Google\u2019s Gary Illyes confirmed that AI content is fine as long as the quality is high. He said that \u201chuman created\u201d isn\u2019t precisely the right way to describe their AI content policy, and that a more accurate description would be \u201chuman curated.\u201d<\/p>\n
The questions were asked by Kenichi Suzuki in the context of an exclusive interview with Illyes.<\/p>\n
AI Overviews and AI Mode Models<\/p>\n
Kenichi asked about the AI models used for AI Overviews and AI Mode, and he answered that they are custom Gemini models.<\/p>\n
Illyes answered:<\/p>\n
\u201cSo as you noted, the the model that we use for AIO (for AI Overviews) and for AI mode is a custom Gemini model and that might mean that it was trained differently. I don\u2019t know the exact details, how it was trained, but it\u2019s definitely a custom model.\u201d<\/p>\n
Kenichi then asked if AI Overviews (AIO) and AI Mode use separate indexes for grounding.<\/p>\n
Grounding is where an LLM will connect answers to a database or a search index so that answers are more reliable, truthful, and based on verifiable facts, helping to cut down on hallucinations. In the context of AIO and AI Mode, grounding generally happens with web-based data from Google\u2019s index.<\/p>\n
Suzuki asked:<\/p>\n
\u201cSo, does that mean that AI Overviews and AI Mode use separate indexes for grounding?\u201d<\/p>\n
Google\u2019s Illyes answered:<\/p>\n
\u201cAs far as I know, Gemini, AI Overview and AI Mode all use Google search for grounding. So basically they issue multiple queries to Google Search and then Google Search returns results for that those particular queries.\u201d<\/p>\n
Kenichi was trying to get an answer regarding the Google Extended crawler, and Illyes\u2019s response was to explain when the Google Extended crawler comes into play.<\/p>\n
\u201cSo does that mean that the training data are used by AIO and AI Mode collected by regular Google and not Google Extended?\u201d<\/p>\n
And Illyes answered:<\/p>\n
\u201cYou have to remember that when grounding happens, there\u2019s no AI involved. So basically it\u2019s the generation that is affected by the Google extended. But also if you disallow Google Extended then Gemini is not going to ground for your site.\u201d<\/p>\n
AI Content In LLMs And Search Index<\/p>\n
The next question that Illyes answered was about whether AI content published online is polluting LLMs. Illyes said that this is not a problem with the search index, but it may be an issue for LLMs.<\/p>\n
Kenichi\u2019s question:<\/p>\n
\u201cAs more content is created by AI, and LLMs learn from that content. What are your thoughts on this trend and what are its potential drawbacks?\u201d<\/p>\n
Illyes answered:<\/p>\n
\u201cI\u2019m not worried about the search index, but model training definitely needs to figure out how to exclude content that was generated by AI. Otherwise you end up in a training loop which is really not great for for training. I\u2019m not sure how much of a problem this is right now, or maybe because how we select the documents that we train on.\u201d<\/p>\n
Content Quality And AI-Generated Content<\/p>\n
Suzuki then followed up with a question about content quality and AI.<\/p>\n
He asked:<\/p>\n
\u201cSo you don\u2019t care how the content is created\u2026 so as long as the quality is high?\u201d<\/p>\n
Illyes confirmed that a leading consideration for LLM training data is content quality, regardless of how it was generated. He specifically cited the factual accuracy of the content as an important factor. Another factor he mentioned is that content similarity is problematic, saying that \u201cextremely\u201d similar content shouldn\u2019t be in the search index.<\/p>\n
He also said that Google essentially doesn\u2019t care how the content is created, but with some caveats:<\/p>\n
\u201cSure, but if you can maintain the quality of the content and the accuracy of the content and ensure that it\u2019s of high quality, then technically it doesn\u2019t really matter.<\/p>\n
The problem starts to arise when the content is either extremely similar to something that was already created, which hopefully we are not going to have in our index to train on anyway.<\/p>\n
And then the second problem is when you are training on inaccurate data and that is probably the riskier one because then you start introducing biases and they start introducing counterfactual data in your models.<\/p>\n
As long as the content quality is high, which typically nowadays requires that the human reviews the generated content, it is fine for model training.\u201d<\/p>\n
Human Reviewed AI-Generated Content<\/p>\n
Illyes continued his answer, this time focusing on AI-generated content that is reviewed by a human. He emphasizes human review not as something that publishers need to signal in their content, but as something that publishers should do before publishing the content.<\/p>\n
Again, \u201chuman reviewed\u201d does not mean adding wording on a web page that the content is human reviewed; that is not a trustworthy signal, and it is not what he suggested.<\/p>\n
Here\u2019s what Illyes said:<\/p>\n
\u201cI don\u2019t think that we are going to change our guidance any time soon about whether you need to review it or not.<\/p>\n
So basically when we say that it\u2019s human, I think the word human created is wrong. Basically, it should be human curated. So basically someone had some editorial oversight over their content and validated that it\u2019s actually correct and accurate.\u201d<\/p>\n
Takeaways<\/p>\n
Google\u2019s policy, as loosely summarized by Gary Illyes, is that AI-generated content is fine for search and model training if it is factually accurate, original, and reviewed by humans. This means that publishers should apply editorial oversight to validate the factual accuracy of content and to ensure that it is not \u201cextremely\u201d similar to existing content.<\/p>\n
Watch the interview:<\/p>\n
\n
Featured Image by Shutterstock\/SuPatMaN<\/p>\n","protected":false},"excerpt":{"rendered":"Google\u2019s Gary Illyes confirmed that AI content is fine as long as the quality is high. He said…\n","protected":false},"author":2,"featured_media":67365,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[554,733,4308,86,56,54,55],"class_list":["post-67364","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-technology","tag-uk","tag-united-kingdom","tag-unitedkingdom"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/67364","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=67364"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/67364\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/67365"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=67364"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=67364"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=67364"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}