Microsoft AI Training Controls: Stay Visible in Bing Search

Yes. Microsoft says websites will be able to block generative AI training while remaining eligible for traditional Bing search results. Publishers can separately control whether their content helps generate answers in Copilot and other Microsoft AI experiences. Announced on 30 October 2026, the preferences begin taking effect on 1 February 2027 as Microsoft recrawls content. For Sydney and NSW publishers, the change separates control over AI reuse from ordinary search discoverability.

ELYMENT Insights | Artificial Intelligence & Digital Business | Sydney, NSW | 2 November 2026

Microsoft Is Separating Two Uses of the Same Website

A website can serve several purposes for an artificial intelligence company.

Its published information might help train a future generative AI model. The same information might also be retrieved to support an answer being generated for a user. Separately, the website might appear as an ordinary result in a search engine.

Those activities are related, but they are not identical.

Microsoft's latest announcement establishes a clearer distinction between them.

On 30 October 2026, Krishna Madhavan, Principal Product Manager at Microsoft Bing, announced new publisher preferences that distinguish between AI model training and AI grounding.

The announcement, published under the title Your content, your choice: clearer preferences for training and grounding, introduces more granular controls over how Microsoft's generative AI systems may use website content.

The central change is commercially significant.

A publisher will be able to express a preference against having its content used for generative AI training without automatically preventing that content from appearing in ordinary Bing search results.

It may also allow its content to inform Microsoft Copilot responses while preventing the same content from supporting certain AI grounding services supplied to third parties.

For Sydney digital publishers, specialist information businesses, professional associations and organisations investing in original research, this creates choices that previously required less flexible controls.

However, the announcement must not be confused with an immediate product release.

As at 2 November 2026, the new preferences have been announced but are not yet effective. Microsoft will begin honouring them on 1 February 2027 as its crawlers discover the relevant settings.

Training, Grounding and Search Are Different Activities

Understanding Microsoft's announcement requires distinguishing three separate processes.

1. AI Training

AI training involves using content to modify the learned parameters of a generative AI model.

Publishers may object to this use because their articles, research, photographs or other original material represent substantial editorial and commercial investment.

Microsoft's new AI Training preference allows a publisher to specify whether relevant website content can be used for future generative model training.

The available preferences are:

• Allow: Permit the relevant training use.
• Disallow: Express a preference that the content must not be used for generative AI training.

Allow is the default when no applicable preference has been published.

2. AI Grounding

Grounding is different from training.

When an AI system generates an answer, it may retrieve information from websites to help make that response more relevant, current or factually supported.

Microsoft calls this use AI Grounding.

For example, someone using Copilot might request a comparison of specialist services in Sydney.

The system may retrieve publicly available information from relevant websites and use that information to help construct its response.

Depending on the experience, the generated answer may also include source links or citations.

Microsoft's new AI Grounding preference allows publishers to decide whether their content can support these experiences.

3. Traditional Bing Search

Traditional search involves discovering, indexing and ranking webpages so users can find relevant information through search results.

Microsoft explicitly distinguishes ordinary indexing and ranking from AI grounding.

Its new AI Training and AI Grounding preferences do not, by themselves, remove a website's eligibility for traditional Bing search results.

However, normal technical requirements still apply.

A website that blocks Bingbot from crawling relevant content cannot expect an AI usage preference to override that restriction.

Remaining eligible for search also does not guarantee indexing, ranking positions or traffic.

The Three AI Grounding Choices Publishers Will Have

The most distinctive part of Microsoft's proposal is the separate AI Grounding preference.

Unlike the AI Training preference, which can be expressed through robots.txt, AI Grounding is controlled at the individual webpage level.

Allow

• What it allows: Content can help inform eligible AI-generated responses.
• Visibility consideration: Retains eligibility for Microsoft-operated and relevant third-party grounding experiences.

DisallowExtended

• What it allows: Content can inform Microsoft-operated AI experiences but not grounding services Microsoft provides to third parties.
• Visibility consideration: Preserves some AI visibility while excluding a category of third-party grounding use.

Disallow

• What it allows: Content cannot be used to help inform generated responses covered by this Microsoft preference.
• Visibility consideration: Reduces opportunities for visibility in AI-generated experiences.

Microsoft identifies Copilot, Microsoft 365, AI Mode and AI Overview among its operated experiences.

Third-party grounding services include Web IQ and grounding through API access in Microsoft Foundry.

The distinction makes DisallowExtended particularly relevant to publishers that want their content discoverable through Microsoft's own AI experiences without making it available through Microsoft's broader third-party grounding services.

