Three weeks.
That's all it took in February 2026 for the AI industry to fracture over advertising.
February 4: Anthropic buys 4 Super Bowl spots to hammer its message: "Ads are coming to AI. But not to Claude." February 9: OpenAI officially launches ads in ChatGPT in the United States. February 18: Perplexity announces it has permanently dropped ads after testing them in 2024. Google, meanwhile, keeps Gemini ad-free while quietly rolling ads into AI Overviews.
This goes well beyond a debate about ad formats. When a user talks to an AI, they're not scanning 10 blue links. They're delegating part of their thinking. They expect one answer, and they need to trust it.
That changes everything.
OpenAI's impossible equation
800 million weekly users. $20 billion in annualized revenue. And a majority of free users that OpenAI cannot convert into subscribers.
The decision to run ads isn't ideological, it's arithmetic. Internal projections point to $1 billion in free-user monetization by 2026, then $25 billion by 2029. Neither subscriptions nor API revenue can cover the cost of serving hundreds of millions of free users. Ads are the only model that scales.
The launch terms tell the story: $60 CPM, $200K minimum commitment, hand-picked launch partners (Expedia, Ford, Adobe, Target). No self-serve buying: every advertiser is manually selected by OpenAI. Everything is calibrated to prevent backlash.
But the core problem remains. In search, when you type "best hotel in Lisbon," your intent is already formed. A sponsored Booking.com link doesn't surprise you. In ChatGPT, an ad can appear on the very first message, while you're still figuring out what you need. It's a bit like asking for a medical opinion and seeing a pharma ad before you've even described your symptoms. The trust context is fundamentally different.
Anthropic and Perplexity: two paths, same conclusion
These two companies end up in the same place through different routes.
Anthropic gets there through math. With a smaller user base, giving up ad revenue costs little. In return, the ad-free promise becomes a decisive B2B argument. CIOs, legal teams, healthcare, finance: every sector where conversations are sensitive needs that guarantee.
To deliver that message, Anthropic went big. Four cinematic spots produced with real actors, aired during the most expensive ad break in the world. One of them shows a guy asking an AI chatbot for a workout program, only to get a sponsored product recommendation instead of an answer. The contrast with Claude, ad-free, does the rest. Result: Claude jumped from 41st to 7th on the US App Store, with a 32% increase in downloads in 3 days. For $8 million in media spend, the return was remarkable.
Perplexity gets there through experience. They tested ads in 2024, then pulled the plug entirely. Their reasoning sums up the whole problem: "A user needs to believe this is the best possible answer. The challenge with ads is that a user would just start doubting everything."
And they're right. For 20 years, we've learned to tell organic results from sponsored links in Google. It's second nature. In a chatbot, there is often a single answer. If it looks influenced, the entire product's value collapses.
Perplexity picked its side: protect the credibility of its answers. They now rely on subscriptions and enterprise (~$200M ARR).
On Anthropic's end, another model is taking shape: agentic commerce. Claude could eventually buy, book, and compare on behalf of the user. No ads displayed, but a transactional layer potentially more lucrative than any CPM. If your AI assistant books a flight directly, who needs a banner?
Google and Meta: the heavyweights are playing a different game
Google is doing something smart that most people miss. Dan Taylor, VP of Global Ads, said it clearly: "There are no ads in the Gemini app, and there are no current plans to change that." But Google already runs ads in AI Overviews and AI Mode within Search.
The distinction is subtle and deliberate. Google found its answer: ads where users tolerate them (Search), a protected conversation where users delegate their thinking (Gemini). With $198 billion in annual ad revenue, it can afford to watch others take the hits first.
Meta is in a tighter spot. Its ad personalization engine remains the most powerful on the market, no one disputes that. But Meta AI is already embedded in WhatsApp, Instagram, and Messenger. Private conversation spaces.
Consider a concrete scenario: you ask Meta AI in WhatsApp for a good Italian restaurant nearby. The response mentions three places. Did Meta get paid to feature one of them? You have no idea. And that uncertainty, in a private message thread, weighs far more than in a feed where ads are part of the scenery.
Meta has the technology. But the shift from feed to intimate conversation moves the bar on what users will accept. That balance is still unresolved.
What this actually means for marketing teams
For 25 years, digital marketing has optimized one thing: capturing attention (impressions, clicks, views). In a world where AI gives one answer, the game shifts: it's no longer about buying visibility, but earning it.
If a user asks Claude which CRM to pick for a B2B startup, and Claude recommends yours without being paid to do so, that's a form of credibility no ad can buy. How do you earn that? With structured content, proof of usage, honest comparisons, and a presence in the sources that LLMs actually consult.
This is exactly what GEO (Generative Engine Optimization) is about: measuring and optimizing a brand's visibility in AI-generated answers. At Nanga, this is what we do. Our platform lets brands track in real time how they appear in responses from ChatGPT, Claude, Gemini, and Perplexity, identify the queries where they're absent, and build a strategy to get recommended.
Metrics to add to your dashboards
These are the indicators we track at Nanga with our clients, and that you should integrate into your reporting:
| Metric | Why it matters |
|---|---|
| Presence in LLM responses for your key queries | This is your "share of recommendation," the AI version of share of voice |
| Quality of traffic from AI responses | Does a visitor coming from an AI recommendation convert better than a classic search click? |
| Citation rate vs competitors | Direct competitive position in the conversational space |
| Trust signals (return visits, session depth) | Detects a perception erosion before it impacts conversion |
What to do in the next 90 days
Month 1: Map. Identify the 10-20 highest-value queries for your business and check what the major LLMs answer. Which competitors get cited? Which sources are used? That's your baseline audit. Our GEO module automates this mapping across every major LLM on the market.
Month 2: Produce. Create "answer-ready" content: comparison pages, decision frameworks, expert FAQs, structured product evidence. The goal isn't to be indexed, it's to be useful inside an AI's answer.
Month 3: Measure and allocate. Set up continuous monitoring of your visibility in LLM responses. Compare with your search performance. Allocate budgets based on actual incremental value, not just volume. This mapping / production / measurement loop is exactly what Nanga helps its clients run day to day.
Bottom line
Four models now coexist: conversational ads (OpenAI), ad-free premium (Anthropic, Perplexity), contextual hybrid (Google), and personalization to reinvent (Meta). No one knows yet which will win.
What is clear: an AI's unpaid recommendation is becoming the most credible channel on the market. The brands that have figured this out are no longer asking how much to spend on ads in ChatGPT. They're asking how to become the answer the AI gives on its own.
Sources: