Does Google penalise content written with AI?

What Google has said, and when

Google has said the same thing several times, in more detail each time:

• February 2023: Google said it looks at the quality of content rather than how it was produced, that appropriate use of AI or automation is not against its guidelines, and that using automation, AI included, mainly to manipulate rankings is spam. It added that AI gives content no special boost.
• March 2024: Google replaced its old policy on automatically generated content with a policy on scaled content abuse, which applies whether pages are made by automation, by people or by both.
• May 2026: its guide to the AI features in search warned that a high number of pages does not make a site better, and that creating pages for every variation of a query, mainly to manipulate rankings or AI answers, breaks the scaled content policy.
• Its current page on using generative AI says AI can help with research and with structuring original content, warns that generating many pages without adding value may break the scaled content policy, and tells publishers to check AI content for accuracy before it goes live.

Read together, the rule is simple. Google's question is whether the page helps the reader, and whether the site is churning out pages to catch searches. Which tool typed the words is not part of it.

Where the line falls

Google's spam policy describes scaled content abuse as many pages generated mainly to manipulate rankings rather than to help users: typically unoriginal content with little or no value, however it was created. Its examples, in our words:

• Using AI or similar tools to produce many pages that add nothing for readers.
• Scraping other sites, feeds or search results and rewording or translating them into new pages.
• Stitching together content from several pages without adding anything.
• Spreading pages across several sites to hide how many there are.
• Pages stuffed with search terms that make little sense to a reader.

The test Google keeps coming back to is purpose and value. Its guidance on helpful content asks whether the people you serve would find a page useful if they came to you directly, rather than from a search. A page that passes that test is fine, whoever or whatever drafted it. A page whose main job is to catch a search is at risk either way.

The local business version of this mistake

For a small business, scaled content rarely looks like a content farm. It looks like this:

• Forty town pages for a cleaning company, each the same text with the town name swapped.
• A page for every pairing of service and suburb: boiler repair in each of twenty areas, then boiler servicing in each of the same twenty.
• A blog that publishes a generic post every day, such as tips for cleaning an oven, written to a word count and never read by the owner.
• Competitors' pages run through an AI tool and reworded.
• The whole site machine-translated into five languages nobody at the business speaks.

Compare a page that earns its place: a page for the second town where you have a team based, with the area you cover from there, the people who work it, jobs you have done nearby and how parking and access work on its streets. One page like that does more than forty swapped names.

One page, written two ways

Say an electrician wants a page about fitting home car chargers. An AI draft written from nothing usually reads like this: "Looking for reliable EV charger installation? Our experienced team provides high-quality installations tailored to your needs, with excellent customer service and competitive prices."

Every sentence could sit on any electrician's site in any country. Google has nothing in it that a thousand other pages do not already say, and a customer learns nothing from it.

The same page built from the electrician's own notes says:

• Which charger makes they fit, and why they stopped fitting one of them.
• What the survey visit checks: the fuse board, the cable run from the board to the drive, and whether the earthing needs work.
• What usually decides the price: the length of cable, digging across a lawn or a path, and whether the board needs upgrading.
• The grant or scheme customers ask about, with a link to the official page rather than a figure that will date.
• Photos of recent installs on local streets, shown with the customers' permission.
• The questions people ask before booking: will it work with my solar panels, and can it go on a shared drive?

An AI tool can turn those notes into tidy paragraphs quickly. It cannot supply the notes.

Google's own checklist: who, how and why

Google's guidance on helpful content asks publishers to look at their pages in terms of who made them, how and why:

• Who: is it clear who wrote the page or stands behind it? For a business site, that is the business, with real names where readers would expect them.
• How: would a reader wonder how the page was made? Google says disclosing AI or automation is useful where someone might ask that question. It also said in 2023 that giving AI an author byline is probably not the best way to do it.
• Why: was the page made to help people, or to attract search visits? Google is clear that the second is the wrong reason.

The same guidance lists warning signs that a site is writing for search engines first. Paraphrased:

• Lots of content on many different topics, in the hope that some of it ranks.
• Heavy automation to produce content on many topics.
• Mostly summarising what others say without adding much.
• Writing to a word count.
• Moving into a topic with no real expertise, because it might bring traffic.

