Each few years, the web optimization trade discovers a brand new option to mass-produce content material and convinces itself that this time it’ll work. That the sheer quantity of pages will overwhelm Google’s means to evaluate high quality. That if you happen to simply publish sufficient, the numbers will carry you.
It by no means works. It has by no means labored. And the individuals promoting you these approaches realize it has by no means labored. They only want it to work lengthy sufficient to gather the bill.
The Sample Has A Identify. It’s Known as “Not Studying”
Let’s stroll by way of the timeline, as a result of apparently, we have to do that once more.
2008-2011: Content material Spinning
The pitch was easy: Take one article, run it by way of software program that swaps synonyms, and out of the blue you might have 50 “distinctive” articles. The phrase “distinctive” was doing a variety of heavy lifting in that sentence. These articles learn like somebody had fed a dictionary by way of a blender. However even when the output had been polished, the premise was damaged. Right here’s what the content material spinners by no means grasped, and what their successors nonetheless don’t: Uniqueness is trivially straightforward to supply. A monkey dropping its palms on a keyboard produces distinctive content material. The string of characters has by no means existed earlier than – congratulations, it’s unique. The laborious half was by no means uniqueness. It was producing uniqueness that’s value one thing. Distinctive and invaluable will not be synonyms, and the hole between them is the place each scaling technique falls aside.
Google tolerated it for some time. Its programs merely hadn’t caught up but. Then Panda arrived in February 2011, hit practically 12% of all search queries, and content material farms watched their site visitors evaporate in a single day … I used to be “lucky” sufficient to look at it occur in actual time. Demand Media, the poster youngster of the content-farm mannequin, reported a $6.4 million loss the next yr.
The lesson was speculated to be clear: You can’t industrialize high quality. Quantity with out substance is a legal responsibility with an extended tail than most budgets can soak up.
2015-2022: Programmatic web optimization
The pitch developed. As a substitute of spinning present articles, you’d construct templates and fill them with structured knowledge. “Greatest [X] in [City]” pages, generated by the thousand, each a skinny wrapper round a database question. A few of these truly supplied worth – if the underlying knowledge was good and the template served real person wants. Most didn’t. Most have been simply doorway pages carrying a greater outfit. Google spent years refining its means to detect and demote templated content material that existed primarily for indexing functions reasonably than for people.
The lesson was speculated to be bolstered: scale works when there’s substance beneath. With out it, you’re simply constructing a much bigger goal.
2023-Current: AI-Generated Content material At Scale
And right here we’re once more. Similar pitch, shinier instruments. “We will produce 500 articles a month!” Fantastic. Are you able to produce 500 articles a month which are value studying? That comprise one thing a reader couldn’t get from the outcomes already within the index? That display any type of experience, expertise, or unique thought?
No? Then you definitely’re not scaling content material. You’re scaling your crawl finances waste.
And the sample recognition failures are beautiful. (This wasn’t delicate. A number of of us observed. No, we weren’t impressed.)
I not too long ago got here throughout an AI visibility device – one which sells itself on serving to you get found by AI programs – that had generated tons of of pages following the sample “greatest web optimization businesses in {metropolis}.” Déjà vu. Anybody who lived by way of programmatic web optimization acknowledges this instantly – it’s the 2017 playbook, besides now the copy is written by an LLM. The template bought a grammar improve and an “it’s AEO” stamp. The technique didn’t.
Lily Ray flagged the same case: a resume web site with 500+ programmatic pages for “resume examples for {profession}.” Each title following the very same system. Close to-identical web page templates. Misused AggregateRating schema. Apparent AI content material all through. Her abstract was three phrases: “Labored till it didn’t.”
That phrase needs to be tattooed on each content material scaling pitch deck. Labored till it didn’t. It at all times does. After which it doesn’t.
The irony of an AI optimization device utilizing mass-generated doorway pages to construct its personal visibility could be humorous if it weren’t so completely on-brand for this trade.
The Qualitative Wall Doesn’t Transfer
Right here’s what each era of content material scalers fails to grasp: Google doesn’t consider content material in isolation. It evaluates content material relative to all the things else within the index on the identical subject.
Publishing 500 AI-generated articles about mortgage charges doesn’t make you an authority on mortgage charges. It makes you the five hundredth supply saying the identical factor in barely completely different phrases. And Google already has 499 of these. It doesn’t want yours.
The qualitative wall is that this: There’s a minimal threshold of real worth – unique perception, lived expertise, particular experience, one thing the reader can’t get elsewhere – beneath which no quantity of quantity helps you. You may publish 1,000,000 pages beneath that threshold. You’ll rank for nothing that issues.
And it will get worse. For the individuals scaling AI content material particularly to realize visibility in AI-powered reply programs, the quantity technique doesn’t simply fail; it actively backfires. A 2025 paper on retrieval analysis for LLM-era programs introduces a metric that measures each useful and distracting passages in retrieval. The discovering that issues right here: Low-utility content material doesn’t sit quietly within the index ready to be ignored. It could actually pull retrieval fashions off-track, degrading the standard of solutions these programs produce. Your 500 skinny articles aren’t simply invisible. They’re noise. And in case your web site additionally has genuinely helpful pages buried in that noise, congratulations – you’ve constructed your individual interference sample. The quantity you thought would assist discovery is actively drowning the pages that may have earned it.
