# How SEO Changed: From Keyword Stuffing in the 1990s to Getting Cited by AI in 2026, a 30-Year Timeline > From directory listings and meta keywords to PageRank, Panda and Penguin, BERT, AI Overviews and AI Mode. We map SEO's five eras on one timeline, separate what changed from what never did, and lay out the order to work in for AI search. | Field | Value | |-------|-------| | Published | 2026-09-04 | | Category | SEO | | Author | Aro Ogata | | Tags | SEO history, Google algorithm updates, AI Overviews, AI Mode, GEO, E-E-A-T, PageRank, AI search optimization | | URL | https://linksurge.jp/blog/en/seo-history-keywords-to-ai-2026/ | --- ## Key Takeaways: Five Eras, One Direction, From Rankings to Citations With AI search news landing every week, I did the opposite of chasing it. I put 30 years of SEO on a single timeline. What looks like an upheaval since 2024 turns out to be the latest step in a direction Google has been moving since the 1990s. Here's what the timeline shows. 1. **SEO has passed through five eras** — Directories and keywords (1994–2002), links (2003–2010), quality (2011–2014), intent and mobile (2015–2021), and AI citations (2022 onward). Each boundary is marked by an update that killed the previous era's winning tactic 2. **What changed is who judges you and what counts as a win** — The judge moved from human directory editors to link math, machine learning and language models. The win moved from "rank #1 and get the click" to "get cited or mentioned in the AI's answer" 3. **Three things haven't changed in 30 years** — Answer the user's actual question, make your pages machine-readable, and prove you're a trustworthy source. Every time a shortcut got closed, these three gained weight 4. **Where we stand in 2026 is measurable** — US zero-click rate 68.01% (SparkToro, Jan–Apr 2026), AI Overviews on roughly 43–48% of searches (Similarweb and Semrush, July 2026), AI Overviews on commercial queries up 71% in six months (Semrush). 54.3% of Gen Z use generative AI when shopping, yet trust search engines most (Rakuten, Sept 2026) 5. **Old techniques aren't discarded, they're stacked** — Crawlable structure, mobile readiness and E-E-A-T proof are still prerequisites in AI search. The 2026 work is adding citation-friendly structure and AI-side measurement on top Let's walk through each era, what it rewarded, what killed it, and then the order to work in now. --- ## 1. 1994–2002: The Directory and Keyword Era ![30 years of SEO: five eras and the updates that defined them](/blog/images/seo-history-keywords-to-ai-2026/en-01.png) Honestly, SEO was simple back then. Search engines had almost no way to understand what a page was about, so they largely trusted the page's own description of itself. Whatever you declared, you got credit for. ### From hand-curated directories to full-text search In January 1994, two Stanford graduate students, Jerry Yang and David Filo, published "Jerry and David's Guide to the World Wide Web." It became Yahoo!, a directory of sites organized by hand into a hierarchy of categories. By September 1994 it listed more than 2,000 sites and handled 50,000 searches a day. "SEO" at the time meant submitting your site to the right category. Then full-text engines like AltaVista arrived and started indexing page bodies and meta tags directly. That's where keyword stuffing was born: pack the meta keywords tag with target terms, repeat them through the body, and sometimes hide extra copies in text the same color as the background. ### 1998: PageRank changed the game In 1998, Sergey Brin and Larry Page published "The Anatomy of a Large-Scale Hypertextual Web Search Engine." Instead of trusting what a page said about itself, PageRank used the web's link structure to compute how important a page was. Judging a page by how many others vouched for it became the center of SEO for the next two decades. As for the meta keywords tag, Google stated publicly in 2009 that it doesn't use it for ranking. The first era's winning move was officially dead. > References: > [History of Yahoo - Wikipedia](https://en.wikipedia.org/wiki/History_of_Yahoo) > [The Anatomy of a Large-Scale Hypertextual Web Search Engine - Stanford InfoLab](http://infolab.stanford.edu/pub/papers/google.pdf) > [Google does not use the keywords meta tag in web ranking - Google Search Central Blog](https://developers.google.com/search/blog/2009/09/google-does-not-use-keywords-meta-tag) --- ## 2. 