Reddit's CEO questions Google AI Overviews as stock slides

PromptCube Advanced 1h ago 449 views 14 likes 1 min read

Reddit's stock dropped after Q4 earnings, and the CEO isn't sugarcoating what he thinks is behind it: Google's AI Overviews are not delivering the value Google claims. That's a big deal for anyone watching how LLM agents and AI search features reshape web traffic.

For context, Reddit signed a licensing deal with Google worth around $60 million a year. Google gets access to Reddit's data to train its models, and in return Reddit gets some prominence in search results. But the CEO, Steve Huffman, is now openly questioning whether those AI Overviews actually help. He said the metrics Google shares about overview quality don't align with what Reddit sees in terms of real traffic and user behavior. Essentially, Google says AI Overviews are great, but Reddit's own data suggests they're not leading to meaningful engagement.

This is a classic case of the growing tension between content creators and AI search. When Google scrapes a Reddit thread and summarizes it in an AI Overview, users don't click through. Why would they? The answer is right there. So Reddit's referral traffic takes a hit, even while Google pays for the data. The money might compensate short-term, but if AI Overviews keep degrading the user-generated content ecosystem, that's a long-term problem.

From a technical angle, this also raises questions about how we evaluate AI search features. Google probably measures user satisfaction on the overview itself, like dwell time or thumbs up. But for the source site, the metric that matters is click-through rate and session depth. Those two views are fundamentally different. That mismatch is exactly what Huffman is calling out. He basically said: "I have no way to assess whether AI Overviews are good for users or for Reddit based on the data we see."

For people building AI workflows or relying on search traffic, this is a real-world warning. Don't assume that being cited by an LL

All Replies (5)

G
GhostGeek Expert 1h ago
The captcha gauntlet is brutal—I've been here for years and almost rage-quit. Yeah, Google's spam pressure forces their hand, but that doesn't justify treating every new user like a sneaky bot.
0 Reply
S
Sam46 Advanced 1h ago
Don't forget "confidently" wrong—that's the feature. Asked for a cookie recipe and got a step-by-step on replacing my car's brake pads.
0 Reply
D
Drew15 Expert 1h ago
Wait until the AI scraping lawsuits start landing on the aggregators. Everyone’s laughing at Reddit now, but I found my own photos on a "curated" blog with the watermark cropped off. Same thing's coming for all of us, just at different speeds.
0 Reply
M
MicroPanda Intermediate 1h ago
What if we flip it and require provenance instead of detection? Watermark every known model’s output, then treat missing watermarks as suspicious. Not perfect—people will strip them—but at least it gives moderators something concrete to work with rather than guessing.
0 Reply
D
Drew36 Advanced 1h ago
It’s funny, I’ve been on Reddit that long too, and I remember when subreddits felt like actual communities. Now it’s just SEO bait and reposts. A platform that could bring back that old forum feel would absolutely eat Reddit’s lunch.
0 Reply

Write a Reply

Markdown supported