We've spent the last six weeks doing something most teams in this space talk about and never finish: manually scoring AI signals across an entire SERP corpus. 350 top-ranking URLs, 35 SERPs, eight signal dimensions, no detection tools just structural and lexical pattern recognition any reader could replicate.

The headline finding will sit uncomfortably with anyone selling "AI-free content" as a service: AI-generated content is already ranking in the top 10 for almost every category we tested. 42% of pages showed moderate-to-heavy AI signals. About 11% looked fully AI-generated with minimal human review. The correlation between AI signal and rank position came in at -0.18 a weak negative. Less AI-looking content ranks slightly better. Slightly. Not reliably.

But within that weak average, a much sharper pattern emerged. The top three is being won by something specific. And it's not "no AI."

Here's what we're tracking this week.

Google announced three bidding updates this week ahead of Marketing Live 2026. The most interesting is Journey-Aware Bidding a Smart Bidding beta that lets Search campaigns using Target CPA learn from both biddable AND non-biddable conversion stages across the full lead-to-sale journey. MQL, SQL, closed deal all reading into the bidding model without each having to be the bidding goal.

This shows up in an SEO/GEO newsletter for one reason: it confirms Google's direction. The same conversion infrastructure that feeds Smart Bidding is the infrastructure that will eventually feed AI search readiness measurement, attribution across AI-cited touchpoints, and the next generation of ranking signals on the organic side. Google is publicly committing to full-funnel optimization across every surface they own.

The teams that get their funnel data clean now offline conversions imported, conversion stages categorized correctly, CRM connected through Google Ads Data Manager will be ready to take advantage of Journey-Aware Bidding in paid AND the next set of signals Google ships on the organic side. The teams still optimizing toward form fills as their bidding goal will be invisible inside an increasingly journey-aware system.

One number worth holding onto from the announcement: Smart Bidding Exploration drove a 27% increase in unique converting users on average across Search campaigns. That isn't a paid mechanic. It's a hint at how much value is currently sitting outside the queries marketers consider efficient and a preview of how much the rest of Google's ecosystem is about to start optimizing the same way.

The single most useful pattern in the data: AI tolerance varies sharply by query type. Treating all queries the same is the most expensive mistake in the dataset.

The split most teams are missing:

Four patterns separated the top three from the rest of page one (present in 82% of top three URLs vs only 31% of positions 4-10):

The two numbers that matter most for your roadmap:

The jump from position 6 to position 2 isn't a polish improvement. It's a step change in operator voice and verifiable proof. AI-assisted content can get to page one. It usually cannot get to the top three. The teams winning the top three know which slot on which SERP tolerates which kind of content. Everyone else is optimizing the wrong thing about 70% of the time.

While we were finishing our SERP study, Cyrus Shepard at Zyppy published a meta-analysis of nearly every academic paper, controlled study, and case study on AI citations from the past two years. He extracted 22 ranking factors most associated with earning AI citations.

The headline he wrote and most people skipped over: most AI citation ranking factors align with traditional SEO, with a few tweaks.

This is the under-discussed truth of the GEO conversation. Cyrus's external meta-analysis and our own internal SERP study are converging on the same answer from opposite directions: there is no separate "GEO playbook." There is SEO with three layers added on top schema completeness, citable structure, and brand presence in the surfaces LLMs pull from.

If you've been running AI search optimization as a separate discipline with its own tools, dashboards, and headcount, that overhead is buying you very little. The teams doing this well in 2026 are running one motion that ships clean structured data, real authorial proof, and proprietary data — and watching that work pay off across Google rankings and AI citations simultaneously.

The most interesting comment buried in the thread on Cyrus's post: Artemii Gorbunov pointed out a finding most of the public research has under-explored — that AI models repeatedly revisit the same canonical company sources before citation even happens. The implication is that "which sources get cited" is downstream of a much earlier decision: which sources the model resolves as canonical in the first place. That's a brand and entity problem, not a content problem. The teams who internalize that distinction will move first.

Most AI visibility dashboards answer the comfortable question: "Where am I being cited?" The one that actually changes strategy is the inverse: "Which AI citations is my content driving without my brand getting any of the credit?"

These are ghost citations. Your URL gets pulled into AI answers 100+ times. Your brand name shows up zero times. The model used your content to recommend your competitor. We covered the mechanic in last week's research roundup, but the operational version is what matters: most teams have never measured their ghost citation rate, and it's almost always higher than they'd guess.

A pattern from a recent audit: a B2B SaaS client had 340+ ghost citations across ChatGPT and Perplexity in a single month. Their own content was doing the heavy lifting for AI to recommend three different competitors. The fix wasn't more content. It was tightening the entity signal schema completeness, brand mentions inside the content body, structured author bylines, and stronger third-party brand presence so the model resolved them as the canonical source on their own pages.

This is the same finding Artemii surfaced in Cyrus's comment thread, just measured from the operator side. If you're tracking AI citations without measuring ghost citation rate alongside, you're looking at the half of the picture that flatters you.