SEO01 What we did (and why you should trust us, mostly)
Between July 2025 and July 2026, we built a tracking corpus designed to survive scrutiny. We selected 10,000 commercial keywords across 12 verticals — finance, health, SaaS, e-commerce, legal, travel, and six others we’ll call “the depressing ones.” Keywords were stratified by volume (head, mid, long-tail) and by intent (transactional, informational, that gray zone where nobody clicks anything).
We ran daily rank tracking on the top 100 organic results for each keyword, logged every confirmed Google update, recorded Search Console click/impression data from 240 anonymized participating domains, and then stared at the spreadsheet until meaning appeared. Meaning appeared reluctantly.
Rank tracking via a major third-party tool, cross-validated against raw SERP scrapes twice weekly. Traffic data contributed anonymously via Search Console API exports. Updates identified via Google’s public search status dashboard + two independent trackers. Volatility = standard deviation of rank position per keyword per week. We excluded branded keywords and SERPs dominated by image/video packs. The full dataset (anonymized) is available on request and a small bribe.
02 The volatility problem
The first thing the data screamed at us — and it did scream — is that ranking volatility is not an event, it is the default state. Across the full corpus, the median keyword changed position 4.2 times per week. Twenty-three percent of keywords moved by 5+ positions in a single week at least once during the year.
The practical implication: a ranking report is a photograph of a blur. Anyone who shows you a stable-looking rank tracker for a commercial keyword is either lying, looking at a branded term, or has refreshed it at the one calm moment of the week.
Stop treating daily rank movement as signal. A single day’s position tells you almost nothing. The trend over 4–8 weeks is the smallest unit that carries real meaning.
03 Where your traffic actually went
This is the part that hurts. We aggregated estimated click share — where a click “lands” on a typical commercial SERP — across the full corpus for the final quarter. The distribution is not flattering to anyone who owns a content site.
Yes, the AI Overview now absorbs the majority of estimated clicks on commercial queries in our set. But — and this is the genuinely useful part — the pages that are cited by the AI Overview share a remarkably consistent structure, which we’ll get to in a moment.
04 The author-headshot effect (we can’t explain this either)
This is where the data gets weird, and we want to be honest that we cannot fully explain it. The single strongest correlation we found — across every vertical, controlling for backlinks, content length, and domain authority — was between ranking stability and the age of the domain’s author bio photo.
Domains whose author headshots were taken before 2019 ranked 2.3× more stably across update windows. We have checked this four different ways. It holds. We have no explanation, and, in the spirit of intellectual honesty, we are not going to invent one. It is almost certainly a proxy for something else (domain age, publishing consistency, an old site that has accumulated trust signals over years). But the correlation is there, loud and confusing.
“We tried to control for content quality. Then we realized there was no content quality to control for.” — Lead Researcher, halfway through their notice period
05 What actually correlated (the useful part)
Stripping out the noise, four factors showed a consistent, repeatable positive correlation with ranking stability and traffic retention across the year. None of them are new. All of them are unfashionable. Here they are, ranked by effect size:
- Topical depth over breadth. Sites covering fewer topics with more pages per topic outperformed. The “one page per keyword” model actively hurt.
- Internal linking density. Median internal links per page correlated with stability more than any single off-page metric. The 240 sites in the top quartile averaged 11+ internal links per page.
- Faster Largest Contentful Paint. LCP under 2.0s showed a small but real stability edge. Under 2.5s is the practical threshold; beyond that, the penalty compounds.
- Freshness on commercial pages. Transactional pages updated within 90 days retained rank more reliably. Informational pages rewarded updates within 12 months.
06 What you should actually do
If you take one thing from 12,400 words and 38 charts, take this: the tactics that worked a decade ago are the tactics that correlate today, and the shiny objects (raw link velocity, content volume, EMD-style hacks) do not. The volatility is real and permanent. Your defense against it is structural, not tactical.
Concretely: pick fewer topics, cover them more thoroughly, wire them together with ruthless internal linking, keep your commercial pages technically fast and freshly updated, and build the kind of site that would still earn traffic if Google vanished tomorrow — because some days, for some sites, it effectively does.
Stop chasing the algorithm. Build the structural fundamentals. The algorithm is a mood ring; the strategy is the same one it’s always been.
— The SEO Bits Data Desk. Dataset available on request. Brad had nothing to do with this one.