Skip to content

SEO and AI

Does Google penalise AI content? What one online shop's data shows

Does Google penalise AI content? What Google actually says, how to tell why your traffic dropped, and what data from a 900-page online shop shows.

Author
Mücahit Arslan · Co-founder
Published
Reading time
11 min read
Rows of kraft-paper product tags on a wooden table, each with the same grey stamp, as a hand presses a wooden stamp onto one more tag.
Contents
  1. What does Google actually say about AI content?
  2. What does scaled content abuse look like on product pages?
  3. What happened to a 900-page shop after the September 2026 spam update?
  4. Why did pages nobody touched lose traffic too?
  5. Why has my organic traffic dropped?
  6. Did sales fall as far as traffic?
  7. What should you do after a drop, and what should you avoid?
  8. How can you use AI for product descriptions without the risk?
  9. How can MAKK Digital help?
  10. What are the limits of this data?

No. Google doesn't penalise content for being written by AI. What it looks at is whether a page is useful to the person reading it. The risk lies elsewhere: rewriting hundreds of product and category pages from the same template in a short space of time. If that adds nothing for the reader, Google can class it as scaled content abuse, which it treats as spam. And the damage doesn't necessarily stay on the pages you changed. In an online shop we analysed in October 2026, rewritten pages and untouched pages each lost around 60% of their Google traffic.

What does Google actually say about AI content?

Google's position hasn't changed since February 2023: appropriate use of AI or automation is not against its guidelines. Using automation, AI included, to generate content mainly to manipulate rankings is.

In autumn 2026 the line got sharper. On 1 October, Google revised its guidance on generative AI content. It now calls it critical to fact-check and review all AI-generated content by hand before publishing, and it spells out that this covers page titles, meta descriptions, structured data and image alt text, not just the body copy. If you've generated a thousand Merchant Center titles with AI, check those the same way. Google's guidance on helpful content also lists this question among the warning signs of content made for search engines rather than people: "Are you using extensive automation to produce content on many topics?"

People often say that "Googlebot caught the AI text". That isn't how it works. Googlebot is just Google's crawler. It fetches pages and brings them back to Google, but it doesn't decide rankings. That's the job of Google's ranking systems and spam detection systems such as SpamBrain, which weigh up whether a page is worth showing.

What does scaled content abuse look like on product pages?

Google's spam policies define scaled content abuse as generating many pages mainly to manipulate search rankings rather than to help people, and they stress that this applies "no matter how it's created". The first example on the list is using generative AI tools to produce many pages without adding value for users.

On a product page it rarely looks like spam. Each description reads fine on its own. Put twenty side by side and the template shows: same length, same subheadings, same number of FAQs, the same paragraph about the shop at the bottom. Some sentences appear word for word on dozens of pages.

Google's guide to its ranking systems says that site-wide signals and classifiers are also used, meaning Google judges the site as a whole as well as single pages. So a template that runs through half your shop may end up being a problem for all of it.

What happened to a 900-page shop after the September 2026 spam update?

The shop is in Turkey and has around 900 pages that appear in Google. Between July and October 2026, the copy or titles of roughly 60% of its product and category pages were rewritten with AI assistance, all from one template. The template gave each page a similar length, similar headings, eight FAQs and a paragraph about the shop. Over the same period, more than 1,000 Merchant Center product titles, a large batch of image alt texts and a series of guides with an identical structure were changed in bulk.

Google's September 2026 spam update started on 24 September and finished on 8 October. The shop's first drop came the next day, 9 October. A second drop followed from 17 October, a date for which Google announced no update.

Before (1–8 October)After (23–25 October)
Google clicks per day864315
Average position5.77.3
Main search termsPage oneMostly page two

That's roughly 60% fewer clicks. October the year before showed no such fall, so seasonality was ruled out. Nothing was technically wrong either. All 26 of the pages that lost the most were indexed, crawled without errors and had the correct canonical tag (the tag that tells Google which version of a page is the main one). There was no manual action in Search Console. The pages hadn't disappeared; Google was simply showing them further down.

Then we looked for repeated sentences. We found 1,149 that appeared word for word on three or more product pages, and together they made up about a third of all the words in the product copy. One line, which translates roughly as "a meaningful gift for the people you love", sat on 51 pages. A sentence about bulk orders for companies sat on 49. Neither told a customer anything about the product in front of them.

