The Maze: ChatGPT has an audience problem that growth alone cannot solve. Similarweb counted 494 million ChatGPT web users from March through May 2026, but 461 million of them also used Google. That makes the shared group 95% of ChatGPT's audience and only 14% of Google's. For advertisers, the asymmetry kills the easy pitch that ChatGPT is a large new pool of people Google cannot reach. OpenAI needs to sell a different moment, not a different crowd.
The same 461 million people tell two opposite stories. Against ChatGPT's 494 million users, the overlap is nearly universal. Against Google's 3.3 billion, it is a minority. Similarweb's report therefore shows distribution power, not a close audience contest. Google can touch most ChatGPT users without ChatGPT. ChatGPT cannot say the reverse.
Overlap measures reach, not substitution. Similarweb defines shared users as people who visit both sites, with the percentage calculated against each platform's own audience. It does not measure duplicated queries, time spent, purchase intent or ad performance. The method also covers web behaviour, not a cleanly deduplicated universe of apps, APIs and embedded products. A buyer can use Google for navigation and ChatGPT for a complex shortlist. Same person, different job.
OpenAI's best case is context, not incrementality. Mike Ryan's original argument is that ChatGPT Ads must reach people in a better moment of intent: after users have explained constraints, compared options and revealed what would change their decision. Similarweb's ad panel gives that thesis some evidence. Leading advertisers' placements moved from turns 7–8 to turns 14–22 in three weeks, putting ads deeper into the research journey. That is a distinctive surface. It is not the same as distinctive reach.
Early testing can still buy learning—even without a reach moat. Similarweb tracked 5,171 ChatGPT advertisers by early June and says early, sustained spend can win disproportionate impression share while the market is thin. That slightly challenges the idea that there is no first-mover advantage. But the advantage is format knowledge, measurement practice and auction learning. It is not 494 million net-new consumers.
Commerce makes the measurement hurdle higher. Product ads need reliable merchant feeds, current inventory and conversion signals. Without them, conversational intent becomes an attractive theory that media teams cannot value. Google already connects reach, shopping data and measurement. OpenAI must prove that a later, richer conversation produces enough lift to offset audience duplication and the cost of building another channel.
Why it matters: ChatGPT Ads should not be bought as “Google reach, plus 494 million.” The real offer is access to a more explicit decision process inside a conversation. That could become valuable, especially for considered purchases, but it demands a different test. Measure incremental conversion, assisted discovery and learning by conversation depth—not impressions that mostly land on people already inside Google's orbit. Reach is the incumbent's moat. Intent is OpenAI's opening.
