How much AI content is acceptable to Google? What the rule actually is
There is no percentage. Google does not publish a threshold, does not run an “AI percentage” check, and has said in plain terms that how a page was made is not the thing it judges. The rule is about the page, not the method, and once you see why a percentage rule is impossible the real test becomes obvious.
Why a percentage cannot exist
A percentage rule would need two things Google has never offered: a reliable way to measure how much of a page was machine-written, and a public cut-off it promises to act on. Neither exists. So-called AI detectors guess at a probability from writing patterns, and they are wrong often enough that no search engine would stake rankings on them. Even if the measurement were perfect, a number tells you nothing about whether the page helps the person reading it. A wholly hand-typed page can be thin and useless; a page drafted with assistance can be the most useful result for a query. Google judges the second thing, not the first.
What Google actually says
The clearest statement came with the March 2024 spam-policy update, which says Google will act on low-value content “no matter whether content is produced through automation, human efforts, or some combination of human and automated processes”. The target is not automation. The target is what Google calls scaled content abuse: “when many pages are generated for the primary purpose of manipulating search rankings and not helping users.” AI makes that easy to do at volume, which is why it gets named, but the abuse is the intent and the output, not the tool.
The test that replaces the percentage
If you want a rule you can actually apply, judge each page on four things, one page at a time:
- Originality. Does this page say something its siblings and competitors do not, or is it a reskin with the town name swapped?
- Purpose. Was it made to help a specific reader, or to occupy a search result?
- Substance. Does it carry real, checkable detail a stranger to the area could not have written, or only generic filler?
- Standalone value. If this were the only page on the topic, would it still earn its place?
A page that passes these is fine whether you typed every word or had help drafting it. A page that fails them is at risk however it was made. That is the whole rule, and it scales to forty pages because you ask it of each one.
Where the risk really sits for location pages
For a business with one page per town, the danger is almost never “too much AI” in the abstract. It is near-duplication: pages that are 90% identical to each other with a swapped place name, which is exactly the pattern scaled content abuse describes. That is measurable. You can compare each page against its closest sibling and see how much text they share, long before Google folds them together. The duplicate-content guide covers how that measurement works, and the scaled content policy explained in plain English covers the intent side. The fix is the same either way: fewer pages, each carrying real local substance.
How Townsmith approaches this
Townsmith ships no AI and makes no external calls. It will not write a page for you and it cannot phone a model. What it does instead is score every page out of 100 before you publish, on the things the four-part test cares about, and warn you when a page leans on its template instead of standing on its own. The quality score documentation sets out exactly what it measures. The advisory gate is just that, advisory: it never blocks you, it tells you which pages are thin and which sit too close to a sibling so you can fix them before search engines do. The optional Pro assistant is off by default, assists rather than generates, and works only from facts you have already typed about an area. It never invents a claim about a place you have not described. The features page lays out the full split between the free engine and Pro.
Common questions
Is there a safe percentage of AI content?
No. Google publishes no threshold and runs no AI-percentage check, because how a page was made is not what it ranks. Judge the page on originality, purpose, substance and standalone value instead.
Will Google detect that I used AI?
Treat that as the wrong question. Detectors are unreliable, and Google has said the method does not matter. A thin, duplicate-heavy page is the risk whether a person or a model produced it; a genuinely useful page is fine either way.
If a tool gives my page a confident-looking AI score, does that mean trouble?
A figure that looks authoritative, such as a percentage from a detector, is a guess about writing style, not a verdict from Google and not a measure of whether the page helps anyone. It tells you nothing about ranking risk. The four-part test does.
Can I use AI to help write location pages at all?
Yes, if each finished page is original and has real local substance. The trouble starts when assistance is used to mass-produce near-identical pages at a scale you could never justify one by one. See the guide on AI content and location-page penalties for where assistance turns into abuse.
Does hand-writing every word keep me safe?
Not on its own. Forty hand-typed pages that all say the same thing with the town name swapped are still scaled content abuse. Originality and real local substance are what protect you, not the absence of a tool. The E-E-A-T guide covers how to show that substance credibly.