AI Content and Search Engines: What We Learned the Hard Way

When ChatGPT burst onto the scene at the end of 2022, two completely opposite reactions swept through the SEO world. On one side, people were convinced that artificial intelligence would make copywriters, content creators and SEO specialists obsolete. On the other, people were terrified that Google would penalise anyone who dared use AI to create content.

Now, more than three years later, we can look at real data and say with confidence that both camps were wrong.

AI has not replaced humans in content creation — and it probably will not in the foreseeable future. At the same time, Google is not blindly penalising all content created with the help of AI. The reality, as always, is considerably more nuanced and far more interesting.

It would be dishonest, however, to pretend that nothing has changed. A great deal of copywriting work has simply evaporated. We believe this is a transitional phase. We do not know exactly what the landscape will look like in ten years. What we are fairly certain about is that genuinely human content — written by people with real experience and something original to say — will become increasingly rare and, precisely because of that, increasingly valuable.

What Google actually says about AI content

Let us start with the facts. In February 2023, Google published a clear and unequivocal position on AI-generated content on their official Search Central blog. The key statement was straightforward: the appropriate use of AI or automation is not against their guidelines, as long as it is not used to generate content primarily to manipulate search rankings.

In plain English: Google does not care whether you use AI to create content. What matters is why and how you use it.

The crucial point is this: content should be created primarily for people, not to manipulate rankings. If you are using AI to produce thousands of pages whose sole purpose is to rank for some keywords and capture clicks, you are in trouble. If you are using AI as a tool to create better content more efficiently, Google has no objection.

A word of caution: we have learned from long experience that official Google statements should always be taken with a generous pinch of salt. Besides, they sometimes change their minds. And indeed…

How Google’s position has evolved (2023–2025)

Throughout 2024 and especially into 2025, the position has become noticeably more nuanced and, in some respects, considerably tougher.

In August 2024, John Mueller signalled the direction by stating that Google aims to connect people with a range of high-quality sites, including small or independent sites that create useful and original content.

In January 2025, Google updated the Quality Rater Guidelines, introducing for the first time an official definition of “Generative AI” and specific instructions on how to evaluate AI-created content. The new guidelines specify that content can receive the lowest quality rating if it is created with little or no effort, lacks originality, or adds no value for visitors. While these guidelines do not directly influence rankings, they are used to train the algorithm. And when Google takes the trouble to put something in writing in the Quality Rater Guidelines, you can be sure the topic is a priority.

Then, in mid-2025, came the real blow. Google began issuing manual actions against sites guilty of what they call “Scaled Content Abuse” — the mass production of low-value content. Affected sites received notifications in Search Console with messages along the lines of: “It appears that pages on this site use aggressive spam techniques, such as scaled content abuse.”

The result? Complete disappearance from the SERPs. Not a gradual decline — total removal.

So let us be clear: Google has not changed its mind about AI itself. It has changed its mind about how easy it is to abuse. The official position remains “AI is fine,” but in practice Google is becoming increasingly aggressive in punishing those who use it badly. And “badly” essentially means “on an industrial scale without meaningful human oversight.”

“AI Content” versus “Content created with AI”

This is the heart of the matter, and it is a distinction that many people still have not grasped.

There is an enormous difference between:

1. AI Content: text generated entirely by artificial intelligence, perhaps from a generic prompt such as “write a 1,000-word article on how to choose an electric razor,” published with no significant human intervention.
2. Content created with AI: text that uses artificial intelligence as a tool in one or more stages of the creative process, while maintaining substantial human involvement in terms of strategy, experience, fact-checking and personalisation.

The first category is destined to fail. The second can work extremely well.

Think about it: even before AI, we all used tools to create content. Spell checkers, digital thesauruses, keyword research tools, readability analysers. All of these are forms of automation that nobody ever considered problematic. AI is simply a more powerful tool in the same category. The problem arises when people think this tool can do all the work for them.

Our experiment (and what went wrong)

We wanted to verify this on the ground, so we ran an experiment. We created and published approximately one thousand articles generated almost entirely by artificial intelligence on one of our editorial projects.

Let us be clear: we are not talking about a couple of throwaway posts to see what would happen. We invested significant time. Images had to be selected, content had to be uploaded to the CMS, basic optimisations had to be done. This was not a cheap experiment.

The result? Worse than zero.

Not only did those articles fail to bring in new traffic, but the site actually lost visitors compared to the previous period. Google did not penalise the site with a manual action. No alarming message appeared in Search Console. It simply… ignored our content. Or worse, it decided that the site as a whole deserved less visibility.

The experiment taught us a lesson worth far more than the money and time we invested: AI can be a valuable ally in the creation process, but content that adds nothing new, that lacks the “salt” of human experience, is worse than content never written at all. It is not neutral — it actively dilutes the quality signal of your entire site.

