By Nersing Pradhan

In February 2011, thousands of website owners woke up, opened their analytics, and found that most of their traffic had gone overnight.

They hadn’t done anything wrong that morning. They’d done something wrong for years, and Google had finally built a system that could see it.

That update was called Panda. It was the first time Google made it obvious that the rules could change without warning, that no amount of traffic was permanent, and that a business built entirely on search could be unbuilt by a single deployment.

There have been hundreds of updates since – Search Engine Journal keeps the running list, and so does Search Engine Land. Most people treat those lists as trivia: names to memorise, dates to recite. That is a waste. Read in order, the history of Google’s algorithm is the only honest rulebook anyone has ever published for building on the web, and it is written in reverse. Every named update is Google closing a door that somebody was walking through.

Learn what was behind each door and you stop guessing about what works. You start seeing the pattern, and the pattern has held for twenty-three years.

Why a beginner should care about any of this

If you’re about to start a blog or market a business online, you might reasonably ask why ancient history matters.

Here’s why. Most of the advice you’ll find was written during one of these eras and never updated. When someone tells you to write 500-word posts stuffed with your keyword, they’re teaching 2009. When someone tells you to buy a hundred backlinks, they’re teaching 2011. When someone tells you to publish 300 AI articles and wait, they’re teaching a strategy Google explicitly outlawed in March 2024.

You can’t tell good advice from expired advice unless you know roughly when each idea stopped working.

Era one: the wild west (1998-2010)

Google launched in 1998 on one genuinely new idea. Instead of judging a page by what it said about itself, judge it by how many other pages linked to it. That was PageRank, and it made Google better than everything else overnight.

It also created the entire SEO industry, because the moment links became currency, people started manufacturing links.

The early updates were Google reacting to whatever people manufactured next. Florida, in November 2003, was the first one to genuinely hurt – it landed weeks before Christmas, gutted a lot of small retailers, and taught the industry that keyword stuffing and doorway pages had an expiry date. In 2005, Google introduced the nofollow attribute to stop comment spam. Jagger went after link networks the same year. Big Daddy rebuilt the crawling infrastructure. Caffeine, which finished rolling out in mid-2010, rebuilt the index itself so that fresh content could appear in results within minutes instead of days.

Through all of this, the game was still fundamentally winnable with brute force. Write more pages. Get more links. Repeat your keyword more times. It was crude, and it worked, and an enormous amount of the SEO advice still circulating today is a fossil from this period.

Era two: the great cleanup (2011-2015)

This is the era that made modern SEO.

Panda arrived in February 2011 and went after thin content and content farms – sites publishing thousands of shallow pages purely to catch search traffic. Google said it affected around 12% of queries, which in practice meant entire businesses evaporated in a week.

Penguin followed in April 2012 and went after the other half of the problem: manipulated links. Paid links, link networks, exact-match anchor text repeated a thousand times. If Panda was about what you published, Penguin was about what you built pointing at it.

Then Google started closing the smaller doors. The EMD update in September 2012 stopped exact-match domains from ranking on their name alone – which is why buying bestcheaplaptops.com is no longer a strategy. Hummingbird, in August 2013, was a rewrite of the core algorithm to understand what a query meant rather than which words it contained. Pigeon reworked local results in 2014. HTTPS became a lightweight ranking signal the same year. Mobilegeddon in April 2015 made mobile-friendliness a factor, which sounds obvious now and caused genuine panic then.

If you want the single lesson of this era: Google spent five years systematically removing every way of winning that did not involve being good.

Era three: the machines take over (2015-2020)

In October 2015, Google confirmed RankBrain – a machine learning system that helped interpret queries it had never seen before. It mattered less for what it did than for what it signalled. From here on, Google’s ranking systems would increasingly be things nobody could fully explain, including the people who built them.

Fred, in March 2017, hit sites that were mostly advertising with a bit of content wrapped around them. Possum reshuffled local results in 2016.

