We sell an AI writing tool, so you should read this with appropriate suspicion. We have tried to write the version we would give a friend rather than the version that sells the most subscriptions, and there is a section near the end telling you when not to use something like ours.
The short answer
Google does not penalise content for being written by AI. It has said so plainly, and its published guidance is about whether content is helpful and original, not about how it was produced. What gets penalised is mass-produced material with no purpose beyond gaming search results — and a human can produce that just as easily as a machine.
So the real question is not "will I get caught". It is "will this be any good", and that depends almost entirely on what happens around the writing rather than the writing itself.
The fear, and why it is mostly misplaced
Most business owners asking this question are picturing a penalty — some switch flipped at Google that buries your site the moment a detector spots a machine. That is not how it works, and the AI detection tools you may have seen are unreliable enough in both directions that nobody serious is making ranking decisions with them.
What Google is actually assessing is closer to: does this page answer the question a person typed? Does it contain anything they could not get from the other nine results? Did anyone stay and read it? Those tests are indifferent to who typed the words.
Here is the uncomfortable part, though. A great deal of AI-written content genuinely does fail those tests. Not because it is machine-written, but because it was produced without anyone deciding what the page was for.
Where AI writing actually falls down
In our experience the failures are consistent and they are almost never about grammar:
- It does not know anything specific about your business. Ask a chatbot to write about your service and it will produce a competent article about that service in general. Every real advantage you have — your particular process, your pricing, the thing you do that nobody else locally does — is missing, because you never told it.
- It writes for topics, not searches. Left alone it will happily produce a lovely article about something nobody types into Google. Beautiful writing aimed at zero demand is still worth zero.
- It hedges. Unprompted, AI writing tends toward the safe middle — "it depends on your circumstances", "there are many factors to consider". Readers came for an answer and a hedge reads as evasion.
- It is confidently wrong about details. A statistic, a consent requirement, a price range, a regulation. It will state these with total assurance and be flatly incorrect perhaps one time in ten. This is the genuinely dangerous failure mode, and it is worse here than most places because these tools know very little about New Zealand.
- It has a tell. Certain rhythms and constructions turn up constantly. Once you notice them you cannot stop noticing them, and a small but growing number of your readers already do.
What separates AI content that works from AI content that does not
Almost all of it comes down to what you feed in and what you do afterwards:
- Real search research first. Decide what question the post answers based on what people actually type, before a word is written. This single step is the difference between content that earns traffic and content that sits there.
- Your specifics, not the generic version. Your prices, your region, your process, your examples. The parts that could only have come from your business are the parts that make the page worth existing.
- A destination. Every post should point somewhere — a service page, an enquiry form, something. A post with nowhere to go is a page that gets read and forgotten.
- A human read before publishing. Not a rewrite. A read, by someone who knows the business, checking that nothing is wrong. Ten minutes. This is not optional and anyone telling you otherwise is going to embarrass you eventually.
Do those four things and the writing being machine-assisted stops being the interesting fact about the page. Skip them and no amount of editing rescues it.
When you should hire a person instead
There are cases where we would tell you not to use a tool like ours, and it costs us nothing to be straight about them:
- When the exact wording is the product. Your homepage, your key sales page, your brand voice. If a sentence has to land precisely, pay a copywriter.
- When you are the expert and the expertise is the point. If you have twenty years of specialist knowledge that nobody else in the country has, an AI does not have access to it. Dictate it into your phone and get someone to tidy it up.
- When accuracy is legally or medically consequential. Health, law, finance, safety compliance. The one-in-ten confidently-wrong problem is unacceptable here regardless of review process.
- When you would not read it before publishing. If you genuinely do not have ten minutes per post, do not publish AI-assisted content. Publish less, written by a person.
Will readers be able to tell?
Some will. More every year. But this matters far less than people expect, because readers are not grading your prose — they arrived with a question and they want it answered.
Nobody has ever left a website because the writing was merely competent. People leave because the page did not answer the thing they asked, or because it took four hundred words to get to the point, or because it was obviously written to rank rather than to help. Those are failures of intent, and they are equally common in human writing.
The way we would put it
AI writing is a good answer to "I know what my customers ask and I have no time to write it down". It is a bad answer to "I do not know what to say". The tool solves the second problem badly and the first one very well, and knowing which problem you have is most of the decision.
For a lot of New Zealand small businesses, the real alternative is not a beautifully crafted post by a professional. It is the blog page that has said "Coming soon" for two years. Against that, a well-researched, human-checked, specific post that answers a real question is not a compromise. It is the first thing that has happened at all.