This isn’t a brand new perception. It’s the identical perception that content material spinners ignored in 2010, that programmatic web optimization factories ignored in 2018, and that AI content material mills are ignoring proper now. The instruments bought higher at producing textual content. The textual content nonetheless has nothing to say.
Google Instructed You. Repeatedly
Google’s spam insurance policies outline scaled content material abuse as producing pages “for the first objective of search rankings and never serving to customers.” They explicitly listing “utilizing generative AI instruments or different related instruments to generate many pages with out including worth for customers” for instance. This isn’t subtext. It’s textual content.
In June 2025, Google started issuing guide actions particularly for scaled content material abuse, concentrating on websites that had been mass-publishing AI-generated content material. Websites throughout the UK, US, and EU acquired Search Console notifications citing “aggressive spam strategies, resembling large-scale content material abuse.” Full visibility drops. Pages didn’t slide down the rankings; they vanished.
The August 2025 spam replace continued the enforcement. Subsequent core updates have saved tightening the screws. Every time, the identical profile will get hit: excessive quantity, low substance, no editorial oversight.
And every time, the affected web site house owners acted stunned. As if Google hadn’t been telling them this for 15 years.
‘However Our Content material Is Rating Properly’
That is my favourite delusion. I’ve seen it at each stage of this cycle. “Our AI content material is rating, so it should be nice.” Claiming “that is rating nicely” is commonly exactly why Google points algorithmic enhancements and guide actions to your web site. In case your low-value content material is rating, the system hasn’t gotten to you but. That’s all it means.
Google aggregates indicators on the web site degree, not simply the web page degree. You may have particular person pages performing whereas the general high quality sign of your web site degrades. And when the enforcement catches up (algorithmically or manually), it doesn’t choose off pages one after the other. It hits the lot.
That is the content material spinner’s fallacy, recycled: “It’s working proper now, so it should be a technique.” Demand Media’s content material was rating too. Proper up till it wasn’t.
Lily captured this completely: “The case research: scaling AI content material is working! The truth:” – adopted by the site visitors cliff that inevitably arrives. Each scaling success story is a snapshot taken earlier than the correction. No person publishes the sequel.
The Economics Don’t Even Make Sense
Put aside the danger for a second. Let’s discuss what you’re truly producing.
5 hundred AI-generated articles a month. Every one must be reviewed for accuracy – as a result of LLMs hallucinate, and publishing incorrect data is a legal responsibility that extends nicely past web optimization. Every one must be checked for originality – as a result of if it reads like all the things else within the index, it offers no added worth; no aggressive benefit. Every one wants editorial oversight to make sure it truly serves the viewers you declare to serve.
In the event you’re doing all of that, the associated fee simply moved – and presumably elevated – whilst you satisfied your self you have been being environment friendly. The “effectivity” of AI content material era evaporates the second you apply the standard requirements the content material truly wants to fulfill.
And if you happen to’re not doing any of that? You’re publishing unreviewed, unoriginal, probably inaccurate content material at scale below your model title. I genuinely don’t perceive how anybody indicators off on that.
Similar Mistake, Higher Instruments
Content material spinning. Programmatic web optimization. AI-generated content material at scale. Three completely different instruments, one similar mistake: treating content material as a producing downside.
Manufacturing produces similar outputs at scale – that’s the purpose. Content material derives its worth from the alternative: from being particular, from being knowledgeable by expertise, from saying one thing the remainder of the index doesn’t. Each try and industrialise it crashes into that contradiction.
You may’t automate specificity. You may’t template expertise. You may’t generate unique thought by operating a immediate by way of an LLM and hoping one thing helpful comes out. And these constraints gained’t be solved by the following mannequin launch. They’re baked into what makes content material value studying within the first place.
The individuals who maintain chasing scale are optimising for the flawed variable. They see “extra content material” as an enter that produces “extra site visitors” as an output. However the perform just isn’t linear. It by no means was. It’s gated by high quality, and no quantity of quantity bypasses the gate.
The Solely Query That Issues
Earlier than you publish something (AI-assisted or in any other case), ask one query: What does this web page supply that the reader can’t already get?
If the reply is “nothing, however we’ll have extra pages listed,” you’re not constructing a content material technique. You’re constructing a legal responsibility. And also you’re doing it with the boldness of somebody who has apparently by no means heard of Panda, by no means checked out what occurred to programmatic web optimization websites in 2022, and by no means learn Google’s personal spam insurance policies.
You may persuade your self for so long as you need. However you’ll solely idiot everybody else for some time.
The wall continues to be there. It’s at all times been there. The instruments maintain altering. The wall doesn’t.
Extra Assets:
This publish was initially printed on The Inference.
Featured Picture: Roman Samborskyi/Shutterstock