Then came the update that changed the language of the industry. In August 2018, a broad core update landed so hard on health, medical and finance sites that the community named it Medic. Google’s guidance pointed toward a concept from its Quality Rater Guidelines: E-A-T, meaning Expertise, Authoritativeness and Trustworthiness, applied most strictly to YMYL topics – Your Money or Your Life, the queries where bad information does real damage.

For the first time, the question wasn’t only what’s on the page. It was who wrote this, and why should anyone believe them.

BERT arrived in October 2019 and improved Google’s understanding of natural language – prepositions, word order, the small words that change meaning entirely. It affected around 10% of queries. Nobody could optimise for it, which was the point.

Era four: experience and trust (2020-2023)

Page Experience and Core Web Vitals rolled out through 2021, making load speed, visual stability and responsiveness into measurable ranking inputs. Small effect, big noise. A lot of people spent that year chasing green scores instead of writing anything.

The Product Reviews updates, starting April 2021 and running as a series through 2023, went after affiliate review content written by people who had clearly never touched the product. Specification tables copied from Amazon. “In-depth reviews” of things the writer had never opened. Google wanted evidence of actual use – photos, measurements, comparisons that could only come from having the thing in your hands.

Then, in August 2022, the Helpful Content Update. And in September 2023, a second one that did far more damage than the first.

The September 2023 helpful content update is the one that broke a generation of niche sites. Sites with years of history, real writers and reasonable rankings lost most of their traffic and, unusually, many never recovered through subsequent core updates. If you have heard bloggers talk about “the HCU apocalypse,” this is it. It is also the moment a lot of experienced people quietly concluded that a business standing on a single traffic source was not a business.

Quietly, in December 2022, Google had added a letter. E-A-T became E-E-A-T, with the new E standing for Experience. Not credentials – experience. Have you actually done the thing you are writing about?

That single letter is the most useful piece of guidance Google has published in a decade, and it is the reason first-hand content still holds up in 2026 while everything else erodes.

Era five: the AI era (2024-now)

March 2024 was the largest single change in years, and it came in two parts. Google’s own announcement is worth reading in full, and Search Engine Land’s breakdown covers what it meant in practice.

The core update absorbed the helpful content system directly into Google’s main ranking systems – no more standalone HCU announcements. Google said it expected to cut low-quality, unoriginal results by around 40%.

Alongside it came three new spam policies, and these matter enormously if you’re starting today:

• Scaled content abuse. Producing content at scale to manipulate rankings is a violation whether it’s written by AI, by humans, or by both. Google was explicit that the method doesn’t matter. The intent does.
• Site reputation abuse, better known as parasite SEO. Publishing third-party content on a trusted domain to borrow its ranking power. Enforced from May 2024.
• Expired domain abuse. Buying dead domains for their backlinks and repurposing them.

Read those three together and you have Google’s answer to the question everyone was asking in 2024: can I just point AI at a keyword list and print money? No.

Then, in May 2024, AI Overviews launched in the US, and the ground genuinely moved.

An AI Overview is the summary Google generates at the top of the results page. It answers the question directly, using content from pages that are, in many cases, no longer getting the click. By May 2025 Google had taken the feature to over 200 countries and more than 40 languages, claiming over 1.5 billion monthly users.

The traffic impact is real, and it’s worth getting the numbers right rather than repeating whichever one sounds most dramatic. Ahrefs studied 300,000 keywords and compared the same keywords before and after: for the number one result on keywords showing an AI Overview, average click-through rate fell from 7.3% in March 2024 to 2.6% in March 2025, which they calculated as roughly a 34.5% drop against what would otherwise have been expected. Seer Interactive, measuring differently across thousands of informational queries, found organic click-through on AI Overview queries falling 61%, from 1.76% to 0.61%.

Different methods, different numbers, same direction. When Google answers the question itself, fewer people click.

