Google Raters Now Evaluate if Content Is AI-Generated

According to John Mueller, who leads the Senior Search Analyst and Search Relations teams at Google, during a talk at Search Central Live in Madrid, Google has instructed its quality raters to identify pages containing automatically generated or AI-created primary content and evaluate these pages as having the lowest possible quality.

This was posted by Aleyda Solis on LinkedIn earlier today:

This alteration was part of the January 2025 update If you didn’t catch all the details from that update regarding the Search Quality Rater Guidelines, here’s a summary of the key modifications in the newest edition.

1. Google unveils a fresh term: Generative Artificial Intelligence

In its most recent update of the Search Quality Rater Guidelines, Google included a definition and framework for generative artificial intelligence for the first time. According to Google’s documentation, this tool is considered helpful yet has the potential to be misused.

The inclusion of Generative AI is found in Section 2.1 (Key Definitions):

Generative Artificial Intelligence is a form of machine learning model capable of producing novel content—such as text, images, music, and code—based on the patterns it learns from the data it’s trained on. Various platforms utilize these models to generate content through generative AI. While this technology can significantly aid in generating material, similar to all tools, it carries potential risks when misused.

2. Google revises and broadens its spam classifications

Google considerably revised the criteria for identifying spammy web pages.

Section 4.6.3 (Auto-generated MC), which previously existed, has been removed by Google. Instead, they have introduced new subsections and intensified their emphasis on scaled, low-effort content, as well as the risks of potential AI abuse.

What's new for 2025 aligns the guidelines accordingly. Google’s Major Search Quality Updates from last year:

  • Expired Domain Abuse (Section 4.6.3): This occurs when "a lapsed domain name is bought and reused mainly to advantage the new site owner by featuring material that offers minimal or zero worth to visitors."
  • Abuse of Site Reputation (Part 4.6.4) : When “third-party content is published on a host site mainly because of that host’s already-established ranking signals, which it has earned primarily from its first-party content. The goal of this tactic is for the content to rank better than it could otherwise on its own.”
  • Scaled Content Abuse (Section 4.6.5): Producing substantial amounts of content "with minimal effort or uniqueness without any editing or manual curation." The use of generative artificial intelligence is cited as an instance of such an automatic tool.
  • MC Developed with Minimal Effort, Lack of Originality, and Limited Additional Value for Users (Section 4.6.6) Here lies a dedicated area for poorly paraphrased material frequently encountered in outputs from generative AI systems and similar automation technologies.

The section Mueller highlighted during his presentation was Section 4.6.6., particularly this portion:

“The Lowest The rating comes into play if everything or nearly everything from the main content (covering text, images, audio, videos, etc.) has been duplicated, restated, included, or embedded. auto or AI generated , or simply copied from elsewhere without much effort, lacking originality and not adding significant value for our website’s visitors. These pages deserve lower ratings. Lowest , even if the page attributes the content to another source." [emphasis added].

So, how might an evaluator determine if content has been auto or AI-generated? Since there isn’t specific guidance for AI-created material, there is some recent direction concerning "paraphrased content" instead.

  • Section 4.6.6: Automated tools can likewise generate paraphrased material by restating or summarizing the information from various webpages.
  • Section 4.6.7: Paraphrased material may be significantly more difficult to identify… Such content tends to:
    • Sure, please provide me with the text you would like to be paraphrased.
    • Show significant similarity to pages from reputable sources like Wikipedia and reference sites.
    • It seems like you want to provide a concise summary of a particular page, for example, from a forum discussion or a news article, but with nothing extra added.
    • Include terms or signs indicating summaries or paraphrases generated by AI tools, for example, phrases like 'As an AI language model.'

3. Google clarifies the difference between low and lowest ratings

This new part provides guidelines for raters on how to handle content that doesn't reach a level severe enough to warrant an rating. Lowest ranking, yet it still merits a Low rating. Here’s the difference:

  • Low Some material is recycled, yet there is still some basic effort made to adapt or alter it.
  • Lowest : Almost all MC is copied or paraphrased with no effort or added value.

The Search Quality Rater Guidelines share examples of repackaged content like:

  • Reposting content from social media with minimal extra commentary or conversation
  • Webpages containing material sourced externally (such as embedded video pages or those featuring 'repinned' images) where the individual responsible for the webpage adds minimal commentary, dialogue, or oversight.
  • "Top" compilations derived from pre-existing evaluations and listings featuring minimal fresh material.

Google aims for raters to identify thin content that attempts to appear original yet fails to meet the criteria for a satisfactory user experience.

4. Google incorporates 'placeholder' material

This updated segment deals with "padding" material—essentially, content that requires minimal effort and has little significance. Such elements might take up significant visual space on a page without contributing effectively to its objective.

Filler can artificially pad out content, producing a webpage that seems abundant yet fails to provide material which website visitors consider valuable.

It highlights that even if material doesn't pose harm, it still has the potential to generate earnings. Low Rating systems that make it more difficult to find genuinely valuable content. Specifically aimed at: pages where crucial information is hidden under advertisements, vague beginnings, or overly long sections.

  • Content that is noticeably inserted and draws attention away from the main character.
  • Pages that seem more extensive or detailed due to extra spacing.

Reviewers are prompted to assess how the arrangement of pages and the structure of content influence a user's capability to accomplish their objective.

5. Google becomes more strict about overly dramatic or slightly deceptive statements.

Google’s Search Quality Rater Guidelines now specifically address overly dramatic or slightly deceptive statements regarding the person behind a web page, even when these assertions do not qualify as complete falsehoods.

The recently added Section 5.6 provides an explanation of:

Misleading details regarding a website or content producer serve as significant grounds for assigning the lowest rating.

However, it cautions that subtler forms of embellishment (such as exaggerated qualifications, fabricated know-how) can be grounds for concern. Low rating:

Occasionally, details regarding the website or content creator may appear inflated or somewhat deceptive, like assertions of individual experiences or skills that come off as overly dramatic or added merely to awe site visitors.

This indicates that raters should depend on what the primary content shows, along with additional research, instead of accepting assertions without question.

E-E-A-T evaluations should rely on the MC itself, insights gathered from reputation analysis, verifiable qualifications, among others, rather than solely depending on assertions like 'I am an authority!' made either through websites or content creators.

If an evaluator determines that the creator’s stated credentials seem more like marketing spin than substance, the document is clear:

“If you find the information about the website or the content creator to be exaggerated or mildly misleading, the Low rating should be used.”

Some other smaller changes

Google also implemented several other small adjustments.

  • Low-quality sections (Part 4.0): Google has included the following statement: "A low rating is necessary if the webpage is designed primarily to profit the site’s owner (such as generating income) without making much effort to enhance user experience or provide significant value."
  • Misleading Purpose of the Page, Misinformation About the Site, Deception in Design (Part 4.5.3): Google updated this part of the document, adding further details which they organized into a table and a bullet-pointed list including examples.
  • New evaluation category: Minimal Recipe 3: This will be provided for recipe pages featuring a significant amount of irrelevant content, interstitials, and advertisements.
  • AdBlocking Add-ons (Part 0.4) Raters are required to "disable all ad-blocking features within your web browser when performing rating tasks." This includes browsers such as Chrome, which may automatically prevent certain advertisements from displaying.
Posting Komentar (0)
Lebih baru Lebih lama