The rule is about quality, not authorship
Google’s stated position is that it rewards helpful, reliable content made for people, however it was produced. Using automation to generate pages purely to manipulate rankings is against its guidance — and that has been true since long before generative AI, because it describes spam rather than a technology.
So “will I be penalised for using AI” is the wrong question. The right one is: does the page I published deserve to rank? Unedited AI output usually does not, and the reason is structural rather than moral.
Why unedited AI output ranks poorly
A language model produces the most probable next words based on what already exists. That is an excellent way to produce something average and a terrible way to produce something distinctive.
Three consequences follow, and each has a direct cost:
- It says what every other page says. If your page is a competent summary of the top ten results, there is no reason to rank it above them. You have added nothing to the index.
- It contains no first-hand knowledge. It cannot tell the reader what happened when you tried this with a client in Kathmandu, because it was not there. That specificity is the thing a reader remembers and a competitor cannot copy.
- It states things confidently that are wrong. Invented statistics, misremembered platform features, features that were removed two years ago. On a marketing page this is embarrassing; on anything touching money, health or law it is a serious risk.
The third one deserves emphasis because it is the most common way AI content actually damages a business. A fabricated figure on your site is a claim a prospect can check, and being caught costs more than the page ever earned.
Where AI genuinely helps
Used as an assistant rather than an author, it is very good at several things:
- Research and structure. Getting from a blank page to an outline, listing the questions a topic should answer, finding the angle you had not considered.
- Editing your own writing. Tightening a draft you wrote, finding the paragraph that repeats itself, flagging jargon. This keeps your thinking and removes your bad habits.
- Repetitive variations. Forty ad headlines within a character limit, meta descriptions for pages you have already written, alternative subject lines. Judgement still picks the winner.
- Understanding rather than producing. Summarising a long report, explaining an unfamiliar technical concept, translating a client’s rambling brief into requirements.
- Genuine automation. Routing enquiries, drafting replies for a human to approve, extracting data from documents. This is where the real time saving is, and it never touches a public page.
Notice the pattern: AI is useful where the output is checked by someone who knows the subject, and risky where it is published directly.
Rules worth keeping
Never publish a figure you did not verify
If a draft contains a percentage, a statistic or a “studies show”, either find the source and link it or delete the sentence. Models invent these fluently and they are the single most damaging thing to let through.
Add what the model cannot have
Before publishing, ask what is in this page that could only come from us. A real example, an opinion, a number from our own accounts, something we learned the hard way. If the answer is nothing, the page is not ready — and that is true whether a person or a model wrote it.
Keep one writer’s judgement in the loop
Someone who knows the subject has to read every published word and be willing to defend it. Not a proofread for grammar — a check that the claims are true and the advice is what you would actually give.
Do not scale publishing just because you can
The temptation is a hundred pages a month. Volume without quality is the exact pattern Google’s guidance describes, and it has a second cost people forget: your own pages start competing with each other for the same searches, and your good page loses to your mediocre one.
Be careful with anything consequential
Advice about money, health, legal matters or safety carries a higher standard, for good reason. If a page could cause harm by being wrong, a qualified human writes and reviews it.
What about AI detectors?
They are unreliable in both directions — they flag human writing and clear genuinely machine-written text — and more importantly, Google has not said it uses them as a ranking signal. Writing to defeat a detector is optimising for the wrong examiner.
The useful test is not “would this be detected” but “would a knowledgeable reader find this worth their time”. That test is harder to pass and it is the one that correlates with ranking.
A workflow that works
- Decide the page’s single purpose and who it is for. A human decision.
- Use AI to list the questions a good page would answer, then cut and add from your own knowledge.
- Write the sections that need your experience yourself. These are the reason the page exists.
- Use AI to draft the connective and explanatory sections where being standard is fine.
- Edit the whole thing in your own voice, removing anything you could not defend.
- Verify every factual claim, and delete any you cannot source.
- Add the specifics: your example, your screenshot, your opinion.
That is slower than generating a page in a minute and faster than writing from nothing, which is roughly the honest value of the technology today.
The short version
AI does not hurt your rankings. Publishing average, unverified pages hurts your rankings, and AI makes it very easy to do that at scale. Use it for the work around the writing, keep a knowledgeable human responsible for what goes live, and never let it invent a number.
If you want help working out where automation genuinely saves you time — usually in operations rather than in publishing — get in touch.