A History of Google Algorithm Updates and What Each One Changed
Read as a list of names and dates this is a reference nobody needs. Read in sequence it is a single argument, because every major update has moved the same way. That direction is the reason this history is worth your time.
They All Point The Same Way
Every major update Google has made has moved in one direction, towards assessing whether a page genuinely serves the person who searched. Twenty years of names, nicknames and industry panic describe a single argument being made repeatedly, in slightly better ways each time.
Why that framing matters. It changes what you do with the list.
A chronology invites you to work out which update hit you. A direction invites you to work out whether your site is on the right side of it, which is the more useful exercise by some distance.
What has never happened. A reversal.
No update has ever moved back towards rewarding volume over substance, keywords over meaning or technique over usefulness. Not once, across the whole record set out below.
What is deliberately not here. A timeline graphic.
There is no date rail and no visual sequence, because the order is the least interesting thing about this material. The dates are attributed in the copy where Google has published them.
The Early History
Six updates sit together here rather than getting a section each, because their significance is in establishing the pattern rather than in anything you would act on today. They are worth a sentence apiece and no more.
Florida, 2003. The first one that frightened everybody.
Widely reported to have devalued crude keyword repetition, which had been enough to rank until then. It arrived just before Christmas, which is why it is still remembered by anybody who was selling online at the time.
Caffeine, 2010. Infrastructure rather than judgement.
Google announced this as a rebuild of how the web was indexed, producing much fresher results. It changed how quickly things could appear rather than what was considered good, which makes it the odd one out in this list.
Venice, 2012. Where local began to matter everywhere.
Recorded in Google's own search quality notes for that period, it began showing local results for searches that carried no place name. For a service business that was the beginning of location mattering without anybody typing a town.
Payday, 2013. The heavily spammed searches.
Confirmed by Google as addressing queries that attracted the most aggressive manipulation. It is the ancestor of the spam work that continues as recurring updates today.
Possum, 2016. An industry name, never confirmed as such.
Practitioners observed changes to how local results were filtered and grouped, particularly by proximity, then named it themselves. Google never described an update in those terms, which is worth knowing before treating any account of it as authoritative.
Fred, 2017. A joke name that stuck.
Sites carrying heavy advertising with little substance were widely reported to have lost ground. The name came from a passing remark by a Google engineer rather than from any announcement. No detailed description was ever published.
Panda
Panda is where content quality started being judged at the level of a whole site rather than page by page. Google's ranking systems documentation records it as announced in 2011 with that nickname and as becoming part of the core ranking systems in 2015.
What it changed about the judgement. The unit of assessment.
Before it, a good page on a weak site could do perfectly well. After it, the character of the whole site began to affect how its individual pages were treated. That principle has never gone away.
What it addressed. Thin material at scale.
Sites built from large volumes of shallow pages, much of it rewritten from elsewhere, produced to occupy searches rather than to answer them. The industry had been doing this deliberately for years.
Why it still matters. The thin content argument descends from it.
Every conversation about whether to publish two hundred more articles traces back to this update. Our material on risky practices covers where that approach still gets sold.
What being folded into core means. There is nothing to wait for.
While Panda ran separately, sites waited for the next refresh. Once it became part of the core systems, the standard simply applied, which is the same path the helpful content system later took.
Penguin
Penguin is the ancestor of the current position on links. Google's documentation records it as announced in 2012 with that nickname to combat link spam and as integrated into the core ranking systems in 2016.
What it changed about the judgement. Links stopped being a count.
Volume had been treated as strength, so the industry bought it. Penguin began discounting the manipulated variety, which meant a link was worth something only if somebody had chosen to give it.
What it addressed. Bought and built links.
Directories nobody read, paid placements, exchanges and networks constructed for the purpose. All of it had been standard practice and much of it was still being sold years afterwards.
Why the 2016 change was significant. Assessment became continuous.
Rather than periodic refreshes, the judgement applied as pages were reassessed. That removed the waiting period businesses used to sit through after cleaning up.
What survives today. The whole position on link buying.
Everything the industry says about earned against acquired references starts here. Our risky practices material covers what is still being sold and what happens when it stops working.
Hummingbird
Hummingbird is where the shift from matching words to understanding meaning begins. Google's documentation describes it as a major improvement to the overall ranking systems, made in August 2013, since when those systems have carried on evolving.
What it changed about the judgement. The question rather than the phrase.
Results began to reflect what somebody was asking rather than which page contained the same words. That is the single most consequential change in this entire history.
What it made obsolete. Writing for the phrase.
Repeating an exact phrase stopped being the route to ranking for it. Pages could rank for wordings they never contained, provided they answered the underlying question.
