Reddit cut a deal with Google. Its AI still ate Reddit's lunch.

Every SEO and GEO team has heard the same instruction for two years: go get corroborated on Reddit. Answer the threads, build the presence, earn the citations, because Reddit is one of the sources AI engines trust more than almost anything else on the internet. So here is the part nobody saw coming. Reddit is now one of the biggest traffic casualties of the exact thing it helped teach those engines to do.

That's not a knock on the strategy — generative engine optimization work built around Reddit still earns citations, and it should stay in the plan. It's a warning about treating any single source, channel, or relationship as protection from what AI Overviews do to the traffic underneath it.

In 2024, Reddit signed a licensing deal with Google worth reportedly around $60 million a year: an ongoing feed of human conversation, handed over so Google's models could train on it, in exchange for a check that landed just ahead of Reddit's IPO. It looked, at the time, like the smartest trade a publisher could make. Get paid for the thing AI was going to take anyway.

Two years later, according to reporting from GHacks Tech News on July 23, 2026, Reddit is reportedly among a group of major publishers, alongside USA Today, Politico, CNN, and Business Insider, weighing whether to start blocking Google's crawlers outright. Not because the licensing money stopped. Because Google's own AI Overviews are cutting into the organic search traffic those same publishers depend on, and the licensing check does not cover that hole.

The company Google leaned on hardest is losing to Google anyway

Sit with that for a second, because it is the whole story in one sentence. Reddit had leverage almost nobody else in publishing has: a direct commercial relationship with Google, real money on the table, and two years of runway to build whatever defense it wanted. And the traffic still went. If a $60 million annual check and a seat at the table were not enough to protect Reddit's search traffic from AI Overviews, that is not a Reddit problem. That is a preview.

This is the piece worth taking home to your own team, whether or not you compete with Reddit for anything. Every marketing org with an organic channel has been telling itself some version of the same story for the last year: the dip is probably algorithm noise, or seasonality, or a bad quarter, and it will smooth out. Reddit had more reason than almost anyone to believe its own version of that story. It didn't smooth out.

There's a specific piece of industry chatter this should quiet down, too — the idea floating around GEO circles that Reddit is somehow insulated from AI-driven traffic pressure because it's the source everyone else is fighting to get corroborated on. Being a trusted source for other people's citations and being protected from AI Overviews on your own site turn out to be two completely different things. Reddit is both, at the same time, and the second one is losing.

It also undercuts a specific comfort a lot of teams have been leaning on without saying it out loud: the idea that being cited by AI engines somehow offsets losing clicks to them. Reddit is cited constantly. It gets pulled into ChatGPT answers, Perplexity summaries, and Google's own AI Overviews as a trusted source of real human opinion, probably more than any other domain on the internet. None of that citation volume shows up as a visit. Being quoted and being clicked are two separate economies, and right now only one of them pays the bills.

The math behind AI Overviews traffic loss

The mechanism is not complicated, and that is what makes it hard to argue with. Research cited in the same reporting found that when a search did not surface an AI Overview, users clicked through to a traditional organic result about 15% of the time. When an AI Overview did appear, that click-through rate dropped to roughly 8%. Same query type, same intent, same searcher. The only thing that changed was whether Google decided to answer the question itself before handing over a link.

Cut a channel's conversion rate nearly in half and nobody would call that a rounding error. But because it shows up dressed as a slightly worse month of organic traffic instead of a labeled event, it is exactly the kind of decline that survives a quarterly review unchallenged. Nobody stands up in the meeting and says the click-through rate on our category's core queries just got cut in half by design. They say traffic was soft this quarter, and everyone nods and moves on.

Four publishers, four bleeding lines

The click-through math turns into real numbers fast once you look at actual publishers instead of an abstract percentage. Comparing June 2025 to June 2026, the same reporting lays out organic US Google traffic declines that should make anyone with a traffic-dependent business a little queasy.

