Guide
Does Google penalize AI content?
Google’s guidance says “Appropriate use of AI or automation is not against our guidelines.” It also says that using automation, including AI, to generate content “…with the primary purpose of manipulating ranking in search results is a violation of our spam policies”. Its spam policies name scaled content abuse, and list “Using generative AI tools or other similar tools to generate many pages without adding value for users” as an example. uprank checks each draft against questions like Google’s, with 13 hard gates including Information gain (at least 70) and Evidence (at least 85), before a version can be approved or sent to your CMS, but it does not test pages against Google.
Updated
What Google says about AI content
Google’s own words, copied exactly from its Search Central pages. Read 2026-10-16; sources are listed below.
Using AI to write is not the test
“Appropriate use of AI or automation is not against our guidelines.”
“…not all use of automation, including AI generation, is spam”
“Our focus on the quality of content, rather than how content is produced, is a useful guide…”
“Using AI doesn't give content any special gains. It's just content.”
Where Google draws the line
“…with the primary purpose of manipulating ranking in search results is a violation of our spam policies”
“Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users.”
“Using generative AI tools or other similar tools to generate many pages without adding value for users”
“Sites that violate our policies may rank lower in results or not appear in results at all.”
How uprank’s quality gates map to Google’s questions
Google asks the question. uprank checks something close to it before a version can be approved. Each row also says what the check cannot do.
Is it original, or only a summary?
Google asks
“Does the content provide original information, reporting, research, or analysis?”
Creating helpful, reliable, people-first content “Are you mainly summarizing what others have to say without adding much value?”
Creating helpful, reliable, people-first content Google also says
“Does the content provide insightful analysis or interesting information that is beyond the obvious?”
Creating helpful, reliable, people-first content What uprank checks
Information gain must score at least 70. A model reads the article text, only the first part of a long article. It is asked whether it adds original value beyond a generic summary, such as a framework, example, comparison or first-hand perspective. Separately, the draft must declare one information-gain element, such as a checklist, comparison or real example. A version with none fails a hard gate.
Gates that block a version
- Information gain ≥ 70
- At least one information-gain element
What it cannot do
The writing model declares the element, so the gate checks that one is declared, and the score is a model’s opinion of how the article reads. Neither shows that the research or experience behind it is yours. Add your own data, examples and experience when you review.
Is it clearly sourced?
Google asks
“Does the content present information in a way that makes you want to trust it, such as clear sourcing, evidence of the expertise involved, background about the author or the site that publishes it, such as through links to an author page or a site's About page?”
Creating helpful, reliable, people-first content Google also says
“Attribution or giving credit to other sources doesn't replace the need for original effort.”
Creating helpful, reliable, people-first contentHow Search Quality raters assess effort. What uprank checks
Research runs before drafting and builds an Evidence Pack. The draft is written from that pack and cites its sources by number. After drafting, claims are pulled out of the draft and each is checked against the source excerpts. Evidence must score at least 85. Claims count by risk: direct support counts in full, partial support counts less, and unsupported or contradicted claims count for nothing. The score also depends on how many credible sources back the article. With 3 or more sources that have a credibility score of at least 40 it keeps its full value. With fewer it is scaled down by up to 30 percent. A citation to a source outside the pack counts as invented and fails a hard gate. Every cited link is checked each time the article is scored. A broken one fails a hard gate.
Gates that block a version
- Evidence ≥ 85
- No invented citations
- No broken required links
What it cannot do
Claims are checked against the sources uprank gathered, so a wrong source can still support a claim. Citing a source is not the same as adding to it. Read the sources for anything that matters.
Is it free of easily-verified factual errors?
Google asks
“Does the content have any easily-verified factual errors?”
Creating helpful, reliable, people-first content What uprank checks
Each extracted claim gets a risk level. The extractor is told that medical, financial, legal, safety and reputation claims are at least high risk. A high-risk claim that the sources do not support, or that they contradict, fails a hard gate. Risk must score at most 20, and each unsupported high-risk claim raises it. A claim the model marks as time-sensitive gets a freshness window in days, counted from when the claim was checked. If the version is scored again after that window, the claim counts as stale and fails a hard gate. This does not catch a fact that was already out of date when it was checked.
Gates that block a version
- High-risk claims are supported
- Risk ≤ 20
- Time-sensitive claims are fresh
What it cannot do
Claims are checked against gathered sources, not against the world. A model extracts the claims, and a claim it misses is not checked. A page with no extractable claims gets a low Evidence score rather than a pass. Read the draft for errors yourself.
