Will Google Penalise AI-Written Content?

This gets argued about far more than it needs to, because Google's position has been consistent and is not complicated.

What Google actually says

The spam policies target content produced primarily to manipulate search rankings rather than to help people. The method of production is not the test.

The companion document, on creating helpful content, asks the questions that actually matter: does this demonstrate first-hand expertise, would somebody feel they had learned something, is it the kind of thing you would bookmark or recommend.

Neither asks who typed it.

Why the distinction holds up

A page written entirely by a person can be spam. Thin, keyword-stuffed location pages predate AI by fifteen years and plenty were written by humans being paid very little.

A page written with AI assistance can be genuinely useful, if someone who knows the subject shaped it and the detail in it is real.

Judging the output rather than the tool is the only position that makes sense, and it is the one Google has taken.

Where the real risk sits for a trade business

Not penalties. The risk is being unremarkable, which costs you quietly and without any notification.

AI is very good at producing the average of everything written on a subject. For most trade topics the average is already poor: vague advice, no numbers, no opinions, no scars.

Publish that and three things happen. It does not outrank the pages that were specific. Nobody reading it is persuaded to ring you. And you have added pages to your site that say nothing, which makes the whole site a bit less distinctive.

That is not a penalty. It is just money and effort spent going nowhere, and it is far more common than any manual action.

What an average page actually reads like

Worth being concrete, because "unremarkable" is easy to nod at and hard to spot in your own draft.

An average page about replacing a hot water cylinder will tell you that cylinders last a certain number of years, that signs of failure include leaks and inconsistent temperature, that you should use a qualified professional, and that costs vary depending on your requirements.

Every sentence is true. None of it was worth publishing, because all of it is already on forty other pages and none of it helps anybody decide anything.

A page worth publishing says what it costs where you are, what the two options actually are, which one fails early and why, how long you will be without hot water, and what the thing is that the last three customers did not expect. That is the difference, and no model can supply it.

The part a tool cannot do

Everything that makes a trade page worth reading is first-hand.

What actually goes wrong on that job. What customers get told that is not true. What the thing costs where you are, in the currency people here use. Why the cheap option fails in eighteen months. The job you walked away from and why.

Fifteen years on the tools taught me that most of what makes a page about plumbing useful is the stuff you only know from doing it. No model has that about your business, because it has never been in the back of your van.

How to use it without producing sludge

Let it structure, not decide. It is good at an outline and at tidying a draft that already contains your knowledge.

Put the specifics in yourself. Numbers, places, the actual sequence of a job, the thing that went wrong last month.

Cut every sentence true of any business in your trade. This usually removes a third of a draft and improves it.

Check the facts. Models state wrong things confidently. Publishing a wrong price or a wrong regulation damages you in a way no algorithm needs to be involved in.

Publish less. Ten pages worth reading beat sixty that are not, and that was true before any of this.

The fastest way to test a draft

Read it and ask one question of every paragraph: could a competitor publish this sentence unchanged?

If they could, it is doing nothing for you. A page where most sentences survive that test is a page worth having, and it is also the page an AI tool is most likely to quote, which is a useful side effect rather than the reason.

The short version

Google does not care how the page was made. It cares whether it was made for a person.

Which means the question is not whether AI will get you penalised. It is whether the page says anything that could only have come from you. If it does not, the tool was never the problem, and a page like that would have failed in 2015 too.

That is also the test for whether an AI will quote you, because the same specificity that convinces a reader is what makes a page worth citing.