Some unofficial bits of information about Google (and sometimes Bing) and their search engines, gathered here and there over the past few days, with this week's lineup answering a few of these questions: Is the LLMS.txt file more useful than the meta keywords tag once was? What is Bing's view on AI Search? Why should you make your site accessible to AI agents? Does Google use its MUVERA and GFM algorithms?
Google compares the LLMS.txt file to the meta keywords tag
Gary Illyes recently stated on Bluesky that the LLMS.txt file, designed to “control” LLMs’ access to web content, strongly resembles the meta keywords tag of the 1990s, which quickly became obsolete and ineffective. He thus underscored the limited interest Google would have in taking this file into account, suggesting it could be ignored in the same way. This remark confirms Google's skepticism toward the initiative, despite the lack of any clear downside to using it.
Source: Search Engine Roundtable
Reliability rating: 
We agree!
A new statement that once again reinforces the uselessness of the LLMS.txt file.
Goossip #2
Bing confirms that classic SEO remains the key for AI Search
Fabrice Canel, product lead at Bing, says that to optimize visibility in AI Search you only need to follow the usual SEO best practices, a view shared by Google. In a recent blog post, his team reminds the importance of XML sitemaps, of the lastmod (deemed crucial by Bing) and the IndexNow technology to ensure recent content is indexed. According to Canel, content visible in AI Search must first be properly crawled and indexed, as in traditional SEO.
Source: Search Engine Roundtable
Reliability rating: We agree!
As long as there are sites from which LLMs will draw their content, SEO fundamentals will remain essential.
Goossip #3
Google recommends testing e-commerce sites' accessibility for AI agents
John Mueller suggests that e-commerce site owners check whether their store is accessible to AI agents, like those used via ChatGPT for shopping. An experiment conducted in Switzerland showed that some sites remained inaccessible because of CAPTCHAs, Cloudflare protections, or maintenance pages. Mueller reminds that these obstacles can harm the user experience and suggests that SEOs include this test in their audits, as the use of AI agents becomes increasingly common.
Source: Search Engine Journal
Reliability rating: We agree!
Designed to perform various tasks on behalf of humans, AI agents can indeed encounter technical blocks. Given their rapidly growing adoption, it is strongly recommended to ensure your site is accessible.
Goossip #4
Google uses a method similar to MUVERA but remains vague about the Graph Foundation Model
During an exchange at Search Central Live, Gary Illyes confirmed that Google uses a method similar to MUVERA, an innovative multi-vector search technique compressed into fixed vectors, which enables fast and precise retrieval at large scale. MUVERA combines efficiency (thanks to single-vector search) and relevance (with multi-vector reranking). However, Illyes remains evasive about the use of Graph Foundation Models (GFM) in Google Search, estimating they would not yet be in production.
Source: Search Engine Journal
Reliability rating: We agree!
An interesting clarification on two important new algorithms recently unveiled by Google.
The article “Goossips SEO: LLMS.txt, AI Search, AI Agents, MUVERA & GFM” was published on the site Abondance.