One honest footnote most articles leave out: Seer’s own 2026 update found that trend partially reversing, with organic CTR on AI Overview queries climbing back from a floor of about 1.3% in December 2025 to around 2.4% in February 2026. The floor may have been the bottom. Nobody knows yet.

The rest of the era has come fast. AI Mode launched as a Labs experiment in March 2025 and went to all US users at I/O in May 2025, expanding through 2025 into most major markets. Core updates arrived in a steadier stream than the old twice-a-year rhythm – March, June and December 2025, then February, March, May and June in 2026 alone.

The March 2026 core update was the loudest of them. Data from SE Ranking, reported by Search Engine Land, found nearly 80% of top-three results shifted, with about 24% of pages that had been ranking in the top ten dropping out of the top hundred entirely. The direction of travel was clear enough that several independent analysts noticed the same thing: in travel, in real estate, in health, visibility moved away from aggregators and broad discovery platforms toward brands, official sources and specialists. In almost every category, the source beat the summariser.

At I/O in May 2026, Google announced AI Mode had passed a billion monthly users and began rolling out search agents and a redesigned, conversational search box.

Worth knowing, because it saves you money: Google’s official guide to generative AI features, published in May 2026, states plainly that optimising for its AI surfaces is not a separate discipline. It is the same work, running on the same ranking systems. You do not need to buy an “AEO” or “GEO” course. The same guide notes that original, non-commodity, first-hand content is what performs in these features – which is the same advice, again, in different clothes.

The pattern, in five lines

Twenty-three years, hundreds of updates, and the same five things keep coming true.

Every shortcut has a shelf life. Keyword stuffing, link buying, exact-match domains, thin affiliate pages, scaled AI content – each worked brilliantly, then didn’t. If a tactic works because Google hasn’t noticed yet, you’re renting.

Google keeps moving toward the thing it can’t fake. Content quality, then expertise, then trust, then first-hand experience. Each step gets harder to counterfeit. That is deliberate.

Recovery is not guaranteed. The old assumption was that a core update could give back what a previous one took. After September 2023, many sites simply never came back. Plan as though a hit is permanent.

Single-channel businesses die at the whim of a single update. Ask anyone who lost 90% of their traffic in one deployment. The people who survived every era were the ones with a list, an audience, a brand – somewhere the traffic wasn’t Google’s to take away.

Being the source beats being the summary. Panda punished the sites summarising other people. So did the reviews updates. So did the September 2023 HCU. So, measurably, did March 2026. AI Overviews are just the newest and most efficient summariser, which means the only durable place to stand is where the summary comes from.

What this means if you’re starting now

Don’t try to reverse-engineer the current algorithm. It’s the wrong target, it changes every six weeks, and by the time you understand it, it has moved.

Aim at the direction of travel instead. Every update for over a decade has pushed toward the same place: original, first-hand, trustworthy work from a recognisable source. That is where Google was going in 2011, and it is where it is going now, only faster.

In practice, for a new site in 2026:

• Write things only you can write. Your own tests, your own numbers, your own photographs, your own mistakes. An AI Overview can summarise a fact. It cannot summarise your experience of running something for six months.
• Build the name, not just the pages. Google increasingly rewards recognisable brands, and so do humans. A search for your name should return something.
• Don’t stand on one leg. Search, video, social, email. If a single update can end your business, you don’t have a business yet – you have a position in a ranking, and positions get taken.
• Watch the updates, but don’t panic at them. Note the date, check whether your traffic moved, and act only if it did.

If you want to follow along in real time, the three places that log every confirmed update are Search Engine Journal, Search Engine Land and Search Engine Roundtable. Bookmark one. You do not need all three.

I don’t know what the next update will be called. Nobody does, including most of the people at Google.

But I can tell you with reasonable confidence what it’ll be closing: whichever shortcut is quietly working right now. That’s the only thing every one of these updates has had in common.

Build something that does not depend on the shortcut and the updates stop being events. They become weather.