SEO has died nine times. Is the tenth different?
by Cătălin Popa · updated 2026-09-10
The term “search engine optimization” was coined in 1997. The first declaration that it was dead also appeared in 1997.
Histories of the field attribute it to Richard Hoy, who called search engines a technology with no future. We were not able to reach the original text, so we give it as an attribution, not a verified fact. But the date matters: the discipline was declared dead in the same year it was given a name.
The nine deaths already over
Each had a real threat behind it. And each ended the same way: the threat became part of the job.
- 1997
The term SEO is coined. In the same year, someone declares it dead.
absorbed: nothing, yet
- 1998
PageRank makes stuffing keywords into meta tags useless.
absorbed: link building
- 2003
The Florida update wipes out thousands of sites overnight.
absorbed: content, instead of tricks
- 2005
A big name in the field declares that engines self-correct, so there is no point any more.
absorbed: the technical side
- 2010
After Caffeine, people write that SEO no longer matters.
absorbed: freshness and speed
- 2011
Panda cuts down the content farms. Forbes runs the headline that SEO is dead.
absorbed: text quality
- 2012
Penguin destroys link schemes.
absorbed: earned links
- 2015
RankBrain brings machine learning into ranking. AI is announced as SEO's replacement.
absorbed: semantics and entities
- 2022
The Helpful Content Update, so no more text written for engines.
absorbed: declared experience
This is not survival, it is appropriation. Every time, whatever was going to kill SEO became a chapter of SEO. And the discipline came out larger, not smaller: the services market was 81 billion dollars in 2024 and 108 billion in 2026.
The pattern nobody names
Every announcement of death rode a number that had lost its context along the way.
There are two cases we can trace back to source, twenty years apart, and they match down to the detail.
The voice case, 2014 to 2020
“Fifty percent of searches will be voice by 2020.” It was repeated for years and attributed to Comscore.
Comscore never published that figure. The original is a statement by Andrew Ng, then at Baidu, given to Fast Company in September 2014. He said that within five years at least half of all searches would be through images or through speech, and the context was the Chinese market. It was picked up in Mary Meeker's 2016 Internet Trends report, where the year 2020 got attached to it.
Along the way it lost the images half, lost the market it was talking about, and acquired a source it never had. An entire sub-industry of voice optimisation was built on that number. It did not happen.
The zero-click case, 2024 to today
“Zero-click searches grew from 56% to 69%.” It is everywhere now, including in the materials selling GEO.
The figure is real and it is Similarweb's, measured between May 2024 and May 2025. But it is measured on news-related searches, not on all searches. That qualifier disappears almost every time the number is quoted.
We are not saying the figure is false. We are saying it has already started circulating without its domain of validity, exactly as ten years ago. The difference between “69% of news searches” and “69% of searches” is the difference between an observation and a panic.
What is genuinely structurally different now
It would be comfortable to stop here, with the conclusion that people are exaggerating again. It would also be wrong. There is a serious argument that this time is different, only it is not the one usually made.
The usual argument is that AI brings less traffic. The real argument is a different one, and it surfaces when you put two measurements side by side.
How much the assistants send back
Conductor measured across 13,770 domains and 3.3 billion sessions, between May and September 2025. Traffic arriving from AI assistants was 1.08% of all traffic. Almost nothing.
How much is lost along the way
Over the same period, organic traffic to publishers fell from a peak of over 2.3 billion monthly visits to under 1.7 billion.
Put together, the two say something neither says on its own.
Every disruption until now has moved the click. To links, to social, to mobile, to the snippets at the top of the page. You could follow it to the new place. This one deletes it: it is absorbed by the answer on Google and barely reappears at the destination at all.
And that is no longer a change of technique, it is a change in the economic contract. Until now, SEO absorbed techniques. It has never absorbed a contract.
The test, with thresholds written now
We do not know which version is true and we will not pretend we do. But the difference between moving and deleting can be measured, and a threshold written in advance is the only kind that can be wrong.
- 1The share of traffic from AI assistants.
Baseline: 1.08%, on the Conductor set. If it passes 5% and holds for twelve months, the click moved, and SEO is absorbing again. If it is still under 2% in September 2027, it moved nowhere.
- 2Organic traffic to publishers.
Baseline: under 1.7 billion monthly visits, down from a peak of 2.3. If it recovers above 2 billion, the loss was an adjustment. If it settles under 1.5, it was a break.
- 3Zero-click, measured on the same segment.
Baseline: 69% on news searches. What matters is whether it plateaus or keeps climbing, and whether the same measurement ever appears for commercial searches. Without the declared segment, the figure cannot be compared with anything.
The thresholds above are not certainties. They are simply written in advance, which means we cannot move them once we know the answer. Exactly what we ask of any supplier, ourselves included.
So did it have another life or not
We do not know. What we do know is that nine times in a row the answer was “yes, and it came out larger”, and nine times in a row the number announcing the opposite had lost its context.
We also know the current argument is better than the ones before it. Not because it is repeated louder, but because it touches something else: not the technique for reaching a person, but the exchange that paid you for doing so.
If the pattern holds, the discipline will absorb generative engines too, and in five years GEO will be a chapter of SEO the way link building is a chapter of 1998. If it does not hold, it is the first time the threat was not a technology but the payment model. A cat can fall from a height many times. It cannot fall into water once.
Where the numbers come from
- 1.08% traffic from AI. Conductor's AEO/GEO benchmark report, 13,770 domains, 3.3 billion sessions, May to September 2025. The methodology is published, which is why we used the figure.
- Zero-click and publisher traffic. Similarweb, May 2024 to May 2025. The declared segment is news-related searches, and we declare it too.
- The voice figure. Andrew Ng's Fast Company interview, September 2014, and the 2016 Internet Trends report. The attribution to Comscore has no source, because there is none.
- The timeline of deaths. From published histories of the field. The year 1997 and the name attached to it are an attribution, not a primary document, and the text says so.
None of the figures above were measured by us. We used them because the article is about how figures travel, so each one comes with its period, its sample and its segment. When we have our own measurements on the Romanian market, they will appear separately and marked as such.
Share this article
Further reading
- What to ask before you hire a GEO expertThe first question on the list is where the number comes from
- The ad repository ChatGPT has to publishThe first time part of the paid layer becomes publicly verifiable
- All the questions explainedIn the order it makes sense to read them
- The measurements we publish ourselvesWith methodology, by city and sector, so they can be checked