Will the reader leave feeling they have learned enough to help achieve their goal?
Google asks
“After reading your content, will someone leave feeling they've learned enough about a topic to help achieve their goal?”
Creating helpful, reliable, people-first content What uprank checks
Practical value must score at least 75. A model is asked whether a reader can act on the article. It rewards actionable steps, decision criteria and usable specifics over abstraction. Search intent is scored by a model as well, on whether the article answers the searcher’s primary question directly and completely. It has no gate and only feeds the overall score.
Gates that block a version
- Practical value ≥ 75
What it cannot do
Both scores are a model’s opinion of one article. They do not measure what readers do, and they are not a stand-in for asking someone in your audience.
Did a person review it, or is it unreviewed output?
Google asks
“Is this content written or reviewed by an expert or enthusiast who demonstrably knows the topic well?”
Creating helpful, reliable, people-first content Google also says
“…using generative AI to produce large amounts of text without manual oversight or curation represents little to no effort”
Creating helpful, reliable, people-first contentHow Search Quality raters assess effort. What uprank checks
A draft the pipeline scores lands in Needs Review whether or not it passed. Approving a version needs the Editor role or higher and a version that has cleared all 13 hard gates. Pushing a version to Strapi checks the same gate. So does recording a live publication request over MCP. The MCP server has no approve tool, so approval is something a person does in the web app.
Gates that block a version
- All 13 hard gates, before approval
What it cannot do
The Editor role is a permission, not proof that the person knows the topic, and uprank does not require a second reviewer. It cannot tell whether the approver read the draft. The review is still yours.
Is it one of many pages made for search?
Google asks
“Are you using extensive automation to produce content on many topics?”
Creating helpful, reliable, people-first content Google also says
“Using generative AI tools or other similar tools to generate many pages without adding value for users”
Spam policies for Google web search “…creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created”
Spam policies for Google web search What uprank checks
Each version is scored and gated on its own, so every page has to clear the Information gain, Evidence and Practical value gates, and no other article on the same site may use the same title. None of the quality gates counts how many pages you publish.
Gates that block a version
- Information gain ≥ 70
- Evidence ≥ 85
- Practical value ≥ 75
- Unique title
What it cannot do
Whether a set of pages is scaled content abuse depends on why the pages were made, and no gate can judge purpose. uprank does not judge your publishing volume, your topic choice or your intent. That judgement is yours, using Google’s questions.
The full list of gates, with weights and thresholds, is on quality gates. Checking claims against sources is how to fact-check AI content, and sourcing is how to add citations to AI-generated content.
What uprank will not do
- Test a page against Google. The gates are uprank’s own and ask questions close to the ones Google publishes. Passing them says nothing about how any search engine will treat a page.
- Decide whether your site breaks Google’s spam policies. Scaled content abuse turns on purpose and on value to readers, and the gates check each article by itself. A set of pages that each pass can still be a set you should not publish.
- Add a byline or an AI disclosure. The Strapi push sends the title, slug, body, excerpt and SEO fields, and no author. If Google’s questions about who made the page and how matter for your readers, add that in your CMS or template yourself.
- Check who reviews. Approval needs the Editor role, and nothing checks that the approver knows the topic or read the draft.
- Prove a claim true. Claims are checked against the sources uprank gathered, and a model extracts the claims.
- Replace a reviewer. Every score on this page is automated, and the Information gain and Practical value scores come from a model’s judgement.
A checklist for AI-assisted content
Works with any AI tool. Each step carries the Google lines it comes from.
Step 1
Write down why the page exists
Name who the page helps and what they can do after reading it. If the honest answer is to attract search visits, do not publish it.
“…creating content primarily to help people…”
Creating helpful, reliable, people-first content Step 2
Add something a summary would not
Add original information, analysis, an example or experience of your own, then reread the page against Google’s question.
“Are you mainly summarizing what others have to say without adding much value?”
Creating helpful, reliable, people-first content Step 3
Source the claims and fix the errors
Tie each factual claim to a source you have read, and fix or cut any claim you cannot support.
“…clear sourcing, evidence of the expertise involved…”
Creating helpful, reliable, people-first content “Does the content have any easily-verified factual errors?”
Creating helpful, reliable, people-first content Step 4
Have a person who knows the topic review it
Name the person who read the final version. A review of the prompt does not count as a review of the page.
