real businesses.

What Smart Entrepreneurs Are Actually Building

While everyone's trying to build the next ChatGPT, smart money is moving toward AI agents – systems that actually complete work instead of just chatting about it.

The numbers for this shift are huge:

• The AI agents market was $5.1 billion in 2024 and is projected to hit $47.1 billion by 2030

• 40% of Fortune 500 companies already use AI agents

• Companies see 50% efficiency improvements in customer service, sales, and HR 

These aren't competing with ChatGPT. They're using AI to solve “niche-specific”, expensive problems that require deep domain knowledge and integration with existing business processes.

The "Caring AI" Approach 

Geoffrey Hinton (godfather of AI who invented the neural networks powering ChatGPT) said: "Don't try to dominate superintelligence; design it to care, like a mother wired to protect her child".

He's talking about building AI that serves people instead of replacing them – and this approach is already working.

The OpenAI research backs this up. The highest-quality interactions happen when people use AI for "Asking" – getting help with thinking and decision-making – rather than just "Doing" – generating outputs.

The LocalLlama community on Reddit (approaching 600k members) represents this perfectly: humans staying in control by understanding and customizing their AI tools.

This is Hinton's "maternal instinct" in action – AI that cares about serving you rather than exploiting you.

What You Should Actually Be Building

Based on what's actually working, here's what you should focus on.

Ask yourself these three questions:

1. Does your AI need deep knowledge of a specific industry that ChatGPT can't just learn?

2. Does it integrate with existing business processes in ways generic AI can't?

3. Does your solution get more valuable as AI gets better, not less valuable?

Focus on augmentation, not replacement. The Stanford research proves this works. Don't build tools that replace humans – build tools that make humans superhuman at specific tasks.

Think local and specialized. As AI capabilities move to personal devices, there's huge opportunity in:

• Privacy-focused AI that doesn't send data to big companies

• Cost-effective local models that beat subscription services

• Customizable AI agents that adapt to specific workflows

Build for "Asking," not just "Doing." Since the highest-quality AI interactions happen when people use AI to think better, focus on tools that enhance decision-making and problem-solving rather than just generating outputs.

The entrepreneurs building tools for local AI deployment, industry-specific agents, and personal AI workflows are positioning themselves perfectly for what's coming next.

Don't try to compete with ChatGPT's general capabilities. Use AI as infrastructure to solve specific, expensive problems in industries you understand deeply.

So, what are you building? More importantly, will it still matter when AI gets 10 times better next year?

Hit reply and tell me about your AI project. I want to see if it passes the "still relevant in 2026" test.

- Aashish

P.S. Do let me know if you're working on AI agents or automation tools, I'd love to feature real examples in future newsletters. 

Your Secure Voice AI Deployment Playbook

• Meet HIPAA, GDPR, and SOC 2 standards

• Route calls securely across 100+ locations

• Launch enterprise-grade agents in just weeks

Download The Free Guide