Using AI for content marketing is fine, but publishing large volumes of it with nobody adding anything is a liability, and the risk is bigger for a small business than for a large one.
Google does not penalise AI content as such. It penalises scale without value, and a small site that fills itself with generic pages has little else to rank on.
This article sets out what Google says, what the evidence on AI-assisted content shows, where we use AI in content marketing ourselves, and the list of things we will not publish whatever the tool can produce.
Does Google penalise AI-generated content?
No. Google's rule is about what the content does for the reader, not how it was made.
Its spam policy on scaled content abuse describes the problem as "creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created".
It lists "using generative AI tools or other similar tools to generate many pages without adding value for users" as an example. The words that matter are "without adding value". You can read the full wording in Google's spam policies for Search.
The enforcement that follows is blunt. Google says sites that break its spam policies may rank lower or not appear at all, and it advises hosting providers of such content to exclude it from Search.
So the question is never "was a model involved?" It is "would a reader miss this page if it were gone?"
What does the evidence say about AI content and rankings?
It says AI content is everywhere in the results and does not hurt by itself, which is exactly why the volume approach looks safe until it is not.
An Ahrefs study published in July 2025 looked at the top 20 results for 100,000 random keywords, about 600,000 pages. It found that roughly 86.5% of top-ranking pages contained some AI-generated content, and the correlation between AI use and ranking position was close to zero, at 0.011.
The authors noted that pages in position one tended to have a little less AI content, and that fully AI-generated pages showed up in the top 20 but rarely at number one. The measurements came from the company's own detector and show correlation, not cause. Search Engine Journal covered the findings in its report on the Ahrefs study.
Read that carefully. It tells you that AI-assisted pages can rank. It does not tell you that a site with 400 near-identical AI pages will, and it says nothing about whether those pages bring in customers.
Why is volume the risk for a small business?
Because a small site's whole case for ranking rests on being specific, and volume dilutes it.
A large publisher can absorb a few hundred thin pages. A 20-page service site that adds 300 generic posts has changed its ratio of useful to forgettable content from nearly all to nearly none. Google's own guidance says "a high quantity of pages doesn't make a website higher quality or more relevant to users".
The same guidance, last updated in July 2026, describes what to aim for as non-commodity content: material with "unique expert or experienced takes that go beyond common knowledge". It tells publishers to avoid recycling content that "could easily be produced by a generative AI model". That is the standard in Google's guide to generative AI features in Search.
A model with no access to your business can only produce the commodity version. Everything that makes a post worth reading, such as a job you did, a figure from your own books or a mistake you made, has to come from you.
We have written elsewhere about how to tell whether your SEO agency is using AI to write your content, and what to ask them.
What else can go wrong besides rankings?
Four things: wrong facts, invented sources, a voice that is not yours, and personal data in the wrong place.
Wrong facts are the obvious one. A model states a fee, a deadline or a regulation with confidence, and it is out of date or invented. If your page says it, you said it.
For clinics, where a post about a treatment can be read as advice, the stakes are higher, which is why healthcare marketing content needs a clinician to check every claim.
Invented sources come next. Models produce plausible statistics and case studies with no origin. A single fake figure that a reader checks and cannot find costs more credibility than a hundred accurate posts earn.
Voice is quieter. A site where every post opens with a question, offers "a comprehensive guide" and ends with a neat summary reads as machine-made within a paragraph, whoever actually wrote it. Readers who do not trust the text do not trust the firm.
Personal data is the one people forget. Pasting client emails or case notes into a public AI tool to "write a case study" is a data protection issue, covered in our guide to UK GDPR and AI tools.
What do we use AI for in content?
We use it for research, outlines, first drafts, structural edits and checks, and every published piece is read, corrected and signed off by a person.
In practice that means a model summarising source documents and call transcripts, suggesting an outline we then change, drafting sections from notes that a person wrote, flagging sentences that match known machine-writing patterns, and checking links and figures against the sources they came from. It also helps repurpose a finished piece into other formats.
None of this replaces the interview. Our content marketing work starts with the person who knows the subject and ends with an editor who has read every line.
How much does AI actually save on content?
It saves time on the first draft and almost none on the parts that make a post worth reading.
A model can turn a page of notes into a rough draft in a minute. But a post that carries your own figures, a real job and a point of view still needs someone to supply those, someone to check them and someone to edit the result so it sounds like a person.
For that kind of piece, the interview and the edit are most of the work, and they stay.
The saving is real where the content is repetitive: reformatting a finished article for email or social, summarising a long call, drafting the metadata, tidying references. Those are the jobs we hand to a model. Handing it the thinking is where the cheap version of content marketing starts, and where the liabilities above come from.
What do we never publish?
We never publish unreviewed AI text, invented evidence, or anything that implies a person or result that does not exist.
The list is short and specific:
- statistics, quotes, studies or case studies we cannot trace to a real source
- client results, testimonials or names that a client has not approved
- AI-written posts under a named expert's byline when the expert did not read them
- bulk pages generated from a template with the town or service swapped
- AI-generated photos presented as real staff, premises or completed jobs
- advice in regulated areas, such as health, tax or legal, that a qualified person has not checked
The sixth matters to accountants and professional services firms, whose readers act on what they find. A tax tip that was true in 2023 and not now is a client complaint waiting to happen.
A final point on disclosure. Google's rules turn on the value of a page, not on how it was made, but your readers may care how a piece was made, particularly for health, money or legal topics.
Many sites now say plainly that AI tools were used for research or drafting and that a named person reviewed the result. If you would be uncomfortable saying that on the page, it is a sign the process needs to change.
How can you test your own content before you publish it?
Ask five questions of every piece, and do not publish if any answer is no.
- Does it contain something only our business could say, such as a number, a case, a decision or a mistake?
- Can every figure and claim be traced to a source or to our own records?
- Has a person who knows the subject read the final published version?
- Would we be comfortable if the reader knew exactly how it was produced?
- Is this page here because a customer needs it, or because it was cheap to make?
If you run these on your existing posts, you may find that the answer is no for a good share of them.
Where does AI content belong in a wider AI plan?
Near the back, behind the automations that save time without putting anything in front of customers.
Ahead of it come the jobs that save time without putting anything in front of customers, such as enquiry replies and invoice chasing. The wider question of what agencies mean by "AI" is in our AI agency guide.
If you want AI inside your content process without the risks above, our AI automation services cover the workflow side, and the team can talk you through what we would keep human.
