Search & SEO

What Actually Ranks in 2026: My SEO Research Notes

July 12, 2026 · by Brendon Kennedy

Every year I do a pass on what actually moves organic rankings, mostly so I can throw out the folklore I’ve absorbed from SEO threads and re-read the primary documentation myself. This is that catalog for 2026 — what Google’s own docs say, what independent research measured, and where I think the honest uncertainty still is. These are research notes, not a playbook I’m selling.

”Helpful content” isn’t a separate system anymore

The biggest structural change is one people still get wrong. The standalone “helpful content system” — the classifier Google ran from 2022 — was folded into the core ranking systems with the March 2024 core update. There is no longer a single switch labeled “helpfulness”; it’s distributed across many signals inside core ranking.

What replaced the mystique is refreshingly plain in Google’s creating helpful content guidance: make “people-first content,” and self-assess with questions like whether a page “clearly demonstrate[s] first-hand expertise and a depth of knowledge.” The E-E-A-T framing (Experience, Expertise, Authoritativeness, Trustworthiness) is a lens quality raters use, and Google is explicit that “trust is most important” — the other three exist to build it. The “who, how, and why” test is the part I keep returning to: who made it, how (including whether AI involvement is disclosed), and why — with “why” being the question that separates content made for readers from content made for rankings.

None of that is a dial you can turn. It’s a description of what the system is trying to reward. I read it as: there’s no shortcut left that isn’t just “be the actually-useful result.”

Core Web Vitals: a confirmed signal, but a small one

This is where people over- and under-index at the same time. Google confirms page experience is real: the page experience documentation states plainly that “Core Web Vitals are used by our ranking systems,” while also warning “there is no single signal” and that great page experience mostly helps when content is otherwise comparable.

The change worth internalizing: Interaction to Next Paint (INP) replaced FID as a stable Core Web Vital. INP measures responsiveness across the whole visit, not just the first tap, and the thresholds are concrete — good is 200 ms or less, “needs improvement” runs to 500 ms, and anything above 500 ms is poor. LCP (loading) and CLS (layout shift) round out the three.

My honest read: CWV is a tiebreaker, not a growth lever. It won’t rank thin content, but a slow, janky page can lose a close race. I treat it as hygiene, not strategy.

Entities and structured data: understanding, not a ranking hack

Structured data is the area most oversold to me. Google’s structured data intro is careful: schema gives “explicit clues about the meaning of a page,” which Google uses “to understand the content of the page, as well as to gather information about the web and the world in general.” Crucially, it enables rich results — and it is not described anywhere as a direct ranking factor.

So the honest framing is: structured data doesn’t push you up the rankings; it makes you legible as an entity and eligible for richer result treatments that can lift click-through. In a world of machine-read results (below), being unambiguously understood is worth more than it used to be — but I’m careful not to call it a ranking cheat code.

The real 2026 shift: AI Overviews break the rank-to-traffic link

The thing that actually changed the game isn’t a ranking signal at all — it’s what now sits above the rankings. Pew Research Center tracked 900 U.S. adults and found that on searches showing an AI summary, users clicked a traditional result in just 8% of visits versus 15% without one — and clicked a link inside the summary only 1% of the time. About 18% of searches in that March 2025 sample produced an AI summary, and people ended their session more often when one appeared (26% vs 16%).

Correlational data always invites “but was it causal?” So the result I weight most is a randomized field experiment that actually removed AI Overviews for some users: it found they cut organic clicks by 38% on triggered queries, with zero-click searches rising from 54% to 72%. That’s an experiment, not a correlation, which is why it changes my thinking rather than just confirming it.

The implication for “what ranks” is uncomfortable but clarifying: ranking #1 and getting the traffic are now two different problems. Position still matters, but the click that used to follow it is increasingly intercepted at the top of the page.

What I’m actually doing

I run zsty, which ranks organically and buys no ads, so this isn’t abstract for me — it’s how the sites I care about live or die. My n=1 for 2026:

• Writing to the “why” test first — if a page only exists to rank, and wouldn’t help the person who actually lands on it, I cut it.
• Treating CWV as pass/fail hygiene, watching INP specifically since the FID-to-INP switch — checking it on the templates that carry the most traffic rather than chasing one global score.
• Marking up entities with schema for legibility, not expecting a ranking bump from it.
• Tracking impressions and answer-inclusion, not just position — because a #1 that gets summarized away is becoming a vanity metric, so the numbers I keep coming back to lean toward reach rather than rank.

The through-line for 2026: the ranking fundamentals got simpler and more honest (be genuinely helpful, be fast enough, be legible), while the payoff got harder (the results page increasingly answers without a click). I’d rather plan around that than keep pretending the old rank-equals-traffic math still holds.

← All writing