Why OpenAI paused ChatGPT ads to fight Google’s Gemini

For years, OpenAI set the pace of the generative AI revolution with ChatGPT.

Their partnership with Microsoft seemed unbeatable, coupling innovation and enterprise distribution. However, the lead they once held is now in the balance. 

Faced with evidence that Google’s Gemini has not only caught up but surpassed capabilities in critical areas, OpenAI CEO Sam Altman made a dramatic decision by declaring an internal “code red.”

Doing so paused all non-essential initiatives to focus exclusively on ChatGPT’s quality. 

The most significant casualty of this shift was OpenAI’s plan to introduce advertising into ChatGPT. 

It’s important to note that OpenAI’s advertising plans are postponed, not canceled. 

Currently, OpenAI cannot effectively monetize its product while simultaneously losing users to a top competitor. 

Retaining a loyal user base by fixing fundamental issues with speed, reliability, and reasoning is the top priority, but that doesn’t mean ads are off the table down the road.

To understand why ChatGPT’s ad plans were put on hold, but will be inevitable in the future, we need to examine how OpenAI fell behind, the steps they are taking to recover, and what this delay means for the future of AI advertising.

The great stumble behind

OpenAI and Microsoft did not necessarily slow down.

Rather, Google’s massive infrastructural investments paid off, revealing weaknesses in the Microsoft-OpenAI alliance.

The primary reason for the shift in performance lies in model architecture. 

Google’s Gemini 3 was built from the ground up as a “native multimodal” model. 

It does not process text, images, video, and code separately. It understands them as intertwined data.

In contrast, ChatGPT combines separate, specialized models:

• GPT-4 for text.
• DALL-E for images.
• Whisper for audio. 

This Frankenstein approach, while initially groundbreaking, has over time become slower, clunkier, and less cohesive than Google’s unified approach.

Google leveraged its advantage of owning its own technology, controlling all the components that comprise Gemini: 

• The custom-designed tensor processing unit (TPU) chips that train the models efficiently.
• The data centers and the model itself.
• The end-user applications, like Android, Gmail, and Docs.

This allows for a level of optimization and cost control that Microsoft and OpenAI struggle to match. 

The Microsoft-OpenAI partnership relies heavily on costly Nvidia GPU integrations, a factor in projected OpenAI losses that could reach $140 billion by 2029, according to Deutsche Bank Research.

Ultimately, the absence of an ecosystem significantly contributed to the shift in ChatGPT’s user sentiment. 

Google seamlessly embedded Gemini into users’ daily workflows, making it feel like one unified assistant across their workspace. 

Whereas Microsoft’s Copilot has been criticized for feeling disjointed, as an add-on feature that can be inconsistent across Word, Excel, Teams, and Windows. 

Recent benchmarks from LMArena showed Gemini 3 outperforming ChatGPT in reasoning, coding, and speed, indicating that a cohesive, integrated machine was beginning to surpass the alliance between Microsoft and ChatGPT.

Dig deeper: OpenAI discusses an ad-driven strategy centered on ChatGPT scale and media partnerships

How ChatGPT and Gemini solve the same problem differently

Here’s an example scenario comparing the processing behavior of both GPTs to illustrate the distinction. 

The goal: You need to stay in Times Square to be near your office, but you hate the noise. You need a “quiet” tech-forward hotel, a verified coworking space nearby (since hotel rooms in Times Square are notoriously small), and a top-rated ramen spot that doesn’t require a 2-hour wait.

The ChatGPT approach 

ChatGPT will likely provide a list of popular, high-volume hotels and restaurants that appear frequently in travel blogs.

• Process: It searches for “Top rated hotels Times Square” and “Ramen near 42nd St.”
• Result: “I recommend the Marriott Marquis or The Knickerbocker. For ramen, Ichiran is highly rated and just a few blocks away. You can find their locations on a map.”
• The gap: It may not realize that Ichiran Times Square often has a 90-minute wait on Tuesday nights, or that the Marriott’s lobby is chaotic for someone trying to take a deep-work Zoom call.

The Gemini approach 

Gemini uses Google Maps and Workspace integrations to find “niche” results that specifically solve for a business traveler’s needs.

