Signal Through the Slop — Week 40, September 2026

AI search is changing how brands get found, recommended, and trusted. Each week, we pull signal from Reddit, LinkedIn, and industry research to track what's actually moving in the space. Here's what caught our attention this week.

What Reddit is saying this week

The story practitioners keep circling back to is the ChatGPT citation collapse, and this week it stopped being a Reddit-only story. Citations per 1,000 ChatGPT answers have fallen sharply since early August not just for Reddit, but for Facebook, YouTube, Forbes, and Wikipedia too, with the steepest single week landing in the last few days. That's a meaningfully different story than "Reddit got caught being gamed." It looks more like the model is narrowing which platforms it treats as trustworthy sources across the board, not punishing one community for bad behavior.

The more useful question buried under the headline is where those citations actually went. Tracking on ChatGPT's citation mix shows the vacated slots didn't move to other social platforms — they moved to brand-owned help centers and product documentation, which jumped from a sliver of citations to close to a third of them in one dataset. If your team's response to "we're losing AI citations" has been to double down on social and community presence, the data says that's probably the wrong fix. The right fix is closer to home: your own documentation, structured clearly, with one claim per section stated in a full sentence.

There's also a nuance worth sitting with before you panic about a citation drop on any single platform: retrieval and citation aren't the same thing. Reporting on ChatGPT's backend behavior suggests it's still pulling from Reddit during answer generation even where it's citing Reddit less often in the visible response. A platform's citation count falling doesn't necessarily mean it stopped mattering to the answer — it might just mean it stopped getting credit for it. Worth remembering before you pull spend or attention from a channel where your actual customers still hang out, just because a tracking dashboard shows a scary line going down.

If you're managing AI search visibility for a brand right now, the takeaway isn't "avoid third-party platforms." It's that any visibility built entirely on borrowed platforms — even ones that feel safe, like Wikipedia or a major publisher — is more fragile than visibility built on your own well-structured content. This week is the clearest evidence yet that the fragile kind gets reshuffled fast, and without warning.

What LinkedIn is saying this week

The long-running "SEO vs. AEO vs. GEO, do we really need three acronyms" argument finally started converging on something more useful than semantic fatigue. The framing gaining the most traction this week treats them as three surfaces built on the same underlying fundamentals — authority, clarity, real positioning — rather than three separate disciplines requiring three separate teams. It's not a new idea, but it's landing now because enough practitioners have lived through a year of acronym whiplash to want a simpler mental model, and one post making that case plainly picked up the highest engagement of anything found this week.

Underneath that convergence, a sharper and more technical conversation is happening about measurement. Google's new AI visibility reporting in Search Console shipped in late August, and practitioners are already pointing out its biggest limitation: it only measures impressions on pages you own. One breakdown circulating this week found that in a sample of real AI citations, roughly 97% came from pages a brand didn't control — listicles on other sites, PR placements, third-party mentions — meaning the new official reporting tool covers a sliver of what's actually happening. If you're building a measurement plan around Search Console's new report alone, you're measuring the part of the iceberg above the water.

There's a live disagreement worth knowing about too: whether tracking raw citation counts is even a good metric in the first place. One well-known SEO voice argued this week that trending citation counts up and to the right is misleading, because a single model update can swing the numbers on its own, and that being retrieved-but-not-visibly-cited still carries value that most tracking tools don't capture. It's a fair challenge to how a lot of GEO reporting gets built right now, and the practical answer is probably "keep the metric, but normalize for known model changes" rather than throwing it out.

Organizationally, AEO is starting to show up as a real, funded job requisition rather than a task someone tacks onto an existing SEO role. Several companies posted openings this week explicitly built around AI-answer visibility as a standalone function — vendor management, citation outcomes, the whole scope of a real job, not a side project. That's a genuine maturity signal: when a function gets its own headcount line, it's stopped being optional.

What the research shows this week

The most rigorous data point this week comes from a controlled field study, not an observational one: researchers put 1,100 Chrome users through a week of forced AI Mode usage and compared it against normal search. External website visits dropped nearly 19 percentage points. Reddit clicks specifically dropped over 21 points, news-site clicks over 12. And user satisfaction, sense of control, and perceived personalization all scored lower under AI Mode than regular search — more people gave negative feedback than positive. That's a real rebuttal to the idea that AI Mode is simply search working better; the people using it, at least in this study, didn't feel that way.

At the same time, Google's AI Overviews keep expanding their footprint. Third-party measurement now puts AI Overviews in close to 40% of U.S. desktop searches, up from roughly a quarter of searches a year earlier — a jump of well over ten points in twelve months. And separately, the share of AI Overviews that include at least one external link jumped from close to zero to more than a quarter of them, in a fairly short window.

Put those two findings side by side and there's a real tension worth sitting with. More citations sounds like good news for the open web. But a controlled study in the same window shows AI Mode usage measurably cuts the clicks that would follow those citations. The likeliest explanation nobody's fully confirmed yet: a growing share of those new links may be routing users into Google's own AI Mode rather than off Google entirely, which would mean citation counts rising while referral traffic doesn't follow. Anyone citing "AI Overviews are linking out more now" as proof that AI search is getting friendlier to the open web is telling only half the story until that gets resolved.

On the regulatory side, the UK's Competition and Markets Authority is pushing Google to put AI assistants directly on search choice screens for Android and Chrome, alongside traditional search engines, with a decision expected in early October. If it goes through, it's a real distribution advantage for LLM-based search products at the exact moment people set up a new device — the kind of on-ramp that's been unavailable to any AI search alternative so far, and one likely to draw similar pressure in other markets if it works.

That's the signal this week — back next Monday with more.