However, Microsoft warns that choosing either DisallowExtended or Disallow reduces brand visibility in AI experiences.

No grounding setting guarantees citations, referral traffic, licensing revenue or payment.

Can a Publisher Block Training but Still Appear in Copilot?

Under Microsoft's announced framework, yes.

The independent preferences allow a publisher to select:

• AI Training: Disallow.
• AI Grounding: Allow.

This combination expresses a preference that content must not be used for future generative AI training while remaining eligible to inform AI-generated responses.

Ordinary Bing search eligibility remains separate.

Consider a fictional Sydney-based business publication that produces original reporting on Australian technology companies.

The publisher may have three objectives:

1. Protect its editorial archive from use in future generative AI model training.
2. Remain discoverable through Bing search.
3. Continue receiving potential exposure when Copilot references relevant published information.

Previously, publishers seeking to restrict AI use through Microsoft's older controls faced a less granular choice.

With the new framework, these objectives can be addressed independently.

The important qualification is that expressing a preference does not prove that a particular article will be retrieved, cited or visited.

Microsoft controls eligibility through the announced preferences, but actual visibility depends on additional retrieval and product behaviour.

What Changes in robots.txt and Page-Level Settings?

Microsoft is participating in work towards a shared AI training preference being developed through the Internet Engineering Task Force's AI Preferences initiative.

The intention is to reduce the number of different platform-specific controls publishers must manage.

The shared preference remains part of ongoing standards work rather than a completed universal standard.

Microsoft's announcement supports expressing an AI Training preference through robots.txt for a website or a particular path.

For example, Microsoft's published syntax for expressing a site-wide training restriction is:

User-Agent: __usage_ai-training
Disallow: /

This is an AI training preference, not an instruction to prevent ordinary Bingbot crawling.

Microsoft also supports page-level controls.

A publisher wishing to express a training restriction on an individual page can use:

<meta name="Bingbot" content="ai-training:disallow">

For grounding, Microsoft's published page-level example is:

<meta name="Bingbot" content="ai-grounding:disallow">

Microsoft also supports corresponding HTTP response headers.

A page-level AI Training preference overrides the relevant robots.txt AI Training preference for that page.

There is no equivalent site-wide robots.txt setting for Microsoft's AI Grounding preference.

These examples illustrate Microsoft's announced configuration, but publishers should check the current Bing Webmaster Tools AI Preferences documentation before implementation.

The timing is essential: Publishing these settings during November 2026 does not mean Microsoft has already begun enforcing the new framework.

Enforcement begins on 1 February 2027 as Microsoft crawls relevant pages and settings.

Four Dates Matter More Than the Announcement Date

Microsoft has published a transition schedule for website owners.

30 October 2026

• Microsoft's announced change: Microsoft announces independent AI Training and AI Grounding preferences.
• Publisher action: Review the announcement and assess existing policies.

15 December 2026

• Microsoft's announced change: Temporary grounding transition preferences become available in Bing Webmaster Tools for eligible legacy-control users.
• Publisher action: Check whether the website qualifies for the transition mechanism.

1 February 2027

• Microsoft's announced change: Microsoft begins honouring the new preferences as it crawls.
• Publisher action: Verify that the intended settings are published and accessible.

1 March 2027

• Microsoft's announced change: Legacy NoArchive and NoCache controls cease being recognised as AI training or grounding preferences.
• Publisher action: Complete the required migration to preserve intended restrictions.

30 September 2027

• Microsoft's announced change: Any remaining temporary grounding preferences expire.
• Publisher action: Ensure permanent page-level controls have been implemented.

The transition is particularly important for websites currently relying on NoArchive or NoCache.

Microsoft says publishers using those controls should not assume their previous AI restrictions will continue automatically.

For eligible publishers that want to maintain grounding restrictions, Microsoft requires the appropriate page-level preference and a temporary grounding preference recorded in Bing Webmaster Tools before 1 March 2027.

The temporary Webmaster Tools mechanism is available only to sites using the relevant legacy controls as of 15 December 2026.

It is not a general configuration tool for every publisher.

Microsoft also states that when it discovers a new AI Training or AI Grounding preference on a site, it will disregard remaining legacy preferences for those AI uses.

The operational implication is straightforward: publishers should plan a coordinated migration rather than adding isolated tags without checking their existing configuration.

The Visibility Trade-Off Is Commercial, Not Merely Technical

Microsoft's new controls create a distinction between protecting content and participating in AI discovery.

For publishers, the appropriate choice will depend partly on how their content creates business value.