2003–2010: The Link Era Once PageRank became the core signal, the battleground shifted to acquiring links. If I had to sum this era up in a sentence: links were currency. ### Florida showed that shortcuts get closed On November 16, 2003, Google rolled out what the industry later named the Florida update. Right before the holiday shopping season, sites relying on keyword stuffing, cloaking and doorway pages lost rankings en masse, and contemporary accounts describe small retailers and affiliates disappearing overnight. It's remembered as the first major algorithm update. ### Paid links and nofollow At the same time, the clear value of links industrialized link building: buying links, reciprocal link pages, mass directory submissions, blog comment spam. In January 2005, Google introduced the rel="nofollow" attribute as a comment spam countermeasure, giving sites a way to say "this link isn't an endorsement." Paid links and link farms nonetheless kept working into the early 2010s. When I started in this industry, backlink counts were still showing up in sales decks as if they were a KPI. As long as the number of links was rewarded, whoever could manufacture the most would win. Google went after that structural flaw head-on in the next era. Why satellite sites and aggressive link building are still dangerous today is covered in our guide to [satellite site SEO risks and safe link building](/blog/en/satellite-site-seo-risks-2026/). > References: > [Google's Florida update: 20 years since the SEO 'volcanic eruption' - Search Engine Land](https://searchengineland.com/google-florida-update-434540) > [Google Florida: The First Major Algorithm Update - Search Engine Journal](https://www.searchenginejournal.com/google-algorithm-history/florida-update/) > [Preventing comment spam - Official Google Blog (2005)](https://googleblog.blogspot.com/2005/01/preventing-comment-spam.html) --- ## 3. 2011–2014: The Quality Era Three updates between 2011 and 2013 ended the link era. This is where SEO started shifting from a contest of tactics to a contest of content and trust. | Update | When | Target | |:---|:---|:---| | Panda | February 2011 | Thin, duplicated, ad-heavy pages | | Penguin | April 24, 2012 | Manipulative link profiles (paid links, link farms) | | Hummingbird | 2013 | Core algorithm rewrite to understand query intent, not just words | ### Panda and Penguin ended "more is better" Panda demoted entire sites built on thin pages mass-produced around keywords. Content farms took the direct hit. Penguin, a year later, targeted the paid links and link farms accumulated during the link era. Assets from the "links are currency" years turned into liabilities overnight. ### Hummingbird brought intent into the equation Hummingbird in 2013 wasn't a penalty. It was an engine swap. Rather than matching "Tokyo kid-friendly restaurant" word by word, Google started interpreting the underlying need: I want to eat out in Tokyo with my kids. Here's what's interesting: this 2013 pivot is a straight line to BERT in 2019 and AI search after 2024. From word matching to intent understanding. SEO became a job about search intent from this point on. For the four types of search intent and how to analyze them, see [What Is Search Intent?](/blog/en/search-intent-seo-guide/). > References: > [A brief history of Google's algorithm updates - Yoast](https://yoast.com/google-algorithm-updates/) > [A Timeline of Google Algorithm Updates - BrightEdge](https://www.brightedge.com/blog/a-timeline-of-googles-algorithm-updates) --- ## 4. 2015–2021: The Intent and Mobile Era With quality standards set, Google began folding in the device, the experience and the language of each search. Machine learning and smartphones are the protagonists of this era. ### Mobile became the default On April 21, 2015, Google expanded mobile-friendliness as a ranking signal, an update the industry dubbed Mobilegeddon. On March 26, 2018, mobile-first indexing officially began rolling out, meaning Googlebot crawls and evaluates the mobile version of a page as the primary one. The old sequence of building the desktop site first and adapting it for phones flipped. ### RankBrain and BERT put machine learning inside search In October 2015, Google confirmed RankBrain to Bloomberg: a machine learning system for interpreting queries it had never seen, described at the time as the third most important