No. Google does not penalise content for being written with AI. It penalises pages produced at scale for the primary purpose of manipulating search rankings rather than helping people, and its spam policy applies that test the same way whether a human or a model typed the words. The method is not the violation. The intent and the value are.
By Cody Wise, Founder of Wise Media. Published 30 September 2026.
Summary
- There is no AI penalty. Google’s published guidance on generative AI content says automation applied to produce helpful, original content is not against its policies.
- There is a scale penalty. The spam policy Google calls scaled content abuse covers many pages made mainly to manipulate rankings, and names generative AI as one way people do it.
- The trigger is volume plus low value, not the tool. Ten thin AI pages are safer than a thousand. A thousand genuinely useful pages are safer than ten thin ones.
- Google ran four spam updates in 2026 according to its own Search Status Dashboard: March, June, August and one that began 24 September 2026.
- Disclosure is a stated expectation. Google’s own content self-assessment asks whether your use of automation is self-evident to visitors.
- The practical rule: if a named person cannot defend every claim on the page, do not publish it.
Table of Contents
- What Google’s policy actually says
- What scaled content abuse means in practice
- The 2026 spam update timeline
- Can Google detect AI writing?
- Two ways Canadian businesses trip this policy without noticing
- A publishing workflow that survives a spam update
- Common mistakes
- Using AI for content: pros and cons
- Frequently asked questions
What Google’s policy actually says
Google publishes a dedicated page on generative AI content. Its position is that appropriate use of AI is not against its guidelines, and that its systems reward high quality content however it was produced. The document does not describe an AI detector, an AI demotion, or a ranking discount applied to machine-written text.
What the same page does say is that using generative AI tools to generate many pages without adding value for users may violate the spam policy on scaled content abuse. That is the sentence every argument about this topic should start from, because it names the actual condition: many pages, and without adding value. Neither condition is about the tool.
The three tests Google applies
| Test | Question Google asks | What fails it |
|---|---|---|
| Purpose | Was this made mainly to help people, or mainly to rank? | Pages built around a keyword with no reader in mind |
| Value | Does it add original information, analysis or first-hand experience? | A rewrite of the current top three results |
| Scale | Are many near-identical pages being generated at once? | Programmatic city, service or product pages with swapped nouns |
Google’s guidance on helpful content adds a fourth practical expectation. In its self-assessment questions it asks whether the use of automation, including AI generation, is self-evident to visitors through disclosure or in other ways, and it asks whether the content demonstrates first-hand expertise such as having actually used a product or visited a place. A model cannot do the second thing. That gap is where most AI-first content loses.
What scaled content abuse means in practice
Google defines scaled content abuse as many pages generated for the primary purpose of manipulating search rankings and not helping users. The policy page lists the patterns it covers, and only one of them mentions AI at all.
- Using generative AI tools to generate many pages without adding value for users
- Scraping feeds or search results and running automated transformations such as synonymising or translating
- Stitching or combining content from different pages without adding value
- Creating multiple sites to disguise the scaled nature of the content
- Generating many pages that make little sense but contain search keywords
Read that list again with a marketing budget in mind. Four of the five patterns predate AI by a decade. Article spinning, scraped feeds and doorway pages were the same violation in 2014. Generative AI did not create a new category of spam. It lowered the cost of the old one, which is why enforcement got louder rather than why the rule changed.
Algorithmic demotion versus a manual action
Two different things get called a penalty and they behave differently.
| Algorithmic | Manual action | |
|---|---|---|
| How you find out | Traffic falls, nothing is reported | A notice appears in Google Search Console |
| What it hits | Pages or patterns, often partially | Often the whole site |
| Recovery | Fix the content, wait for reprocessing and the next update | Fix, then submit a reconsideration request |
| Typical trigger | Quality systems and core updates | A reviewer confirming a policy breach |
If your traffic dropped and Search Console shows no manual action, you are looking at a quality or a relevance problem, not a punishment. That distinction decides the whole recovery plan. Our guide to local pack versus AI search visibility covers the other common cause of a drop that looks like a penalty and is not one.
