Ecomrank.ai
Start Free

AI generated content and Google: what is allowed in 2026

August 8, 2026·11 min read
Illustration of a home linen workshop where a long row of identical folded cloths fills one side of the workbench, a single carefully stitched piece sitting on the other side, a wicker basket on the floor, wide empty space at the bottom

You are holding off on using AI to write your posts because someone told you Google punishes it. The story of AI generated content and Google fits in one sentence: there is no ban, there is action against pages produced in bulk that give the reader nothing. The difference sounds small, and it decides everything you will do with your store blog this year.

TL;DR

  • Google does not judge how a text was written, it judges what the page gives the person reading it.
  • The rule that concerns you is called scaled content abuse: many pages produced with little or no added value, whatever tool made them.
  • A store publishing three posts a week from real customer questions is not the target. A store dumping three hundred pages built from a keyword list is.
  • Ecommerce carries extra duties, on images and on the product data you send to Google, and almost nobody knows about them.
  • Telling readers how your content is produced is not required for a blog. Google suggests it, and it costs you nothing.

Table of contents

AI generated content and Google: what the rules actually say

Google publishes its position on a dedicated documentation page, last updated on 10 December 2025: Guidance on Generative AI Content. There is no ban. There is a quality requirement, the same one that applies to a hand written text, plus a warning: using generative AI tools to produce many pages without adding value for users may violate the spam policy on scaled content abuse.

In other words, the question the engine asks is not "who typed this". It is: does this page give something to the person who just searched for it. A weak post written by a freelancer paid 60 dollars a piece and a weak post generated in thirty seconds have exactly the same problem, and that problem existed long before AI.

That is good news for you, because you run a store. You hold information nobody else can write: what buyers ask before they order, what comes back through support, how your products behave after six months of use, why you picked one material over another. That raw material is what makes the difference, not the tool that strings the sentences together. The full path, from a search to an order, is laid out in our post on what a blog actually brings an online store.

The red line: many pages that add nothing

The rule that applies has a name in Google's documentation: scaled content abuse. Its official definition, on the Google spam policies page last updated on 15 May 2026, is one sentence: "Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users."

Two words carry the whole thing. Many, because the rule targets volume, not a single piece. And primary purpose, because it targets intent: pages made to occupy space rather than to answer.

Google is also clear that this is not an AI rule. It targets large amounts of unoriginal content that provides little to no value to users, no matter how it was created. Hand spun content farms were already caught by it ten years ago. AI did not create the problem, it made it far cheaper.

For a store, the practical translation is simple. Three hundred pages generated from a purchased keyword list, all built on the same template, with nothing from your trade inside: that is the case being described. Two or three posts a week, each answering a question your buyers genuinely ask: that is not, even if a machine drafted them.

What the rule changes for an online store

It moves the dial you should be working on. It is not writing speed, it is what you add to the text. Here is what a store can bring that a generator on its own will never find.

  • The real questions. From your inbox, your support replies, your comments, your market stall conversations. They are phrased in your customers' words, not in industry vocabulary.
  • Your actual products. A soap maker explaining how to store a cold process bar talks about their bars, their ingredients, and how long they really last in a customer's bathroom.
  • Your own numbers. The return rate on one model, the question that comes up three times a week, the size that runs small. Nobody else has them.
  • Your photos. Taken at your place, on your products, in your conditions.
  • Your mistakes. What you tried that did not work is often the most read part of a post.

A post that contains those five things looks like no other page on the web, even if a machine wrote the connecting sentences. A post that contains none of them looks like every other page, even if you wrote it yourself on a Sunday night. Picking the subject is a separate job: our method for pulling ten subjects out of your own catalog is in blog post ideas for an online store.

The extra rules that apply to ecommerce

This is the part almost nobody quotes, and it concerns you directly if you send your catalog to Google, for example to appear in Shopping. Google's guidance on AI generated content carries two requirements written with "must", not with "you may want to".

  • AI generated images must carry IPTC metadata using DigitalSourceType with the value TrainedAlgorithmicMedia. In plain terms, an invisible technical label saying the image came from a model.
  • AI generated product data, such as title and description attributes, must be supplied separately and labelled as AI generated.

These apply to the product data you send to Google, not to the text of your blog. They change over time, and the official page is what counts: check it before you touch your product feed, or ask whoever manages your store to do it. If you generate product visuals, the first line applies to you as well.