ranking factor after links and content. On October 22, 2019, BERT arrived, reading words in a query bidirectionally to grasp context. In my own rank tracking, the post-BERT shift was obvious. Results moved less on "did you include the target keyword" and more on "did you actually answer the question." ### Core Web Vitals turned experience into metrics The page experience update that started rolling out on June 16, 2021 added Core Web Vitals, LCP, FID (later replaced by INP) and CLS, as ranking signals. Load speed and layout stability went from nice-to-have to measured criteria. > References: > [Mobilegeddon - Wikipedia](https://en.wikipedia.org/wiki/Mobilegeddon) > [Google begins to roll out mobile-first indexing - TechCrunch](https://techcrunch.com/2018/03/26/google-begins-to-roll-out-its-mobile-first-index/) > [Google Turning Its Lucrative Web Search Over to AI Machines - Bloomberg](https://www.bloomberg.com/news/articles/2015-10-26/google-turning-its-lucrative-web-search-over-to-ai-machines) > [Timing for bringing page experience to Google Search - Google Search Central Blog](https://developers.google.com/search/blog/2020/11/timing-for-page-experience) --- ## 5. 2022–2024: The AI Citation Era, Part One. "Is It Helpful?" and AI Enters the Results Page Now that we've covered the foundations, let's take the fifth era, AI citations, in two parts. Between 2022 and 2024, Google started asking who content was written for, and generative AI got built into the results page itself. ### The Helpful Content Update and E-E-A-T On August 18, 2022, Google launched the Helpful Content Update to reward content written for people rather than for search engines. In December 2022, the Search Quality Rater Guidelines added "Experience" to E-A-T, making it E-E-A-T. First-hand experience, having actually used, tested or lived the thing, was now written into the rubric. The March 2024 core update went further. Google announced it had reduced low-quality, unoriginal content in results by 45% (up from an expected 40%), alongside new spam policies against scaled content abuse, site reputation abuse and expired domain abuse. ### AI Overviews launch, and ChatGPT enters search On May 14, 2024, at Google I/O, AI Overviews rolled out in the US, reaching Japan on August 15, 2024. An AI-generated summary now sits at the top of the results page with links to its cited sources. A new layer appeared above "rank #1": being cited in the summary. On October 31, 2024, OpenAI launched ChatGPT search. Together with Perplexity, founded in August 2022, "search and generate an answer" became mainstream outside Google, and SEO's target stopped being a single results page. For more on E-E-A-T and primary sources in AI search, see [how AI search is reshaping SEO](/blog/en/ai-search-seo-japan-2026/). > References: > [Our latest update to the quality rater guidelines: E-A-T gets an extra E for Experience - Google Search Central Blog](https://developers.google.com/search/blog/2022/12/google-raters-guidelines-e-e-a-t) > [Google Search: New updates to address spam and low-quality results - Google](https://blog.google/products-and-platforms/products/search/google-search-update-march-2024/) > [Google I/O 2024: New generative AI experiences in Search - Google](https://blog.google/products-and-platforms/products/search/generative-ai-google-search-may-2024/) > [Introducing ChatGPT search - OpenAI](https://openai.com/index/introducing-chatgpt-search/) --- {{linksurge-card}} ## 6. 2025–2026: The AI Citation Era, Part Two. Where We Stand Today ![Five eras compared: who judges, how you win, what you measure](/blog/images/seo-history-keywords-to-ai-2026/en-02.png) On May 20, 2025, Google opened AI Mode to all US users, with Japanese support following on September 9, 2025. It's a conversational search with no ten blue links. This is the part everyone calls "changing too fast," so let's anchor it in numbers published in 2026 rather than impressions. | Metric | Value | Source and period | |:---|:---|:---| | US zero-click rate | 68.01% | SparkToro and Similarweb, Jan–Apr 2026 (60.45% in 2024 on the same panel; the 58.5% published in 2024 used a different Datos panel and isn't directly comparable) | | AI Overviews share of searches | ~43–48% | Similarweb and Semrush, as of July 2026 (13–20% a year earlier) | | AI Overviews on commercial queries | +71% in six months | Semrush, Nov 2025–Apr 2026, 600K+ US