Why did pages nobody touched lose traffic too?

This is the finding that matters most. The 27 category pages with rewritten copy lost 65% of their clicks, and the 59 category pages that nobody touched lost 60%, almost as much.

Page by page, the picture is the same. Of 271 pages that had more than 15 impressions a day in the first two weeks of October, half kept 29% of their impressions or less. Product pages, categories, guides and the home page all fell by the same proportion.

So it looks as though Google rated the whole site lower, rather than demoting individual pages. That fits with what Google says about site-wide signals. It can't be proven, because Google doesn't tell site owners why they dropped. But on the data available it's the strongest explanation: a large part of the site was rewritten from one template within a few months, and the day after the spam update finished, the whole domain was marked down.

If you run a shop, don't assume that only the new pages will suffer when things go wrong.

Why has my organic traffic dropped?

The pattern in Search Console usually tells you, because each cause leaves a different trace. Google's own guide to debugging traffic drops suggests the same approach: look at the last 16 months, compare with the same period a year earlier, and check which pages and which search terms lost out.

CauseWhat it looks like in Search Console
Seasonal dip or lower demandYour position holds and fewer people search. The same dip shows up a year earlier.
Google's AI OverviewsPosition and impressions hold, the click-through rate falls.
Technical problemPages drop out of the index, URL Inspection shows errors, the fall starts when the error does.
Changes to specific pagesOnly the changed pages lose, the rest stay put.
Site-wide reassessmentEvery type of page loses by the same proportion, position falls, the date lines up with a Google update.

Impressions are the number of times your pages were shown in search results, and click-through rate is the share of those that led to a click. Position is where your page ranked in the results; Search Console reports it as "average position".

Be careful with that average. When you drop, you vanish completely for some search terms. Those terms fall out of the calculation, so what's left looks better than it is. In the shop we analysed, 38% of the lost impressions came from searches where the shop no longer appeared at all.

To do this for your own site, you need to compare individual pages and search terms one by one. Across several hundred pages that takes days by hand, and a wrong diagnosis means fixing the wrong pages. We can run this analysis for you; more on that at the end.

Did sales fall as far as traffic?

Not at first. Visits from organic (unpaid) search fell by 53% compared with the same days in September. Purchases from organic search went from 5.0 to 4.3 a day over the same period, a drop of around 14%. That's first-week data, and a firm conclusion needs a longer window.

The gap comes down to the kind of traffic that was lost. The pages that lost the most visits were guides: pages people read to find something out and usually close without buying. In those results, the shop's place went to reference sites and forums, lawyers' and doctors' websites and government pages. On questions like these, Google appears to prefer a specialist source to a shop's blog. Where people were searching to buy, the big marketplaces and specialist retailers moved up.

So there's no need to panic, but don't relax either. Once category pages slip to page two, the sales loss grows over time. And guides are often where a future customer first comes across your brand.

What should you do after a drop, and what should you avoid?

First rule: don't rewrite everything again. If a bulk change caused the problem, another one is likely to make it worse. Google's advice after a spam update is plain: review the spam policies and bring the site in line with them. It adds that a site may improve once the automated systems learn, over a period of months, that it complies.

What to avoid

  • Rewriting every page from scratch.
  • Changing URLs or restructuring the site. That carries risks of its own, which we cover in our guide to website migration SEO.
  • Deleting pages that still bring in traffic.
  • Disavowing backlinks (telling Google to ignore links from other sites) without a clear reason.
  • Making several big changes in the same week. You won't be able to tell which one worked.

What the shop did

  1. Stopped producing new content from the template.
  2. Deleted the repeated filler sentences on 508 product pages rather than rewriting them. Product names, URLs, titles and images stayed as they were, and the copy got 13% shorter on average.
  3. Took 53 off-topic guides with no traffic out of Google's index.
  4. Rewrote the 20 most-searched guides using official sources.
  5. Added the real shop address, phone number and opening hours to 48 of its guides.

Where things stand

It's too early to say. Cleaned-up and untouched product pages kept falling at the same rate for the first few days. The rewritten guides, on the other hand, started appearing more often within days, and one climbed from position 33 to position 5. That could simply be a short-lived boost for new content. The commercial pages haven't recovered yet. We'll update this section after the mid-November measurements.

How can you use AI for product descriptions without the risk?

