This article is not legal advice. It is an editorial guide built on Google's published documents. The short answer: Google's spam policies do not ban AI on their own. They target pages produced in volume for the primary purpose of manipulating search rankings, without adding value for users. A newsroom that uses AI to help an editor and a site that pushes feeds into thousands of near-empty pages stand on opposite sides of that line.
This guide is for publishers, editors and content managers. For the general workflow of AI-assisted writing, our guide to AI-written articles and human editing comes first; this one is about news sites specifically.
What the spam policy says about scaled content
Start with Google's own wording in the spam policies for Google web search (Opens in a new tab). Scaled content abuse is defined as many pages generated for the primary purpose of manipulating search rankings. The document describes the practice as typically focused on large amounts of unoriginal content that provides little or no value to users, no matter how it is created.
The listed examples
The policy gives examples, and several matter to publishers: using generative AI tools or similar tools to generate many pages without adding value for users; scraping feeds, search results or other content to generate many pages; stitching or combining content from different web pages without adding value; creating multiple sites with the intent of hiding the scaled nature of the content; and creating many pages whose content makes little or no sense to a reader but contains search keywords.
Purpose and value carry the weight
Two phrases do most of the work: "primary purpose" and "without adding value". The text does not say AI is prohibited. It names AI-generated pages with no added value as one example inside a larger description of low-value volume. Whatever the production method, the thing being aimed at is unoriginal, low-value content made in bulk. So the useful question is not "did we use AI?" but "did we make this page for the reader or for the page count?"
How violations are found
The document says Google detects policy-violating practices through automated systems and, as needed, human review, and that human review can result in a manual action. There is no stated page-count threshold, so nobody can honestly say that a given number of generated pages is safe.
The News policies page
The Google News content policies (Opens in a new tab) do not set separate rules for AI-generated, scraped or republished material; they only refer to the Search spam policies. They do say that violating content can be removed from news surfaces, and that repeated or egregious violations can cost a site its eligibility to appear there. Read the two documents together.
For scaled content in general, see scaled content abuse and programmatic SEO risks; this article stays with the feed, wire and editor-control problems of news sites.
Check the current text
Policy wording and examples can change. Open both pages again before you rely on this summary.
A working test: assistance or mass production?
Can a news site use AI for drafts, and where does the limit sit? The documents suggest the limit is in how the tool is used. The test below is our own suggestion and not an official Google classification.
Four questions
- Would this page exist if search engines did not? If the honest answer is no, be careful.
- Has an editor read the full text against its sources before publication?
- Does the page give a reader something the source material alone does not: reporting, context, verification?
- Is the number of pages limited by what editors can actually review, or by what the tool can generate?
Four honest yes answers describe assistance. Two or more no answers describe a drift toward the pattern the policy names.
Google's helpful content guidance
The helpful content guidance (Opens in a new tab) accepts that content can include automated, AI-generated and AI-assisted material. It asks whether the use of automation is evident to visitors through disclosures or other means, and it says using AI primarily to manipulate rankings violates the spam policies. It also describes generating large amounts of text with generative AI without manual oversight or curation as little to no effort, and states that giving credit to other sources does not replace the need for original effort.
| Signal | Looks like assistance | Looks like mass production |
|---|---|---|
| Reason for the page | A reader need | Coverage of keywords or inventory |
| Review | Editor reads text against sources | Nobody reads before publishing |
| Added value | Reporting, local context, verification | Rewording of a source |
| Template | Structure follows the story | One pattern across many topics |
| Volume | Bounded by editor capacity | Bounded by tool speed |
Example scenario: templated school closure pages
Example scenario: A regional site builds one page per school from a data table, each page identical apart from the school name and a few fields, and publishes them without review. The pages offer nothing a visitor could not read in the original table, and nobody has checked them. That is close to the pattern the policy describes. A single reviewed story about closures across the region, written with context and sources, is a different thing, even if software helped to draft it.
Editorial review that actually checks something
Fluent text can still be wrong. Google asks informational content to be factually accurate, so review is the central control, not a formality.
What the editor must verify
- Every factual sentence traces to a primary source or the reporter's own confirmation.
- Names, titles, places, dates and figures match the source.
- Each quote is real and attributed to the right person.
- No detail has crept in that the sources do not contain.
- The headline matches the story.
- The piece adds context beyond the source.
Citing a source is not the same as checking it
The guidance is clear that attribution does not replace original effort. An AI draft that cites sources can still misstate what they say, or cite pages that do not exist. The editor needs to open each source and confirm the claim sits there.