keywords | | AI Overviews on transactional queries | −5% in six months | Same study | | Gen Z using generative AI when shopping | 54.3% | Rakuten Group, Jul–Aug 2026, n=637 | | Gen Z's most trusted source | Search engine, 42.1% | Same survey | ### A citation layer now sits on top of rankings More than two-thirds of searches end without a click, and nearly half show an AI summary. But on transactional queries, AI Overviews are actually shrinking, so the classic results page still owns the "buy" and "price" stage. As Rakuten's survey shows, users shortlist with AI and verify with search. Rankings didn't become irrelevant. A new layer formed on top of them. That's the accurate reading of where we are. ### Cited sources rotate within weeks The hard part of this era is volatility. Promptwatch data shows Reddit's share of ChatGPT Search citations fell 86.4% in four days in August 2026, cause unconfirmed. And Azoma's Q2 2026 analysis found ChatGPT's shopping citations were 41% earned media, 37% retailers, 19% UGC and 3% brand websites. Polishing your own site alone doesn't reach most of what gets cited. What struck me most this past year is that you can no longer talk about SEO without measuring the AI side. Your Google rankings can be steady while competitors take your mentions in ChatGPT and Perplexity, and you lose at the shortlisting stage without ever seeing it. [LinkSurge](https://linksurge.jp/)'s GEO monitoring records brand mentions and cited URLs across ChatGPT Search, Gemini, Google AI Overviews, Claude and Perplexity on a schedule, which is how I make that invisible loss visible. For the Gen Z shopping data in depth, see [54% of Gen Z Ask AI Before They Buy](/blog/en/gen-z-ai-shopping-search-trust-2026/). For how AI search works under the hood, see [What Is Query Fan-Out?](/blog/en/query-fan-out-guide-2026/). > References: > [Google launches AI Mode to all U.S. searchers with new features - Search Engine Land](https://searchengineland.com/google-ai-mode-us-searchers-455654) > [Google begins rolling out AI Mode in Japanese (Japanese) - Current Awareness Portal, National Diet Library](https://current.ndl.go.jp/car/257890) > [In 2026, Less than One Third of Google Searches Still Send a Click - SparkToro](https://sparktoro.com/blog/in-2026-less-than-one-third-of-google-searches-still-send-a-click/) > [About Half Of Google Searches Have AI Overviews - Search Engine Roundtable](https://www.seroundtable.com/half-google-searches-ai-overviews-41780.html) > [AI Overviews are expanding across commercial intent search [Study] - Semrush](https://www.semrush.com/blog/ai-overviews-commercial-search-study/) > [Rakuten Announces Results of Gen Z Generative AI Usage Survey (Japanese) - Rakuten Group](https://corp.rakuten.co.jp/news/press/2026/0904_01.html) > [Reddit's ChatGPT Search citations fell 86% in four days - Search Engine Land](https://searchengineland.com/reddit-chatgpt-search-citations-fall-report-485473) > [What Sources Does ChatGPT Cite for Products? Azoma Publishes the Data - GlobeNewswire](https://www.globenewswire.com/news-release/2026/08/26/3351631/0/en/what-sources-does-chatgpt-cite-for-products-azoma-publishes-the-data-and-explains-how-brands-get-recommended.html) --- ## 7. What Changed, What Didn't, and the Order to Work In ![What changed in 30 years, and what never did](/blog/images/seo-history-keywords-to-ai-2026/en-03.png) Let's turn the timeline into a plan. Looked at as a whole, the changes point one way. ### What changed | Dimension | 1990s | 2026 | |:---|:---|:---| | The judge | Directory editors, word matching | Language models interpreting intent and choosing sources | | What you optimize | Meta tags, on-page keywords | Answer-extractable structure, third-party mentions | | The win | Position in a list | Position plus citation or mention in AI summaries and answers | | The competition | Pages targeting the same keyword | Every citation candidate: review sites, UGC, retailers | | Measurement | Rank tracking | Rank tracking plus per-platform AI mention rate and cited sources | ### What didn't change Three things have stayed at the center of evaluation for 30 years and gained weight every time a shortcut closed. - **Answer the user's actual question** — Hummingbird's "intent," the Helpful Content Update's "for people," and AI answers' "extractability" are all restatements of this one point - **Make it machine-readable** — Crawlable HTML, heading hierarchy and structured data are today's version of the category submission from the directory era: the work of making machines understand you - **Prove you're a trustworthy source** — PageRank's endorsements from others, E-E-A-T's experience and authority, and AI citations' third-party mentions all measure the same question by different means ### The order to work in Old techniques don't get thrown out. They get built on. Here's the sequence I recommend. 