We are deliberately not naming the project in question. Partly out of self-preservation (the site has since recovered, after a thorough cleanup), partly because the specific domain is irrelevant. What matters is the lesson.

When AI actually helps

It would be wrong to conclude from our experiment that AI is useless or harmful for content creation. Quite the opposite. Used correctly, it can be an extraordinary tool. Here is where it genuinely shines.

Research and planning

AI excels at helping you generate ideas for new content, identify different angles on a topic, create structured outlines, suggest alternative titles and analyse the search intent behind a query. If you need to write an article on “how to choose WordPress hosting,” you can ask an AI to suggest ten questions users might be asking. Some will be obvious; others might surprise you and provide interesting angles you had not considered.

Getting past writer’s block

Anyone who writes for a living knows the feeling: sometimes you get stuck. You know exactly what you want to say but cannot find the right words. In these moments, AI can be like that colleague you brainstorm with: it does not give you the final answer, but it helps you get unstuck.

Optimisation and revision

AI can help you check grammar and syntax, suggest synonyms to avoid repetition, analyse text readability and identify unclear passages. These are all tasks where AI can save precious time, freeing you to concentrate on the more creative and strategic aspects.

Translation and localisation

If you need to create content in multiple languages — and as a Malta-based agency working with Italian clients, we do this constantly — AI can provide a solid first draft that you then refine. It is infinitely faster than starting from scratch, and the quality of AI translation has improved dramatically in the past two years. Just never, ever publish an AI translation without a native speaker reviewing it.

Variations at speed

You have written a perfect meta description but want to see some alternatives? AI can generate ten in a few seconds. The same goes for titles, introductions and calls-to-action. This is perhaps the single most time-efficient use of AI in a content workflow.

When AI causes damage

Now let us look at when AI becomes a problem. Fundamentally, it happens every time you think it can completely replace human work.

Industrial-scale content farming

We have seen dozens of cases of sites that published thousands — sometimes tens of thousands — of AI-generated articles within days. The strategy was simple: flood Google with content in the hope that at least a percentage would rank.

Google’s March 2024 Core Update swept away many of these sites. Not because they used AI, but because the content was useless, generic and devoid of real value for users. If you are considering this approach, please do not. We tried a milder version of it with our thousand articles, and even that was enough to cause damage.

Content without experience

The addition of the “E” for Experience in Google’s E-E-A-T framework was not accidental. Google wants to reward content that demonstrates direct, lived, real experience.

An article about “the best restaurants in Malta” written by someone who has never set foot on the island will always be inferior to one written by someone who lives here and has actually eaten at those restaurants. And AI, by definition, has no direct experience of anything.

Factual errors and hallucinations

AI can generate false information with absolute confidence. It can invent statistics, cite non-existent studies, attribute quotes to people who never said them. Without careful human review, these errors end up online and damage your site’s credibility — sometimes irreparably.

Homogenised content

One of the more subtle problems with AI is that it tends to produce content that is remarkably similar. If everyone uses the same tool with similar prompts to write about the same topic, the result is a mass of nearly identical articles. Google does not like duplicates, even when they are not technically duplicates. We have a colourful Italian expression for this kind of content — we call it the “Assyrian-Babylonian” approach: every article starts from the dawn of civilisation and methodically works its way to the present day, padding out the word count without adding any genuine insight.

Before ChatGPT, this type of content already existed — it was just slower to produce. Now it can be generated by the thousand in a single afternoon. The scale has changed. The underlying problem has not.

A practical framework

Based on our experience — including the expensive failure — here is how we now approach AI in our content workflow:

Use AI for: research, brainstorming, first drafts of structured content, translations, meta descriptions, outline creation, grammar checking, generating variations.

Never use AI for: final copy without human review, topics requiring direct experience, anything published without fact-checking, anything where your professional reputation is at stake.

The golden rule: if you could not tell the difference between your article and one generated by a random prompt, your article is not good enough. The human element — your experience, your opinion, your specific knowledge — is what makes content valuable. AI should amplify that, not replace it.

What this means for your website

If you run a business website, you are probably wondering: should I use AI for my content or not?

The honest answer is: it depends on how you use it. If you are thinking of using AI to quickly generate dozens of service pages, blog posts and product descriptions with minimal human involvement, we would strongly advise against it. You might see short-term results, but you are building on sand.

If, on the other hand, you are thinking of using AI to make your content creation process more efficient while maintaining quality, editorial control and the unmistakable voice of someone who actually knows what they are talking about — then absolutely, go ahead.

The websites that will thrive in the coming years will not be those that produce the most content, nor those that avoid AI entirely. They will be the ones that combine human expertise with AI efficiency in a way that produces something genuinely useful for the reader.

We learned this the hard way, with a thousand articles and a traffic graph pointing in the wrong direction. You do not have to repeat our mistake.

Get in touch →