Bylines and invented people
Google names fabricated creator profiles, including made-up names, AI-generated headshots and false credentials, as deceptive. Putting a non-existent reporter's name on an AI draft falls squarely there. A real editor who takes responsibility for the text is both more honest and more accountable. We cover role splits in the writer and editor workflow guide.
| Check | Question | If it fails |
|---|---|---|
| Accuracy | Is each fact in its source? | Verify or cut the sentence |
| Added value | Does the piece go beyond the source? | Hold it, or add reporting |
| Quotes | Are they real and correctly attributed? | Remove them |
| Disclosure | Is AI use stated where readers would expect it? | Add the statement |
| Byline | Does a real, accountable person sign it? | Fix the byline |
Decision rule
A simple rule works: if an editor's name is on a story, that editor has read it and seen its sources. If the team cannot meet the rule at its current output, the output is too high and publishing should slow down. This is our process advice, not a Google requirement.
Wire feeds, RSS and the copying problem
News sites face a risk that most other sites do not: a constant supply of ready-made material. The spam policies list scraping feeds, search results or other content to generate many pages, and stitching or combining content from different pages without adding value, among scaled content abuse examples. Scraping also has its own entry, described as taking content from other sites, often by automated means.
Licensed copy versus scraped copy
A contract with a news agency gives you the right to publish its copy. That right and Google's evaluation are separate matters. A licence grants permission; Google's systems look at what a page offers readers. Contract terms are a legal question.
Ways to add real value
If you carry agency copy, real additions include local context, a reporter's confirmation, links to your own earlier coverage, an explanatory box and filling gaps in the source. Only add these if they are real. Swapping synonyms to reword a source adds nothing.
Networks of sites
The policy also lists creating multiple sites to hide the scaled nature of content. A publishing group that runs the same stories across several domains with light edits should assess itself against that line. Each site needs its own audience and its own contribution.
| Source | Risk | Safer practice |
|---|---|---|
| Licensed wire copy, unchanged | No original contribution | Add local context or verification |
| RSS items auto-published | Scraping and no added value | Turn off auto-publish, route to an editor queue |
| Reworded copy of a rival story | Unoriginal, possibly deceptive | Report and verify independently |
| Same story on several domains | Hiding scale | Produce distinct content per site |
Example scenario: an automated earnings recap
Example scenario: A finance portal publishes a recap for every company in a data feed the moment results arrive, using one template, with no human involvement. Hundreds of near-identical pages appear, each telling the reader only what the feed says. A better design lets the software prepare a draft for a handful of companies the newsroom actually covers, has an editor add context and checks every figure against the filing, and skips the rest.
Disclosure, labelling and a safer workflow
The aim of the workflow is to use AI where it helps without sliding into the pattern the policy describes. Treat the steps below as suggestions to adapt.
Five steps
- Write down what AI may do (summaries, headline options, first drafts) and what it may not (invent facts, quotes or sources).
- Require an editor to read every draft against its sources.
- Record what the story adds beyond the source: reporting, verification, local context.
- Disclose AI involvement where readers would reasonably expect it.
- Sample published pages regularly and fix or remove weak ones.
When to disclose
Google's guidance asks whether the use of automation is evident to visitors through disclosures or other means, and suggests adding such disclosures when it would be reasonably expected. It does not prescribe wording. Decide in your editorial policy where readers would expect to be told, and keep the statement short and plain.
This draft was prepared with an AI tool and edited by our staff against the cited sources.
A label is honesty, not armor. A labelled page that was mass produced without added value is still the kind of page the policy targets.
Pace and capacity
The number of stories you publish a day should not exceed what editors can actually read. The documents give no number; your own capacity sets it. Output that outruns review is a sign the control no longer works.
Cleaning up what already exists
Audit the archive for pages that are thin, repetitive or unread. Mark which ones were produced in bulk, which passed editorial review and which give readers something. Merging, improving or removing weak pages matters as much as producing new ones. If you have already lost traffic, our guide on diagnosing traffic loss after a Google update lays out where to start.
Who does what in the newsroom
A policy is easier to follow when each role knows its part. The split below is a suggestion, and small teams may combine roles as long as the duties stay visible.
The desk editor
The desk editor owns the read-through. Before anything goes live, this person opens the sources, compares facts and decides whether the piece adds anything. The desk editor can stop a story, and nobody should be able to override that without a recorded reason.
The reporter
Reporters supply what a tool cannot: first-hand confirmation, a phone call, a visit, a document obtained. When a story depends on an AI-assisted draft, the reporter's contribution is what lifts it above a rewrite of a source.
The managing editor
The managing editor owns the policy itself: what AI may be used for, what is off limits, what must be disclosed and who approves exceptions. This role also receives reports of breaches and decides what happens next.
The technical owner
Whoever runs the publishing system controls what can be automated. Feeds should never connect to public pages without passing through an editor queue, and the technical owner is the person who makes sure that stays true.
| Role | Owns | Must not |
|---|---|---|
| Desk editor | Source check and publish decision | Sign a story they have not read |
| Reporter | Original confirmation and context | Pass off unverified tool output as reporting |
| Managing editor | Policy, exceptions, disclosure rules | Leave rules unwritten |
| Technical owner | Feed and automation settings | Connect a feed straight to live pages |
A second worked example: a weekend sports digest
Example scenario: A local site wants a weekly digest of amateur league results. Scores arrive in a spreadsheet from the league office. The software drafts a table and a few sentences per fixture. A sports editor checks each score against the league's sheet, adds a line about a notable match from the reporter's own attendance, corrects a misspelled player name and publishes one digest, labelled as prepared with software and reviewed by staff.