1. **Audit the foundation** — Crawlable? Solid on mobile? Passing Core Web Vitals? Heading hierarchy and structured data in place? [LinkSurge](https://linksurge.jp/)'s AI Ready audit checks these plus llms.txt and AI-oriented semantic structure, item by item 2. **Check AI Overviews and cited sources for your key queries** — Is an AI summary appearing for your priority keywords, and if so, who's cited? The AI Overview analysis takes a keyword and your URL and returns whether an overview appeared, the cited URL list, and whether you're in it 3. **Track AI mentions per platform on a schedule** — Weekly mention rate and cited domains for shortlist-style queries across ChatGPT, Gemini, Perplexity and AI Overviews 4. **Work the third-party sources that get cited** — Review sites, comparison articles and UGC where competitors are cited and you aren't come first 5. **Put rankings and AI mentions in the same table** — Watching one without the other hides where the journey breaks --- ## Frequently Asked Questions ### What were the biggest turning points in SEO history? Three stand out. PageRank in 1998 made links from other sites the core signal. Panda, Penguin and Hummingbird in 2011–2013 shifted the game from volume and tactics to quality and intent. And the AI Overviews rollout in May 2024 added a citation layer above rankings. In each case, tactics that worked right up to the update stopped working right after. ### Do old SEO techniques still work? Keyword stuffing, meta keywords, paid links and link farms don't. Google stated in 2009 that it doesn't use meta keywords, and Penguin in 2012 targeted manipulative links. The technical foundations, though, still apply in AI search: crawlable structure, mobile readiness, heading hierarchy and internal linking. ### Do Google rankings still matter in the AI search era? Yes. Semrush's November 2025 to April 2026 study found AI Overviews on transactional queries actually fell 5%, so the classic results page still owns the purchase stage. Rakuten's 2026 survey found search engines remain Gen Z's most trusted source at 42.1%. Rankings didn't become irrelevant; an AI citation layer was added on top of them. ### How is GEO different from SEO? GEO, generative engine optimization, aims to get cited or mentioned in answers from ChatGPT, Perplexity, Google AI Overviews and similar systems. SEO targets position on a results page; GEO targets being one of the sources an AI consults when it composes an answer. The two aren't opposed. Crawlable structure and proof of trustworthiness are shared foundations. See our [beginner's guide to LLMO](/blog/en/llmo-guide-beginners-2026/) for more. ### What has stayed constant in SEO for 30 years? Three things: answering the user's real question, making pages machine-readable, and proving you're a trustworthy source. Directory category submissions, PageRank backlinks, E-E-A-T's experience and authority, and third-party mentions in AI citations all measure these same three by different means. Every time a shortcut closed, their weight increased. --- ## Conclusion: The Next Page of the Timeline Is Built by Stacking Thirty years of updates show a direction. From self-declaration to endorsement by others, from word matching to intent, from rankings to AI citations. Every era closed the previous era's loophole and kept the sites that actually answered people's questions. That's why the first job in AI search isn't chasing a new trick. It's auditing the foundation. Confirm crawlable structure and E-E-A-T proof, then find out whether AI Overviews appear for your key queries and who gets cited, then record AI mentions per platform. Work in that order and what you build survives the next update. LinkSurge's [AI Ready audit](https://linksurge.jp/ai-ready) and [GEO monitoring](https://linksurge.jp/geo) cover the foundation check and the AI-side tracking in one place. Write the next page of the timeline with numbers in front of you.