The 2026 spam update timeline
Google records every confirmed ranking update on its public Search Status Dashboard. Here is 2026 to date, which matters because it tells you whether a traffic change lines up with an enforcement event or with something on your own site.
| Update | Type | Started | Rollout |
|---|---|---|---|
| March 2026 spam update | Spam | 24 March 2026 | About 19 hours |
| March 2026 core update | Core | 27 March 2026 | About 12 days |
| May 2026 core update | Core | 21 May 2026 | About 12 days |
| June 2026 spam update | Spam | 24 June 2026 | About 2 days |
| August 2026 spam update | Spam | 18 August 2026 | About 2 days and 16 hours |
| September 2026 spam update | Spam | 24 September 2026 | In progress at time of writing |
Four spam updates in nine months is a faster cadence than the two or three a year Google ran before generative writing tools were widely available. Take the pattern seriously, but check dates before you blame one. A drop that starts a week before an update is not caused by that update.
Can Google detect AI writing?
The honest answer is that it does not need to. Google has never published an AI detection score, and no public documentation describes one. What its systems do measure is much easier: whether a page contains information that exists nowhere else, whether anyone links to it or cites it, whether visitors stay and act, and whether a site suddenly starts producing hundreds of similar pages.
Third-party AI detectors are not evidence of anything either. They produce false positives on edited human writing and false negatives on lightly rewritten model output, and Google does not use them. Optimising to beat a detector is optimising for the wrong test. The real test is whether the page contains something the model could not have known.
Two ways Canadian businesses trip this policy without noticing
Programmatic city pages
The most common version in Canada is a service business generating one page per city or neighbourhood: plumber in Calgary, plumber in Airdrie, plumber in Okotoks, each identical except for the place name. That is the textbook shape of scaled content abuse, and it is generated far faster now than it used to be. Location pages are not banned. Location pages that differ only by a swapped noun are exactly what the policy describes.
A location page earns its place when it carries something true only of that location: the actual service area boundary, local permit or licensing detail, response times, named projects, local pricing conditions. If you cannot write three sentences that would be false in the next town over, the page should not exist as a separate URL.
Machine-translated French
Canadian businesses serving Quebec or bilingual markets often run the whole English site through machine translation and publish it as a second language tree. Google’s spam policy names automated translation of scraped or existing content as one of the scaled content patterns. Machine translation as a first draft that a fluent human then edits is normal practice. Machine translation published untouched at scale is the pattern the policy is written about, and it also reads badly to the customers you are trying to win.
A publishing workflow that survives a spam update
Use AI where it is strong and a human where the risk is. Seven steps, and the order matters.
- Start from a real question. Pull it from sales calls, support tickets or a forum thread in your industry, not from a keyword tool alone. A question nobody asks does not need a page.
- Gather primary sources first. Government pages, standards bodies, official documentation, your own data. Do this before any drafting, so the draft is built on facts rather than decorated with them afterwards.
- Add what only you know. The job you quoted last month, the failure mode you see constantly, the number your own projects produce. This is the part a model cannot fabricate and the part that makes the page citable.
- Draft with AI if it helps. Structure, transitions and first passes are fine. Treat the output as a draft from a fast junior writer who has never met a customer.
- Verify every claim. Every figure, date, statute and price gets checked against the source and linked. If a claim cannot be sourced, cut it rather than soften it.
- Put a named human on it. A real byline with a real role. Google’s expertise questions ask whether it is self-evident who authored the content.
- Publish deliberately, not in bulk. Two defensible pages a week beats two hundred in a weekend, and the second pattern is the one the policy is looking for.
This workflow also happens to be what gets a page cited by AI answer engines rather than just indexed by Google. Models summarising a topic reach for sources that state specific, verifiable, attributable facts. Generic text is the least citable thing on the internet. If you want the fuller picture of how that side works, see our guides on configuring AI crawler access in robots.txt and fixing incorrect information about your business in AI answers.
Common mistakes
- Publishing volume to hit a content quota. A number of posts per month is an input target. It rewards the exact behaviour the policy punishes.
- Letting AI invent statistics. Fabricated figures and imaginary studies are the fastest way to lose trust with both readers and reviewers. Every number needs a source you have actually opened.
- Removing the author. Publishing under a company name or a bare username removes the one signal that costs nothing to provide.
- Rewriting the current top result. If your page is a synthesis of the first three results, it adds nothing and will be treated accordingly.
- Assuming a drop is a penalty. Check Search Console for a manual action first. Most drops are not penalties.
- Deleting everything after a decline. Mass deletion destroys pages that were working. Audit, improve the weak ones, consolidate overlaps.
- Hiding the AI involvement. Google’s own self-assessment asks whether automation use is self-evident to visitors. A short editorial note costs you nothing.
