A working timeline · 1999 → 2026
Every major ranking and retrieval system, with what changed, how it compressed, and what survived. Curated, dated, sourced. The longest continuous SEO archive in Hebrew or English.
Before transformers, before quality classifiers, before user signals at scale. The web ranked itself with hyperlinks. Practitioners spent a decade learning to read graph centrality the way Google did.
Five major engines, no consensus ranking algorithm, no PageRank. Each one weighted text differently. The practice begins.
Citation-graph centrality becomes the dominant ranking signal. The link economy that defines the next decade begins here.
Black-hat tactics that had worked since 1999 stop working overnight. The first time SEOs collectively realise the platform has a position.
First professional certification in paid search. I take the exam in spring 2004 — among the first Israelis to do so. The certification remains in good standing twenty-two years later.
First structural admission that the link graph can be edited. Comment spam declines. Authority signals are now selective by default.
Images, video, news, maps integrated into the SERP. The page is no longer ten blue links. The visibility unit fragments for the first time.
The link-graph alone stops working. Quality classifiers, semantic understanding, mobile-first indexing — Google starts asking what a page means, not just who links to it.
Sitewide quality scoring goes into production. Content farms collapse. The first algorithm that operates at the site, not the page, level.
Decade-old link-building tactics get punished retroactively. PBNs die. The first algorithmic action with a recovery curve longer than the penalty itself.
First semantic ranking system at scale. Conversational queries become legible. The substrate that BERT will later perfect.
Live diary of the rollout from a Tel Aviv office at 03:00. The actual impact was smaller than the industry feared; the lesson was bigger than the impact.
First confirmed machine-learning ranking signal at Google. Tail queries get a different relevance model. Practitioners learn what "embedding" means.
Sites in health, finance, legal categories get rebaselined. E-A-T becomes a hard ranking constraint, not a soft guideline.
BERT, MUM, helpful-content. Four years where Google's underlying language model swallowed the surface-level rules. Practitioners who measured kept their leverage; those who chased keywords lost it.
First major transformer in production. Prepositions suddenly mattered. Conversational queries shifted by 38%. The Hummingbird thesis, six years on, fully realised.
The most volatile core update on record. We use it as the canonical reference for what algorithmic noise looks like under load.
Officially announced, partially deployed. The first signal that retrieval would not stay text-only. Foreshadows the 2024 generative shift.
Sitewide downranking with a multiplicative effect. Resamples on a 60–90 day cadence. The most consequential algorithm of the decade for practitioners.
Spam classifier moves entity-level. Co-occurrence patterns become the signal. PBNs die quietly. Pattern detection scales beyond per-link analysis.
The substrate shifts entirely. Ranking becomes retrieval. Retrieval becomes generation. The unit of competition moves from the page to the claim. Algoholic flagship launches into this transition.
First Google I/O confirmation of generative search. The framing as "experimental" turns out to mask a deep architectural rewrite.
Generative summaries default-on for US users. The first time a Google product surfaces a claim, not a page, as the user-facing artifact.
Statement-level retrieval scored against an internal "trustworthy summary" head, not against page authority. The decoupling becomes irreversible.
Twenty-six years of practice meets the first measurement framework that actually models the substrate. Algoholic Vol. III publishes.