Business Insider losing more than 85% of a traffic line in a year is not a trend, it is closer to a shutdown notice for that channel. USA Today Co.'s CEO has been reported as saying, in effect, that the company has reached its limit, which is not much of an overstatement when your organic search traffic drops by nearly half in twelve months. Politico and CNN's declines look almost mild sitting next to those two, and they are still the kind of number that would trigger a leadership review at nearly any company that isn't a wire service.

The range matters as much as any single figure. A 23% decline and an 85% decline are not the same event happening at different speeds to different newsrooms. They're evidence that this isn't a fluke hitting one unlucky domain, and it isn't evenly distributed either. Whatever combination of query mix, content format, and topical footprint sits behind that spread is exactly the kind of thing a blended traffic number will never surface on its own, which is the whole argument for measuring closer to the query than the property.

“A quarter of your traffic disappearing without a fight is a bad year. Losing it slowly, to a cause nobody can name in the boardroom, is a measurement failure dressed up as a market shift.”

Why your own AI Overviews traffic loss might be invisible right now

Here is the part of the story that matters more than Reddit, or USA Today, or any single publisher's number. All four of those declines happened inside the same measurement problem most marketing teams are still running today: one blended organic-traffic line that goes up or down, with no built-in way to say how much of the move belongs to AI Overviews specifically versus a Google algorithm update, a seasonal dip, or a content team having a slow month.

That blended number is comfortable, and comfort is exactly the danger. It lets a slow bleed sit unaddressed for a year, the way it apparently did at these publishers, because there is always a plausible alternative explanation on hand. Rankings wobble constantly. Seasonality is real. Any one soft month can be explained away without anyone having to say the sentence AI Overviews are eating our clicks out loud, in a room, with a number attached to it.

If you sell into B2B or software categories, this concentration pattern should worry you more, not less. AI Overviews tend to cluster hardest on informational, comparison-shaped queries, which is most of what top-of-funnel B2B and SaaS content strategies are built around. The exact query types you've spent three years building topical authority on are frequently the ones most likely to get fully answered before the click ever happens.

None of that is a knock on the publishers named here. It's a description of the default state of AI-era analytics for almost every organic channel right now, including yours. The instrumentation most teams have simply was not built to separate AI Overview attribution from everything else moving the number, so it doesn't, and the gap gets read as ordinary volatility until someone finally runs the numbers a year later.

What to build instead of a hunch

You don't need new tooling to close this gap, and you don't need to wait for Google to hand you better data. Search Console's generative AI impression data already tells you more than most teams are pulling out of it. Most of what's missing is a matter of restructuring how you already look at that data, at the query level instead of the site level.

This is not a big lift. It's an afternoon of query segmentation and a habit of checking the split monthly instead of glancing at a total once a quarter. The output is not a nicer chart, it's a number leadership cannot wave away, because it's tied to specific queries instead of a trend line anyone can attribute to bad luck. If you want the fuller model for tying that number to revenue instead of just sessions, that's the same instrumentation gap we walk through in how AI traffic actually shows up in your analytics, and it's worth pairing with a proper reporting and analytics build so the split survives more than one quarter of someone remembering to check it.

Here's the honest reframe. You don't have Reddit's leverage. You don't have a $60-million-a-year check from Google, and you don't have a direct line to renegotiate anything when the numbers turn. But you also don't have Reddit's traffic base to burn through while you figure it out. A company the size of Business Insider can lose 85% of a channel and still be a going concern having a genuinely brutal year. Most sites can't take that kind of hit and still call it a bad year.

That's the actual argument for fixing your instrumentation now instead of waiting for the number to get undeniable. Not because the decline is guaranteed at your scale, but because right now most teams cannot tell the difference between a decline and a bad quarter, and that uncertainty is expensive in a way that never shows up as its own line item. Reddit had two years, a $60 million cushion, and a direct relationship with Google before this became a story it couldn't manage quietly. Build the query-level view this month, and you won't need any of the three.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.