AI visibility

AI-generated content and SEO: what Google rewards and what it removes

Ilyes Massoudi15 September 20266 min read

The short answer

Google does not penalise content for being written by AI; it evaluates quality and usefulness regardless of production method, and that has been its stated policy since 2023. What it explicitly penalises, since the March 2024 update, is 'scaled content abuse': publishing many generated pages with no value in order to manipulate rankings, and it has removed entire sites for it. The practical problem in Arabic is that generated text reads like translation, repeats generalities and sometimes invents. The safe method: use the tool for the draft and the structure, add the numbers, examples and experience yourself, and review every sentence as if you were signing it.

What Google’s policy actually says

The common fear is that Google “detects” generated text and penalises it. The published policy says something else: Google rewards high-quality content however it is produced, and evaluates it by the same standards as any content, above all experience, expertise, trustworthiness and usefulness to the reader. Automation is not a violation in itself.

What is explicitly a violation, since the March 2024 update, is scaled content abuse: producing many pages, by tool, cheap editor or copying, whose main purpose is manipulating rankings rather than serving the reader. We have seen sites lose most of their traffic within weeks, and some removed from the index, because they published hundreds or thousands of similar generated articles.

The line, then, is not between human and machine. It is between content that says something and content that fills a gap.

Why generated Arabic gets into trouble faster

Models write Arabic better than ever, and still fall into three patterns a Gulf reader spots in the first line:

It reads like translation. English sentence structure, textbook formal terms nobody types into a search box, sometimes needlessly vowelled.

It repeats generalities. “In our fast-paced digital world it has become essential to…” and then nothing. The text is long and carries no number, no name, no example.

It invents. Figures, statistics and sources that look convincing and do not exist. In a medical, legal or financial article that is not an SEO problem but a liability problem.

The result is that an unreviewed generated Arabic article competes for a query with text indistinguishable from ten other generated articles on the same query, and gives Google no reason to choose it.

A workflow that uses the tools without the losses

The tools are excellent at what is mechanical and weak at what is specific to you. Divide the work on that basis:

Stage	Who does it	What it produces
Choosing the target query and its volume	You, from data	One intent per page
Finding the questions readers ask	The tool collects, you choose	A list of headings
First draft and structure	The tool	A skeleton without padding
Numbers, examples, cases	You, from experience	What a competitor cannot copy
Opinion and recommendation	You	The sentence that says what to do
Checking every claim	You	No figure without a source
Final read aloud	You	Arabic sentences, not translated ones

One rule: nothing is published under your name that you have not read in full. Text that starts from a tool and passes through these stages is human content in every sense that matters to Google and to the reader.

Signs of content that will lose

Review your pages against this list, and you will know which are at risk:

It contains not one specific number or example from real work.
Your company name could be swapped for any other without changing anything.
It answers the same question ten other pages on your site answer.
Its title is a question and there is no direct answer in the first paragraph.
It was published by the dozen in a single week.
No human name is responsible for it.

A page showing three of these signs is one to delete or merge, not to fix by adding.

Scaled content and programmatic pages

Pages generated programmatically from data, one per city, product or combination, are not a violation in themselves; some of the largest sites are built on them. The violation is when the pages are copies with one word swapped: “cleaning company in Riyadh”, “cleaning company in Jeddah”, same text. Google treats that as duplicated content at scale and demotes the whole site, not just those pages.

The test: do you have genuinely different real data for each generated page that deserves a page? If the answer is no, one good page beats fifty.

What the audit checks

In the Digital Presence Audit I review pages with low impressions and near-identical content, identify what deserves rewriting and what deserves deletion or merging, and connect that to the Google updates that coincided with any traffic drop. The right deletion sometimes produces a faster effect than any addition.

Common questions

Should I disclose that an article was written with AI help?

Google does not require it, but recommends disclosure where a reader would reasonably expect it. The practical rule: if the name signed on the article is a real person, that person is responsible for every sentence in it, whether they wrote or reviewed it. Never sign your name to text you have not read.

I published hundreds of generated articles and my traffic dropped. What do I do?

Delete or merge everything that says nothing another page does not say. The sites that recovered after the 2024 updates are the ones that cut their content down to what deserved to stay and rewrote it with real experience. Adding on top of weak content does not fix it.

Can AI-detection tools identify generated text?

Not reliably, and Google does not depend on them. What marks generated text for reader and engine alike is the absence of specificity: no numbers, no examples, no opinion, no experience. Add those four and the question disappears, whether the text started from a tool or from a blank page.