Why businesses still get this wrong. The old advice persists.
Instructions to use an exact keyword a set number of times are descended from the era this update ended, over a decade ago. They are still being given.
Where the semantic direction leads. Everything since.
Language understanding, entity recognition and the systems behind current results all sit on this foundation. Our advanced material covers where that has arrived.
Pigeon
Pigeon brought local results closer to the main ranking systems. Google confirmed a change to local search in 2014 without giving it that name, which was coined by a search publication and then stuck.
What it changed about the judgement. Local stopped being separate.
Local results began drawing on the same signals as everything else, so the strength of a website started mattering to whether a business appeared in local results rather than only its listing.
Why that was significant for small businesses. The website began to count.
A trade business could no longer rely on a listing alone. The site behind it became part of the assessment, which is where most of our clients' work sits.
What it looks like now. Two connected systems.
Local and organic results still behave differently and they are no longer independent. Our local material covers how that works in practice for a business serving a town or a county.
Why the name is worth flagging. Attribution matters.
Plenty of confident advice about this update describes something Google never published. Knowing the name came from a publication tells you how much weight to give any particular account of it.
Medic
This is the update that made expertise visibly matter for consequential subjects. Google described a broad core update in August 2018. The industry named it Medic because health and medical sites moved most noticeably. The standard it exposed applies far more widely than that.
What it changed about the judgement. Who is behind the page.
For subjects affecting somebody's health, money or safety, the credibility of the source became part of the assessment. A well written page from nobody in particular stopped being enough.
Which businesses this reaches. More than medicine.
Anything where bad information causes real harm. Legal advice, financial services, insurance, gas work, electrical work, structural work and childcare all sit inside the same expectation.
What it asks for in practice. Visible credentials.
Named people with relevant qualifications, registrations where a trade requires them and content written by somebody who does the work. Our on-page material covers how that is shown rather than asserted.
The current position. The event has gone and the standard has not.
There is no Medic update to be hit by. What persists is the expectation of demonstrable expertise, which every core update since has continued to apply.
BERT
BERT is where language understanding took a substantial step forward. Google announced it in October 2019 as a way of better interpreting how words work together in a search, particularly the small ones that change a meaning entirely.
What it changed about the judgement. Prepositions started counting.
The difference between travelling to a country and travelling from it began to be understood. So did the difference between doing something for somebody and doing it to them. Results shifted accordingly on searches phrased conversationally.
What it meant for writing. Natural phrasing stopped being a risk.
Pages written the way people speak became more findable rather than less. The awkward keyword grammar the industry had taught became a liability instead of a technique.
The boundary with our other material, stated deliberately. Two treatments on purpose.
This entry covers BERT as a dated event in the sequence. Our advanced material covers it as a machine learning system and what it means for how content is understood. That duplication is intentional and neither version should be merged into the other.
Why it belongs in this argument. Same direction again.
Understanding a question better means rewarding pages that answer it properly. Every language improvement moves the assessment further towards substance and further away from phrasing.
Page Experience
This is the entry most likely to be described out of date elsewhere, because the branding has changed and one of its component measures has been replaced. Google began rolling the page experience position out from 2021, built around a small set of measures covering loading, responsiveness and visual stability.
What it changed about the judgement. How a page behaves counted.
For the first time, the experience of using a page became part of the assessment rather than only what the page said. It was described from the outset as a consideration between comparable results rather than a route past better content.
What has been superseded. One of the three measures.
Google announced that Interaction to Next Paint replaced First Input Delay as a Core Web Vital effective March 2024. It also stated that Search Console would stop reporting the older measure. Any guide still describing that original set is describing something retired.
What else has changed. The reporting.
The standalone page experience report has been removed from Search Console. The measures themselves are still reported, so the change was to the dashboard rather than to the substance.
The current position, stated plainly. Necessary and not sufficient.
Google's published guidance is that good scores do not guarantee good rankings and that page experience involves more than these measures alone. Treat them as hygiene rather than as a lever.
Why this entry carries a warning. Confident nonsense circulates.
Claims about tightened thresholds and new ranking weight appear regularly without appearing in Google's documentation. We record only what Google has published, which is why this entry is shorter on specifics than others you will read.
Helpful Content
This is the most consequential recent entry and the one with the most important change to record. Google's ranking systems documentation states that it was announced in 2022, designed to ensure people see original helpful content written for people. The same document records that in March 2024 it became part of the core ranking systems.
What it changed about the judgement. Purpose became the test.
The question shifted to whether material was made primarily to serve somebody or primarily to gain search traffic. That is a question about intent rather than about execution. It cannot be met by writing the same thing more carefully.