• Process: It identifies hotels with “Quiet/Urban” room ratings, maps the walking distance to a professional WeWork, and checks real-time “Popular Times” data for restaurants.
• Result: “I’ve optimized your stay for Times Square. I recommend LUMA Hotel Times Square. Their ‘Urban Rooms’ are specifically designed to be soundproofed from the street noise.
• The integrated logic:
  • Coworking: Since LUMA’s lobby is small, I’ve mapped a 4-minute walk to WeWork at 1460 Broadway. It has a roof terrace with views of the square where you can work with your dog (based on past search history for ‘Pet-Friendly’ offices).
  • Restaurant strategy: Instead of the main Ichiran, I suggest Raku on 9th Ave. It’s a 10-minute walk away, much quieter, and currently has a 4.7-star rating with ‘low wait times’ on Tuesday evenings.
  • Action taken: I’ve added the walking route from LUMA to WeWork to your Google Maps ‘Favorites’ and saved a draft of this itinerary to your Google Calendar.”

In this scenario, Gemini is like a helpful teammate who doesn’t just talk, but actually gets things done. 

ChatGPT acts more like a smart book that tells you facts.

On the other hand, Gemini:

• Connects in real-time to see live traffic.
• Checks your calendar so you aren’t late.
• Even helps you write emails or stories directly in your own documents.

It can remember huge amounts of information and see what is happening right now, solving real-life problems instead of just answering questions.

---

The ‘code red’ response

The “code red” at OpenAI is an acknowledgment that new features and revenue experiments are meaningless if the foundation is crumbling, forcing them to pivot back to the basics.

OpenAI released GPT-5.2 in December as part of a broader push to narrow Gemini’s lead in complex reasoning and coding tasks. 

The directive inside OpenAI is clear: 

• Stop the hallucinations.
• Improve the speed.
• Make the model feel intuitive and personal again. 

They are moving away from a passive chatbot that simply “talks” and possesses agentic capabilities to a system that can reliably execute complex tasks on behalf of the user, an area where Google is currently leading.

Microsoft faces an equal challenge on the product side. 

Their priority is to unify the Copilot experience so that it stops feeling like five separate AI tools and starts functioning as a cohesive system that understands user context across different applications. 

Microsoft must also solve its data silo problem. 

Google wins on personalization because it has access to users’ emails, calendars, and location data. 

Microsoft will likely need to leverage Office 365 data more effectively – and securely – to offer a similar personalized experience, beyond just reselling OpenAI’s models.

Survival precedes monetization: The logic of the pause

The decision to delay ads is a significant indication of the severity of OpenAI’s competitive crisis. Introducing paid ads can cause friction in a user’s experience. 

When a product, in this case ChatGPT, is the market leader, users will tolerate friction as the cost of access. 

However, now that ChatGPT is no longer the leader, users are being tempted by Gemini’s speed and better integration within the product’s free tier. 

As a result, introducing ads into ChatGPT would likely cause user churn. 

OpenAI realized that if they rolled out ads while their product quality was in the balance, they wouldn’t just fail to make money – they might permanently stifle their growth.

As a result, retention must come before revenue. 

OpenAI needs to stabilize the loss of its user base by achieving equal or better performance over Gemini. 

Once trust is re-established and users feel that ChatGPT is their go-to again, can they afford to introduce advertising?

Dig deeper: Sergey Brin: Google ‘messed up’ by underinvesting in AI

Designing ads for conversational AI

While the delay is significant, it is crucial to recognize that the pause is temporary. 

The financial strains facing OpenAI is what makes advertising inevitable in the future. 

To achieve profitability, it needs to generate hundreds of billions in revenue by the end of the decade. 

Subscriptions alone from users will likely not be enough; monetizing their free tier with ads will become a necessity.

However, the “code red” and the pressure from Google will also likely change how those ads eventually appear. 

The delay buys OpenAI time to develop ad formats that are less intrusive and more trustworthy. 

This means that recent tests offering irrelevant app suggestions in the middle of conversations, is no longer going to work.

Future ChatGPT ads will need to be integrated and contextually relevant to avoid driving users to Gemini. 

The goal will be to introduce ads natively without breaking the conversational flow of the product.

Ultimately, the pause on ChatGPT ads is a stepping stone in its growth to contend in the  competitive AI space. 

It is a necessary gamble for OpenAI to build a better “brain” fast enough to counter Google’s. 

The potential ad revenue is the prize waiting at the end, but OpenAI must hone in on product quality before they can claim it.