Using AI for content: pros and cons
| Pros | Cons |
|---|---|
| Cuts drafting time so more budget goes to research and editing | Confident fabrication of facts, quotes and sources |
| Consistent structure, headings and internal linking at scale | Produces the average of what already exists, which is the opposite of original |
| Makes translation, repurposing and formatting cheap | No first-hand experience, which Google explicitly asks for |
| Good at outlines, FAQs and schema scaffolding | Tempts teams into volume, which is the actual policy risk |
If your traffic already dropped: a nine-point audit
- Check Search Console for a manual action. This is the first click, not the last.
- Compare the drop date against Google’s Search Status Dashboard rather than a third-party volatility tracker.
- Separate impressions from clicks. Flat positions with falling clicks is a snippet or SERP feature problem, not a quality problem.
- List every page published in the six months before the drop and count how many were near-duplicates.
- Identify pages with zero clicks and zero links over twelve months. They are the audit candidates.
- Check for unsourced statistics and dead outbound links across the affected section.
- Confirm every page has a named author and a visible last-updated date.
- Consolidate pages that compete with each other into one stronger page with redirects.
- Improve first, delete last. Deletion is the right call only for pages with no reader intent behind them at all.
Frequently asked questions
Does Google penalise AI-generated content?
No. Google’s published guidance states that appropriate use of AI is not against its guidelines and that its systems reward quality however content is produced. What is penalised is scaled content abuse, which Google defines as many pages made primarily to manipulate rankings rather than help users. AI is named as one method of doing that, not as the violation itself.
Do I have to disclose that I used AI?
Google does not require a disclosure label to rank, but its own content self-assessment asks whether the use of automation is self-evident to visitors, and it recommends explaining how content was created where that makes sense for your audience. A one-line editorial note stating that drafts are AI-assisted and reviewed by a named person satisfies the spirit of it and costs nothing.
How many AI-assisted posts can I publish before it becomes scaled content abuse?
There is no published number, because the test is value and intent rather than count. A useful working rule: if you could not defend each page individually in front of a customer, you are publishing too fast. Volume is only a signal because low value at volume is what the policy is written to catch.
Can Google tell if content was written by ChatGPT?
Google has never published an AI detection signal and its documentation describes none. Its systems measure originality, usefulness, citations and site-level patterns instead. Third-party AI detectors are unreliable in both directions and are not used by Google, so a detector score is not a ranking diagnosis.
Are AI-generated location or city pages allowed?
Only if each page carries information that is genuinely specific to that location. Pages that differ only by a swapped place name match Google’s description of many near-identical pages generated to manipulate rankings. Fewer, deeper location pages outperform a long tail of templated ones in both search and AI answers.
Was the September 2026 spam update about AI content?
Google confirmed a spam update beginning 24 September 2026 on its Search Status Dashboard but does not publish the specific targets of each spam update. Assume it enforces the existing spam policies, including scaled content abuse, rather than introducing a new rule. Check your dates against the dashboard before attributing any change to it.
Should I delete my old AI-written posts?
Improve them first. Add primary sources, first-hand detail and a named author, then consolidate anything that competes with another page. Delete only pages that no reader would have a reason to land on. Mass deletion after a drop usually removes pages that were still earning.
The short version
Stop asking whether AI wrote it. Ask whether the page contains anything that was not already on the internet, and whether a named person is prepared to stand behind every number on it. Those two questions predict how a page performs through a spam update far better than any rule about tools.
The businesses losing ground in 2026 are not the ones using AI. They are the ones that used it to publish faster without changing anything about how they research, source or edit. The gap between those two behaviours is the entire content strategy now.
Want content that holds up through the next update?
We build search and content systems for Canadian businesses: primary-source research, named authorship, schema, internal linking and a publishing cadence that compounds instead of triggering enforcement. Tell us what you are working on through the Wise Media intake form, or look at the website growth packages first. If you are still deciding how to build the site itself, our comparison of an AI website builder versus a web design agency in Canada is the better starting point.
Sources
- Google Search Central, guidance on generative AI content
- Google Search Central, spam policies for Google web search
- Google Search Central, creating helpful, reliable, people-first content
- Google Search Status Dashboard, ranking update history
Editorial note: Wise Media uses AI assistance in research and drafting. Every article is structured, fact-checked against primary sources and edited by a named human before it is published.