What it inherited from Panda. Site level assessment.
A body of unhelpful material affects how a site is judged generally rather than only the pages concerned. Eleven years apart, the same principle applied more sharply.
Why the 2024 change matters to you. There is no separate recovery.
While it ran separately, businesses waited for the next run of it. Now it is part of the core assessment, so recovering from this and recovering from a core update are the same piece of work.
Where the evergreen version lives. Its own page.
This entry records what happened and when. What Google means by helpful content covers what the standard actually asks of a page, which has not dated and is the more useful read.
Product Reviews
A series of updates from 2021 onwards addressed review content specifically, rewarding first hand evaluation over restated manufacturer copy. Google has since stated that the reviews system is updated on an ongoing basis rather than through announced updates.
What it changed about the judgement. Evidence of having used the thing.
A review needed to show that somebody had handled the product. Measurements, photographs of the actual item, comparisons drawn from use and an account of where it falls short.
What it addressed. Reviews written from the specification sheet.
Pages describing products nobody at the publisher had touched, assembled from the manufacturer's own words and the reviews already ranking. An enormous category at the time.
Why it matters outside retail. Same test, wider application.
Any page recommending anything faces the same question, which is whether the recommendation comes from experience. Our on-page material covers product and service pages against that standard.
The current position. Not something to wait for.
Since the system runs continuously without announcements, there is no reviews update to watch for. The standard applies whenever a page is reassessed.
Spam Updates
These are the only entries here that repeat. Google announces spam updates periodically on its Search Status Dashboard, most recently including one recorded as beginning in March 2026. Each one improves detection against its published spam policies.
What they change about the judgement. Enforcement rather than standards.
A spam update does not raise the bar for everybody. It improves how reliably specific behaviour is identified, which is why they affect a narrow set of sites sharply rather than everybody a little.
How they differ from a core update. They name the behaviour.
A spam announcement describes what it addresses. A core announcement names nothing, because there is nothing specific to name. That difference in wording is the quickest way to tell which you are reading about.
Why we do not call this a punishment. Because the language matters.
Detection improving is not the same as an action being taken against a particular site. Conflating the two sends businesses looking for an appeal that does not exist. Our risky practices material covers the case where something genuinely has been applied to a site.
What The Sequence Tells You
A business trying to anticipate the next update is asking the wrong question. The specifics are unknowable and the direction has been consistent for over a decade, which between them tell you exactly where to put your effort.
What the whole record adds up to. One instruction, restated.
Be genuinely the best answer for the searches you want. Panda, Penguin, Hummingbird, Medic, BERT and the helpful content system are all versions of Google getting better at identifying that.
What the pattern of folding into core says. Standards become permanent.
Panda in 2015, Penguin in 2016, the helpful content system in 2024. Systems that prove their worth stop being events and become part of the assessment, which means nothing here ever gets relaxed.
What this means for the sites that get moved. They were exposed already.
Positions built on volume, on manipulated references or on being the least bad option were always dependent on the assessment not improving. Improvement is the one thing the record guarantees.
Where this argument continues. The future-proofing page.
Preparing your website for future updates takes the direction described here and sets out what building towards it involves. This page is the evidence for that page.
Keeping This Current
A history page carries a particular obligation, which is that parts of it stop being true without anybody touching the page. So this one has a named owner and a review schedule rather than a publication date and good intentions.
Who owns it. Andrew Odgers, Managing Director.
One named person is accountable for whether the statements here are still accurate, rather than the page belonging to nobody in the way most published histories do.
How often it is reviewed. Every quarter and after any named rollout.
Last reviewed in August 2026. Each review checks the status of every system described here against Google's current documentation rather than against what this page said last time.
What the review looks for. Superseded positions.
Systems folded into core ranking, measures replaced, reporting retired and anything described here as current that has quietly stopped being so. Those changes are what make old guides on this subject actively misleading.
What we will not do. Refresh the date and change nothing.
Updating a publication date to look current is the sort of thing the standards described on this page exist to catch. If the substance has not changed, the review date says so and the content stays as it is.
Every update
has moved the
same way.
Build that way.
Nobody can tell you what the next rollout addresses. The record can tell you where it will be pointing, because nothing in this history has ever moved back towards rewarding volume over substance. That is enough to build against without guessing.
What the record consistently rewards:
If your site already reads that way, the next rollout is somebody else's problem.
Every guide.
One position.
How updates work, the history of what each one changed, whether an update caused your drop, what recovery genuinely involves, what helpful content means now and how to build a site that rollouts leave alone.