One more point of plain common sense: a generated product image is not a product photo. A buyer who receives something different from the picture feels cheated, and the return costs you more than the photo session would have. On a product page, the basics of getting a product page to rank come before any question of tooling.

Should you tell readers your posts were written with AI

Google suggests considering it, without requiring it for a blog. The wording is careful: consider adding information about how your content was created, in a way that makes sense for your audience. It is a recommended practice, not an obligation.

Our view, and it commits this site too: one line is enough, and it pays for itself. Something like "this post was drafted with an automated tool, from our own exchanges with customers" is honest, takes two seconds to read, and ends the argument before it starts. Readers who do not care will not notice it. Suspicious readers will value that you hid nothing.

This blog works exactly that way. The posts you are reading are produced by the machine we are building, published on our own domain, and measured in public at day 7, day 14 and day 28: indexing, impressions, clicks, position. When a measurement says nothing, we write that it says nothing. That is the only honest way to show what automatically produced content actually does, instead of promising it.

Six questions to know if a post is worth publishing

Run every post through this filter before you hit publish. If you answer no three times or more, do not publish it: it will do you more harm than good.

  1. Would a customer want to thank me for this page? The shortest and the hardest test.
  2. Could a competitor publish this page word for word? If yes, there is nothing of you in it.
  3. Does the answer arrive in the first two sentences? A post that keeps the reader waiting for six paragraphs is not answering, it is filling.
  4. Did the question come from a real person? An email, a message, a conversation. Not a row in a keyword spreadsheet.
  5. Does the post lead somewhere? To a product, to another post, to a concrete next step.
  6. Could I defend every claim in it? If one sentence makes you uneasy, the tool probably invented it.

A blog that fails all six is usually not punished by Google: it is simply ignored. That is the most common outcome by far, and we covered it in why a store blog gets no traffic.

The mistakes that cost you

  • Publishing in bulk to "get ahead". That is precisely the behaviour described by the scaled content abuse policy. Volume never makes up for missing value.
  • Generating variants of the same post for every city, colour or size. Those pages look too alike to deserve separate existence.
  • Letting a false claim through. A tool that invents a duration, a price or a standard exposes you to much worse than a ranking issue: you are responsible for what you publish.
  • Believing that "humanising" the text fixes it. Removing AI turns of phrase from an empty post gives you an empty post without AI turns of phrase.
  • Forgetting the product data. You can be spotless on the blog and out of line on your product feed, which is the part that touches your sales.

Frequently asked questions

Does Google penalise posts written by AI?

No, not for being written by AI. Google's documentation is explicit: the problem is pages produced in quantity with no added value for users, whatever the tool. A useful post stays a useful post.

Can Google detect generated text?

Google says its anti spam systems go after spam however it is produced. It publishes no detection guarantee, and more importantly, that is the wrong question. Building a strategy on "will it show" means betting against your own reader, who can tell perfectly well when a page teaches them nothing.

How many generated posts per month is too many?

There is no published threshold. The useful marker is not a number, it is an origin: as many posts as you have real questions to answer. A small store easily has forty or fifty in reserve. The day you write to fill a quota, you have crossed the line, whatever the count says.

Can I generate my product descriptions?

Yes, as long as they are accurate and specific to your product. If you send that data to Google for Shopping, the documentation asks that generated titles and descriptions be supplied separately and labelled as such. Check the official page before you change your feed.

Should I delete the generated posts I already published?

Not automatically. Reread them with the six questions above. The ones answering a real demand deserve to be improved, not erased. The ones with nothing from your trade in them can be merged into one fuller page, which beats ten empty ones.

Where to start this week

Take the last five posts on your blog, whoever wrote them, and run them through the six questions. Twenty minutes will tell you whether your problem is a tooling problem or a raw material problem. Nine times out of ten it is the second one: the subjects did not come from your customers.

Then open your inbox and write down the last ten questions someone asked you before buying. That is your editorial plan for the next two months, and no spam policy can hold it against you.

That is exactly what Ecomrank does: write posts around your catalog and your buyers' questions, publish them on your store blog, link them to your product pages, then measure honestly what they produce at day 7, day 14 and day 28. Ecomrank is live: the trial is free for 7 days and you can cancel anytime.

Start your free trial