How to use AI and stay on the right side

Used this way, AI saves time without making the page worse:

• Start from your own material: job notes, quotes you sent, photos, your price list, the questions customers ask on the phone.
• Let the tool suggest a structure or a first draft from that material.
• Have someone who knows the trade check every fact, price, date and claim. AI tools write what sounds likely, which is not the same as what is true, and Google tells publishers to check AI content for accuracy before it goes live.
• Add what only you know: how you price a job, what tends to go wrong, what you would tell a friend.
• Check the page title and description too. Google says the same review applies to them, because they appear in search results.
• Never invent reviews, figures, qualifications or awards.

On the sites we build, the words are written with you and nothing is published until you approve it. We never invent reviews, numbers or claims you cannot back up, as the websites page says.

The bigger risk is being wrong

For most small businesses, wrong information is a bigger danger in AI-written pages than any penalty. A draft can give the wrong opening hours, a price you stopped charging, a promise you never made or a rule that has since changed, and it will say each one with total confidence.

In some trades a wrong sentence does real harm. A dental clinic page that promises results, or a law firm page that gets a deadline wrong, can mislead a patient or a client, and professional and advertising rules apply to what a business publishes, whoever drafted it. Have the person responsible for the advice read those pages before they go live.

How we apply this to our own site

This site has a page for each of 24 trades, a page for each job we do and a growing set of guides like this one. Pages that share a structure are exactly the pattern Google's policy describes when the content behind them is thin, so every page is held to the same rules:

• If a paragraph could be pasted into another trade's page unchanged, it is rewritten or cut.
• Each trade page names the questions, documents and limits of that trade, and what the chatbot must never do in it.
• Every page goes through a script that checks it: no two pages may share a title, a description or a short answer, and a list of words that mark text as generated is rejected.
• No statistic, price or result goes on a page unless it comes from a named source.

That is the decision Google's policy asks every site to make: publish a page when it is worth reading, and not before.

AI images and product listings

The same thinking applies to pictures. An AI image of a finished kitchen you never fitted shows customers work you did not do. Use photos of your own jobs, premises and team.

Online shops have firmer rules. Google's guidance says AI-generated product images in Merchant Center must carry the IPTC metadata that marks them as AI-generated, and that AI-generated product titles and descriptions must be labelled as such. If you sell online, check your product feed with whoever runs it. More for online stores.

Two other policies that catch mass-produced sites

Scaled content is not the only spam policy that bulk pages run into:

• Expired domain abuse: buying an expired domain, such as the old address of a closed local business, and filling it with low-value pages to borrow its past standing in search.
• Site reputation abuse: publishing someone else's pages on a site mainly to ride on that site's standing in search. Google updated this policy again in August 2026.

A small business usually meets these as offers: an "aged domain with authority" for sale, or a fee for placing sponsored articles on your site. Turn them down. Google's spam policies say sites that break them can rank lower or drop out of results, and a review by a person at Google can lead to a manual action.

Signs your pages have a problem, and the fix

Look in Search Console for these:

• Many of your pages under "Crawled - currently not indexed", especially pages that look alike.
• Pages that are indexed but get no impressions at all for months.
• A manual action for scaled content abuse in the Manual actions report. This one is rare and serious.

The fix is the same in each case. Merge pages that say the same thing, rewrite the ones worth keeping with real detail, and remove the rest, with a redirect where an old address had visitors or links. A page you need for another reason can stay on the site with a noindex tag. Why your website is not on Google explains the indexing reports.

Should you use AI to write your website?

Yes, as a drafting tool, working from your own facts, with a person who knows the business checking every line. No, as a way to make more pages than you could write yourself. Can AI build your business website covers the rest of what AI does well and badly on a site.

A site with eight pages that each answer real questions does more for a local business than eighty that repeat each other, in search and in front of the customer. If you want to know what those eight should be, what a small business website needs goes through them page by page, and how a local business shows up in Google's AI answers covers the newer search features.