AI Score: How Google Could Detect AI-Generated Texts

The goal is to algorithmically detect mass AI content that has not undergone any editorial revision and to limit its visibility.

Background

In 2024, Google came under pressure because numerous niche sites generated content automatically and in large quantities using artificial intelligence (AI). This practice led to an oversupply of low-quality content in the index.

According to a study of around 79,000 (predominantly information-oriented) websites, about 1.9% received a manual action; a high AI share correlated conspicuously with these actions (correlation, not proven causation).

What Is the AI Score?

The AI Score, according to the leaked documents, is a hypothetical evaluation mechanism that Google could use to identify web pages with a high AI share. It is presumably a site-wide evaluation value that is calculated on the basis of several sub-metrics.

The potential components of this score include:

Normalized Site Rank (NSR)What is it?The Normalized Site Rank is a ranking factor that, according to the leaks, measures a website’s general performance compared to other sites in the same industry. It is strongly based on user behavior and is apparently determined via user signals that could also originate from the Chrome browser.
What Feeds into the NSR? Why Is This Relevant?
The NSR helps Google assess whether a website is useful, trustworthy, and high-quality – or whether it merely delivers mass-generated content without added value. A conspicuously large amount of new content without recognizable user engagement can worsen the score. What is it?
Site Authority measures how “trustworthy” and “established” a domain is – a kind of reputation value. It is partly derived from the NSR data but also incorporates other signals. What Influences Site Authority? Why Is This Important?
High authority means that Google trusts the website – which in turn leads to better rankings. AI-generated mass content on a previously high-quality site can jeopardize this trust bonus. What is it?
The Effort Score is intended to evaluate the actual effort that went into creating a piece of content. According to the leaks, an AI is even used here to estimate this effort algorithmically. What Feeds into the Score? Important:
“Publish and pray” does not work – content should be deliberately revised to increase the score. Unrevised AI content is presumably penalized significantly. What is it?
This is a group of individual ratings that relate specifically to the content. It concerns text, image, and video – each with its own sub-metrics. Examples: Important:
This score is extremely user-centric. The better the content meets visitors’ expectations, the higher the rating. What is it?
The Clutter Score measures how “cluttered” or “disruptive” a page is built – particularly with regard to advertising, pop-ups, and other UI obstacles. What Has a Negative Effect? Why Is This Important?
A tidy, mobile-friendly page increases user satisfaction – and that is directly rewarded. What is it?
These scores attempt to detect manipulative or low-quality content algorithmically. There are several sub-metrics: Note:
Especially AI-generated texts that are published without editorial revision run the risk of performing poorly in these scores. They capture characteristics such as “gibberish” (meaningless content) or keyword stuffing. The goal is to algorithmically expose manipulative content.

Click-Through Rate (CTR): How often do users click on your page’s search result compared to other hits?
Dwell Time: How long do users stay on the page? Early bouncing (a high bounce rate) is negative.
Page Interactions: Is there scrolling, clicking, interaction?
User Experience (UX): Good structure, navigation, and mobile optimization help.
Sitewide Authority Score: A kind of overall value for the trust in the website (similar to Domain Authority).
Comparison with Competitors: The score is normalized, which means: your page is compared against other pages in your niche.

Site Authority

Quality and originality of the content
Consistency across all pages (sitewide quality)
Backlinks from trustworthy sources
Brand awareness / brand signals
Avoidance of spam or manipulation

Effort Score

Content Design: Structure, paragraphs, subheadings, reading flow.
Research Effort: Content with verifiable sources and depth is rated better.
Scannability: Bullet points, info boxes, tables help.
Multimedia: Images, videos, and audio break up walls of text.
Post-Publishing Activities: Was the content optimized after publication or simply “blasted out”?
Readability: Clarity of language, avoidance of empty buzzwords or hollow phrases.

Content Scores

Title Match Score: Does the text match what the title promises? Clickbait is rated negatively.
Quality Click Signals: How often do users click back? Is the page perceived as helpful?
Video & Image Quality: High-resolution, relevant media content has a positive effect.
Semantic Match: Does the content match the search queries?

Clutter Score

Pop-ups that are difficult to click away
Excessive advertising, especially above the fold
Autoplay videos
Unstructured page layouts

Spam Scores

Gibberish Score: Rates meaningless, empty content (e.g., automated texts with no substance).
Spam Word Score: Watches for keyword stuffing, i.e., excessive repetition of main terms.
Duplicate Content Checks: Recycled content or copied blocks of text have a negative effect.

Sources and Assessment

The existence and concrete implementation of an AI Score has not been officially confirmed by Google. The information comes from leaked documents as well as analyses from third-party sources, including SEO Südwest, On-Page.ai, and the leaked internal API reference GoogleApiContentWarehouse (not official Google documentation, but rather an internal reference that was accidentally leaked via a GitHub repository, uncovered in May 2024 by Rand Fishkin/Mike King).

For proper context, it is advisable to clearly separate the signals documented in the leak (siteAuthority, NSR, ClutterScore, SpamScores) from their interpretation – namely the summary as an “AI Score” as well as the assumption that an AI estimates the “effort.”

Significance for SEO and Content Creation

The leaks suggest that Google differentiates between AI-generated content – content that is high-quality and has evidently been editorially managed is not penalized across the board. The AI Score could thus be a central mechanism with which Google distinguishes between “mass-produced content” and “quality content.”

Quellen