Bing Expands AI Abuse Rules, Adds GEO to Guidelines

Microsoft has significantly updated its official webmaster guidelines to address how content performs in both standard search results and within its AI-powered Copilot responses. This revision marks a formal shift, recognizing that a website’s visibility now extends beyond traditional rankings to include its potential as a source material, or “grounding,” for generative AI answers. The updated document provides new clarity for site owners navigating this dual landscape, offering specific directives for optimization and outlining what constitutes abuse in an AI context.

A major addition is detailed guidance on how existing meta directives now control content appearance in AI experiences. The `NOARCHIVE` tag, for instance, will prevent a page from being used in Copilot responses altogether. Using `NOCACHE` limits Copilot to displaying only a basic URL, title, and snippet, potentially reducing the value of a citation. The guidelines advise against applying `NOCACHE` to content intended for rich AI citations. This builds upon the recent introduction of the `data-snippet` attribute, giving publishers finer control over which text sections Bing can display or reference.

For the first time, the guidelines formally introduce the concept of Generative Engine Optimization (GEO). Bing defines GEO as efforts focused on making content eligible for use as a grounding reference in AI-generated responses. The company is careful to note that, similar to how SEO doesn’t guarantee top rankings, GEO does not guarantee citations. This terminology aligns with Microsoft’s recent launch of an AI Performance dashboard for webmasters, placing GEO into official policy alongside the new analytics tooling.

The language around AI-generated content has also been softened considerably. Previously, all “machine-generated content” was deemed malicious and subject to penalty. The new policy refines this stance, stating that large-scale content generated without oversight, quality control, or editorial review often lacks usefulness and may be excluded. This change brings Bing’s policy closer to Google’s updated spam approach, which targets content made primarily to manipulate rankings rather than all automated content.

The update introduces entirely new sections dedicated to optimizing for AI grounding. Recommendations include stating facts directly for independent verification, using clear and consistent entity names, and focusing each URL on a single topic. Bing recommends placing essential information near the top of a page, as single-topic pages with clear structures are more likely to be selected as grounding sources.

Definitions of abuse have been expanded to cover AI-specific manipulation. The section on keyword stuffing is now titled “Keyword Stuffing and Artificially Engineered Language,” explicitly covering content engineered to trigger AI responses or citations. Prompt injection has been elevated to a full section called “Prompt Injection and AI Manipulation,” detailing attempts to interfere with or manipulate the language models powering Bing and Copilot.

Several older, more technical sections were removed in the rewrite, including detailed sitemap format support and specific JavaScript rendering guidance. Microsoft did not issue a separate announcement; the changes are simply live on the Bing Webmaster Guidelines page.

A critical insight from the update is the note that a decline in traditional click-through rates may not indicate lost visibility. Content might now surface as citations within Copilot answers without generating a direct click. Therefore, tracking impressions and citation eligibility is becoming as important as monitoring click data. While Google has stated its AI Overviews respect certain snippet controls, it has not yet provided the same granular, directive-by-directive breakdown for AI experiences that Bing now offers.

(Source: Search Engine Journal)