The same site could have set up a page for every fixture, filled by script, and published hundreds of them each week without anyone reading. Both versions use the same data. Only the first one is built around the reader, and only the first can be explained to a skeptical outsider in one sentence.
What the example shows
- Automation handles the repetitive part, and a person handles judgment.
- One useful page replaces many thin ones.
- Verification happens against the original data, not against the draft.
- Disclosure is brief and matter-of-fact.
What readers should be able to see
The more openly a page shows how it was made, the easier it is to judge on its merits. This also matches the transparency that Google News asks for, namely clear dates and bylines and information about the authors and publisher.
On-page essentials
- A real byline from an accountable person.
- A clear publication date and, after a meaningful change, an update date.
- Named sources, linked where possible.
- A brief disclosure where AI involvement would be expected to be stated.
- A way to report a mistake.
A last test before publishing
Ask whether a reader would lose anything if the page did not exist. If the honest answer is no, the page probably exists for the count. This is a commonsense check, not a measurement, but it separates reader-first pages from volume pages quickly. When several near-identical pages cover one subject, one strong page usually serves readers better.
Handling weak pages already online
For existing bulk-made pages you have three options: improve them with real value, merge similar ones into a stronger page, or remove them. When you merge or remove, plan where the old addresses should point. Try the change on a small sample first and watch the results before applying it widely.
Records, accountability and expectations
A workflow has to be provable. A year from now you should be able to say how a page was made, who read it and what it relied on.
A production record
Keep a short record per story: the tool used, who prepared the draft, which editor reviewed it, which sources were opened and whether a correction followed. The record makes problems traceable and shows where the process is weak. For service providers and clients who work together on this, the editor review checklist for AI-assisted article writing offers a complementary list.
Who is accountable
When a tool makes a mistake, the publisher is responsible. A model error does not remove the editor's duty to verify. Put that principle in writing and train editors on it.
What to expect
Following the policy gives no ranking guarantee, and Google's documents promise none. What you can do is stay away from the behavior the policy targets and give readers real value. If traffic drops, do not assume AI is the cause: check which pages and sections were hit before drawing conclusions. If you suspect a manual action, the manual actions report in Search Console shows whether one exists; menu names may differ if your interface uses another language.
Common mistakes
- Publishing AI output nobody has read.
- Never opening the sources the tool cited.
- Using a made-up or non-existent byline.
- Wiring an agency or RSS feed straight to auto-publish.
- Treating a label as protection.
- Leaving weak, bulk-made pages in the archive.
A boundary of this guide
Google's documents do not rule on the legal consequences of content production, copyright, personal data or contract duties. For questions in those areas, speak to a qualified professional.
Frequently Asked Questions
Does Google ban AI-generated content on news sites?
The spam policies do not ban AI on its own. Scaled content abuse is many pages generated for the primary purpose of manipulating rankings, and using generative AI to generate many pages without adding value is one named example.
Is it safe to draft with AI and have an editor revise?
The documents do not use the word safe. A draft that an editor reads against its sources, with real added value for readers, differs from the mass production the policy targets. No guarantee can be given.
Do I have to label AI-assisted stories?
Google's helpful content guidance suggests disclosing automation when it would be reasonably expected and does not prescribe a format. A label is for honesty and does not protect low-value bulk content.
Is publishing licensed agency copy unchanged a problem?
A licence gives you permission to publish. Google, however, looks at what a page offers readers, and many pages with no added value approach the scaled content examples. Contract terms are a legal matter; this article is not legal advice.
Can I turn an RSS feed into automatic article pages?
The policy lists scraping feeds to generate many pages as an example of scaled content abuse. We advise against auto-publishing feed items without editor review and original contribution.
How does Google detect violations?
According to the policy document, through automated systems and, as needed, human review, which can lead to a manual action. The document gives no page-count threshold.
Related guides: News Site SEO Checklist, Infinite Scroll or Pagination? A News Archive Guide, How to Write a Corrections and Updates Policy for a News Site.
Conclusion
What decides the matter on a news site is not the tool but the purpose and the value delivered to readers. Google's spam policies aim at many pages produced to manipulate rankings without adding value, and they name generative AI, feed scraping and content stitching among the examples. To stay on the right side, tie AI drafting to editor review, source verification and original contribution, keep agency and RSS feeds away from auto-publish, and disclose AI use where readers would expect it. The documents offer no page-count threshold and no guarantee, and this guide is not legal advice; for contracts, copyright or enforcement in a particular case, consult a qualified professional. One rule to remember: a story an editor has not read and checked against its sources should not go live.
If you want editor-reviewed content support, you can look at the content writing service on Akua Adforce; scope and terms are listed on the service page.
Post a comment
Please log in to leave a comment. Login