Subscription Publishers

A subscription publication may be particularly concerned about whether AI-generated answers reduce the incentive to access its original reporting.

Restricting grounding could align with a decision to reserve more value for subscribers.

However, doing so may also reduce the publication's visibility within AI-assisted research experiences.

The decision should be evaluated alongside subscription acquisition, referral traffic and licensing strategy.

Professional Associations and Research Organisations

A NSW professional association may publish public guidance because broad distribution is central to its purpose.

Allowing grounding could help its authoritative material become part of relevant AI-generated answers.

It may nevertheless decide that future model training is a separate use requiring different treatment.

Business and Specialist Information Websites

A Sydney professional services organisation may publish technical explanations, market commentary and original research to attract qualified enquiries.

Allowing grounding may support brand discovery in AI-assisted research.

But appearing in an AI-generated answer does not necessarily produce a website visit or commercial enquiry.

The relevant performance measures therefore extend beyond search rankings.

Elyment's JL Turf Group digital platform case study illustrates how technical search visibility, structured website content and AI citation reporting can be considered together.

The broader question of how external AI interfaces change customer acquisition is examined in Elyment's analysis of BMW's integration of its vehicle configurator into ChatGPT.

Those developments concern AI-assisted discovery and customer journeys. Microsoft's new announcement addresses the earlier question of whether publishers authorise particular uses of their content in the first place.

How Publishers Can Measure the Impact

One difficulty with AI publisher controls is establishing what has actually changed after implementation.

Microsoft already provides AI visibility reporting through Bing Webmaster Tools.

Its AI Performance reporting includes information about citations, cited pages and grounding queries across supported AI experiences.

In October 2026, Microsoft also announced additional topic-level reporting through its Topics Portfolio beta.

These capabilities give publishers a starting point for assessing AI visibility.

A useful evaluation should distinguish:

• Search visibility: Whether pages continue to appear in ordinary Bing results.
• AI citations: Whether content appears as a referenced source in supported AI experiences.
• Referral traffic: Whether users subsequently visit the publisher's website.
• Commercial outcomes: Whether discovery produces subscriptions, enquiries or other measurable value.
• Policy compliance: Whether the intended technical preferences are published and recognised after crawling.

Publishers should avoid treating an increase in citations as equivalent to an increase in revenue.

They should also avoid assuming that a decline in citations proves a particular control has taken effect.

Changes in user demand, competing content, retrieval behaviour and reporting coverage can all influence the results.

AI visibility reporting is useful evidence, but it is not a definitive audit of model training or content reuse.

A Practical Implementation Plan for Sydney Publishers

Organisations managing large websites, multiple brands or extensive content archives should treat the change as a content-governance exercise rather than a single SEO adjustment.

Five steps are particularly important.

Step 1: Audit Existing Content Controls

Review robots.txt, relevant HTML meta tags and HTTP response headers.

Identify any existing NoArchive or NoCache configuration.

Check whether restrictions are applied consistently across content types, subdomains and publishing platforms.

Step 2: Classify Content by Intended Use

Not every webpage necessarily requires the same policy.

A business may distinguish between:

• Public information intended for maximum discoverability.
• Original editorial analysis.
• Subscription-supported material.
• Licensed or syndicated content.
• Research that carries specific commercial value.

Different categories may justify different training or grounding preferences.

Content ownership and contractual restrictions should be considered before a publisher authorises uses of material it does not exclusively control.

Step 3: Implement Preferences at the Correct Level

Determine whether the training preference belongs in robots.txt, individual webpages or both.

Apply grounding preferences at page level.

Coordinate the implementation across website developers, content teams and any relevant publishing-platform providers.

For businesses managing numerous websites, centralised configuration records can help reduce inconsistent settings.

Step 4: Verify Deployment and Crawling

Confirm that settings are visible in the live HTML or HTTP response headers.

Check for conflicts with existing crawl restrictions.

Remember that Microsoft must discover the relevant settings through crawling before they become effective under the new framework.

IndexNow may help notify participating search engines that content has changed, but Microsoft does not guarantee a particular recrawl date.

Step 5: Review Visibility After February 2027

Establish a reporting baseline before the new preferences take effect.

After implementation, monitor traditional search performance and AI citations separately.

Document any changes to preferences so that later performance comparisons can be interpreted against the relevant configuration history.

This governance approach is consistent with the wider challenge of managing business information across search, AI applications and customer-facing systems.

Elyment has examined related discovery challenges in its analysis of Google Gemini becoming a new discovery channel for service businesses.