“Is this content written or reviewed by an expert or enthusiast who demonstrably knows the topic well?”
Creating helpful, reliable, people-first content “…using generative AI to produce large amounts of text without manual oversight or curation represents little to no effort”
Creating helpful, reliable, people-first contentHow Search Quality raters assess effort. Step 5
Say who made it and how
Add a byline where readers would expect one, and a note on how AI was used where they would expect that.
“Do pages carry a byline, where one might be expected?”
Creating helpful, reliable, people-first content “Is the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways?”
Creating helpful, reliable, people-first content “AI or automation disclosures are useful for content where someone might think "How was this created?"”
Google Search's guidance about AI-generated content “Consider adding these when it would be reasonably expected.”
Google Search's guidance about AI-generated content “Giving AI an author byline is probably not the best way to follow our recommendation…”
Google Search's guidance about AI-generated content Step 6
Check each page on its own when you publish many
A page made the same way as many others still has to pass the steps above on its own. If it does not, it is not ready.
“Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users.”
Spam policies for Google web search
Who approves which version is the AI content approval workflow, and keeping the voice consistent is how to keep AI content on brand.
Common questions
- Does Google penalize AI content?
- Google’s guidance on AI-generated content says “Appropriate use of AI or automation is not against our guidelines.” It also says that using automation, including AI, to generate content “…with the primary purpose of manipulating ranking in search results is a violation of our spam policies”. On sites that violate its spam policies, Google says “Sites that violate our policies may rank lower in results or not appear in results at all.” It says scaled content abuse is typically focused on “…creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created”.
- Is AI content bad for SEO?
- Google says “Using AI doesn't give content any special gains. It's just content.” It also says “Our focus on the quality of content, rather than how content is produced, is a useful guide…” So Google’s stated test is the quality of the content, not the tool that wrote it. Google also tests the purpose, as Google’s spam policies say. Asked whether to use AI to generate content, Google says: “If you see AI as an essential way to help you produce content that is helpful and original, it might be useful to consider.” It continues: “If you see AI as an inexpensive, easy way to game search engine rankings, then no.” This page does not predict how any page will perform in search.
- What is Google’s AI-generated content policy?
- Google’s guidance on AI-generated content points to its spam policies and to its people-first content guidance. The spam policies define scaled content abuse: “Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users.” The people-first page asks creators: “Does the content provide original information, reporting, research, or analysis?” and “Does the content have any easily-verified factual errors?” Sources, with the dates they were read, are listed at the foot of this page.
- Should AI-generated content be labelled as AI-generated?
- Google’s guidance says “AI or automation disclosures are useful for content where someone might think "How was this created?"” It adds: “Consider adding these when it would be reasonably expected.” The people-first page puts its question under this condition: “If automation is used to substantially generate content…” The question is: “Is the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways?” uprank does not add a disclosure or a byline to an article. Add them in your CMS or template if your readers would expect them.
- How does uprank check AI content against what Google says?
- uprank does not test against Google. It runs 13 hard gates on every version: Information gain at least 70, Evidence at least 85, Practical value at least 75, Risk at most 20, and integrity checks including no invented citations and no unsupported high-risk claims. A version that has not cleared all of them cannot be approved, pushed to Strapi or recorded as a live publication request. The table on this page shows which Google question each gate is closest to, and what it cannot do.
Google sources
Every claim about Google on this page comes from these pages. Read 2026-10-16.
- Google Search's guidance about AI-generated contentGoogle Search Central BlogPosted 2023-02-08. Read 2026-10-16.https://developers.google.com/search/blog/2023/02/google-search-and-ai-content
- Spam policies for Google web searchGoogle Search Central documentationLast updated 2026-08-28 UTC. Read 2026-10-16.https://developers.google.com/search/docs/essentials/spam-policies
- Creating helpful, reliable, people-first contentGoogle Search Central documentationLast updated 2026-10-05 UTC. Read 2026-10-16.https://developers.google.com/search/docs/fundamentals/creating-helpful-content
Keep reading
- Quality gates: every dimension, weight and hard gate.
- How to fact-check AI content: check each claim against its source before you publish.
- How to add citations to AI-generated content
- AI writer that doesn’t hallucinate sources
- AI content approval workflow: review before you publish
- How to keep AI content on brand
- How to update old blog posts with AI without breaking facts
- FAQ: product, agents, tokens and publishing.
- Agent setup guides: connect a coding agent over MCP.