You don't have to stop using AI. Just feed it real product information instead of a template to fill. Most shop platforms now come with a built-in AI writer that can fill hundreds of pages in an afternoon, and that's what makes it so easy to overdo.

  • Facts first, then copy. Dimensions, materials, what it's for, care, who it suits. If there's nothing more to say about a product, keep the description short rather than padding it out.
  • No word count targets. A rule like "300 words per product" produces filler.
  • No boilerplate on every page. Delivery, returns and information about your shop belong on their own page or in the footer, not in every product description.
  • No one-size-fits-all FAQs. Only add questions customers actually ask about that product.
  • Read the title, meta description and alt text too. Google's October 2026 guidance asks for a manual check on these as well.
  • Start small. Try the new copy on 20 to 30 products first, watch Search Console for four weeks, then expand. Don't change the whole shop in one week.
  • Count the repeats. Before publishing, check how many pages each sentence appears on. A sentence that sits on 50 pages tells neither customers nor Google anything.

How can MAKK Digital help?

After a drop, the costliest mistake is acting before you know the cause. Before a big rewrite, it's failing to record where you stand, so there's nothing to compare against later. We can take on either job for you:

  • Diagnosis: using your Search Console and Analytics data, we work out whether the drop comes from a seasonal dip, a technical fault, AI Overviews or a site-wide reassessment.
  • Template audit: we count how many pages each sentence appears on and work out how much of your copy is filler.
  • Clean-up plan: page by page, we decide what gets cut, what stays and what comes out of the index, then carry it out without touching URLs or titles.
  • Product copy: we write product and category copy based on keyword research and the facts of each product, in English, German and Turkish if you need it.
  • Monitoring: we track the effect in Google regularly and report back in plain language.

You'll find the details on our SEO and performance marketing page. If the shop itself needs rebuilding, see web design and e-commerce. If your traffic has dropped, or you're planning to rewrite your catalogue with AI, get in touch. And if you'd like to know how you show up in AI answers as well as in Google, start with our free AI visibility check.

What are the limits of this data?

This is one shop, and one case doesn't make a rule. Google doesn't tell site owners why their traffic dropped, so what we describe here is the explanation the data supports best, not a proven cause. The bulk rewrite and the drop happened in the same period, but Google may have made other, unannounced changes at the same time. Recovery hasn't been measured yet. The figures come from Google Search Console and Google Analytics as of 28 October 2026. We'll update this article after the November 2026 measurements.

Frequently asked questions

Google doesn't say it tries to work out who or what wrote a page. Its spam policies judge content by its value to users, no matter how it was created. But a pattern repeated across hundreds of pages is visible to automated systems, whoever wrote it. In the shop we analysed, a third of the product copy was made up of sentences that appeared word for word on many pages, and that repetition is the risk.

Not in itself. Google states that appropriate use of AI isn't against its guidelines. AI content becomes a problem when it's produced at scale without adding anything: generic descriptions, repeated sentences, the same FAQs on every page. Since October 2026 Google also asks for a manual check of AI-generated titles, meta descriptions and alt text before they go live. Accurate, product-specific copy that a person has checked is fine.

Probably not. Most of them swap words for synonyms and reshuffle sentences. Google's spam policies list automated transformations such as synonymising among their examples of scaled content abuse. If no new information is added, the result is the same content in different words. What does help is adding facts about the product that weren't there before.

Start with the Manual actions report in Search Console, though it only covers one of two kinds. A manual action means someone at Google has reviewed your site, and you'll see a notice there. An algorithmic drop comes with no notice; your pages simply appear lower down. Most drops are the second kind, and the shop we analysed had no manual action. So no notice doesn't mean no problem. Check your clicks and impressions over time.

Google gives no fixed timeline. Its page on spam updates says a site may improve once the automated systems learn, over a period of months, that it complies with the spam policies. So traffic won't come back in the week you make the fix. After a clean-up, allow several weeks without other major changes and track the results by page type.

Google doesn't require it for search. Its helpful content guidance says AI disclosures are useful where someone might reasonably wonder how a piece of content was created. With product descriptions, that's rarely what a customer is asking; they want to know whether the details are right. Accuracy matters more than a label. Legal requirements can differ by country and are a separate question.

Questions about this topic?

Write to Mücahit Arslan directly. You'll be talking to the person who wrote this article.

Contact 

More articles