What Microsoft's New Preferences Do Not Guarantee

The announcement introduces meaningful publisher choice, but its limitations are important.

• It does not take immediate effect. The announced implementation begins on 1 February 2027.
• It does not guarantee Bing rankings. Eligibility for ordinary search results remains subject to normal search requirements.
• It does not guarantee Copilot citations. Allowing grounding means content can be considered for relevant AI experiences, not that it will necessarily appear.
• It does not create a payment arrangement. Microsoft's preferences do not establish licensing fees or revenue-sharing rights.
• It does not retroactively undo completed training. The announcement concerns preferences governing future training use.
• It does not automatically govern every AI provider. The shared AI Training signal is being developed through industry standards work, while Microsoft's page-level extensions and AI Grounding preferences are Microsoft-specific.
• It does not override crawl restrictions. A separate setting that prevents Bingbot access may still affect discoverability.

These distinctions are important for any publisher assessing whether technical controls provide adequate protection for its commercial content strategy.

The Larger Shift Is Towards Publisher Choice

Microsoft's announcement reflects a wider debate about the relationship between AI developers and the websites whose content helps support their products.

Publishers want their work to be discoverable.

They may also want to decide whether that work contributes to future AI models, appears in generated answers or becomes available through third-party AI services.

Those objectives do not always align.

The separation of training and grounding preferences acknowledges that a publisher's willingness to participate in search should not automatically determine its preferences for every form of AI use.

For Sydney and NSW organisations, the next few months provide an opportunity to establish a clearer internal policy before Microsoft's transition begins.

That means involving the people responsible for publishing, technology, intellectual property and commercial performance.

The final choice should reflect the organisation's business model, rather than simply following the most restrictive setting available.

Review Your AI Content and Search Visibility Settings

Managing websites, original content or digital publishing assets? Elyment can help review AI-related workflows, website configuration, content governance and the operational requirements of digital visibility.

Request a Digital Strategy Review

The Decision Before February 2027

Microsoft's new controls do not eliminate the tension between content protection and digital distribution.

They give publishers a more precise way to express their preferences.

A website can seek to prevent its content from contributing to future generative AI training while retaining eligibility for ordinary Bing search results.

It can also make a separate decision about whether that content should inform Copilot responses and other supported AI-generated experiences.

The next milestone is 1 February 2027, when Microsoft begins honouring the new preferences as it crawls.

For publishers already using NoArchive or NoCache, the 1 March 2027 transition deadline requires particular attention.

The strategic question is no longer simply whether AI systems can access published material. It is which uses a publisher wants to permit, what visibility it is prepared to exchange, and how those decisions will be implemented and measured.

Sources and References

Microsoft Official Announcements and Documentation

1. Microsoft Bing – Your Content, Your Choice: Clearer Preferences for Training and Grounding (30 October 2026)
2. https://blogs.bing.com/search/2026/10/your-content-your-choice-clearer-preferences-for-training-and-grounding/
3. Microsoft Bing Webmaster Tools – New AI Preferences FAQ
4. https://www.bing.com/webmasters/help/new-ai-preferences-2t1cc16d
5. Microsoft – Introducing AI Performance in Bing Webmaster Tools (February 2026)
6. https://blogs.bing.com/webmaster/2026/2/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview/
7. Microsoft – Introducing Topics Portfolio Beta in AI Performance (October 2026)
8. https://blogs.bing.com/webmaster/2026/10/introducing-topics-portfolio-beta-in-ai-performance/

Industry Standards

1. Internet Engineering Task Force (IETF) – AI Preferences Working Group
2. https://datatracker.ietf.org/wg/aipref/about/

Elyment Related Articles and Case Studies

1. Elyment – JL Turf Group Sydney Sports Surfaces Case Study
2. https://elyment.com.au/blog/jl-turf-group-sydney-sports-surfaces-case-study
3. Elyment – BMW Put Its Car Configurator Inside ChatGPT: Is Conversational AI Becoming the New Sales Website?
4. https://elyment.com.au/blog/bmw-put-its-car-configurator-inside-chatgpt-is-conversational-ai-becoming-the-new-sales-website
5. Elyment – Google Is Connecting Gemini to Angi, Thumbtack and Wix: Is AI Becoming the New Front Door for Service Businesses?
6. https://elyment.com.au/blog/google-is-connecting-gemini-to-angi-thumbtack-and-wix-is-ai-becoming-the-new-front-door-for-service-businesses
7. Elyment – Contact and Digital Strategy Review
8. https://elyment.com.au/contact/