AI training opt-out is coming—but visibility rules differ

Google, Microsoft and OpenAI announced plans on October 30, 2026, to honor a shared AI training opt-out signal through robots.txt. That could reduce the need for separate training instructions for participating providers, but it does not give marketers one uniform control over AI-answer visibility. Implementation is still ahead, and provider-specific rules remain central to the decision.

The AI training opt-out is announced, not live everywhere

The first distinction to make in a client update or internal briefing is status: this is an announced approach, not a completed rollout or finalized industry standard. A shared preference should not be presented as protection already operating across the participating companies.

Google calls the mechanism a draft proposal and aims to support it in 2027. Microsoft will begin honoring its new preferences on February 1, 2027, as it crawls. OpenAI plans implementation in the coming months without naming an exact launch date.

OpenAI also describes participation as voluntary, not universal. Its existing crawler controls will remain available and respected. Google likewise says company-specific controls will coexist with the shared signal.

For marketers, the practical implication is straightforward: keep existing provider-specific controls while implementation is pending. Use the announcement to review policy and prepare changes, rather than treating it as a reason to replace current instructions immediately.

An agency update should distinguish what has been announced, when each provider expects to act, and which decisions still need client approval. That is more useful than telling every account that a single new setting will resolve its content-use concerns.

Google’s proposed default makes explicit choices important

Google’s proposal deserves particular attention because of its connection to Google-Extended. If a website expresses the shared training preference without explicitly specifying Google-Extended, Google plans to treat that preference as a Google-Extended disallow.

That makes the absence of an explicit provider-specific instruction important. Before approving a shared restriction, marketers should review whether the site already specifies Google-Extended and whether that setting reflects the business’s intended policy.

Google says websites explicitly allowing Google-Extended remain eligible to appear through grounding in Gemini Apps. Grounding means using content to inform AI answers. Eligibility is not a promise that a page will appear, so avoid presenting an allow setting as a visibility guarantee.

The policy conversation should therefore cover more than whether the business wants its content used for training. It should also address whether the business wants content used in AI answers and how Google-specific exceptions should be handled.

A shared training preference is not a shared visibility policy.

For an in-house team, this calls for agreement before a developer changes the configuration. For an agency or freelancer, it calls for account-specific approval rather than copying the same restriction across unrelated sites.

Microsoft separates training from AI-answer visibility

Microsoft’s announced approach separates training permissions from page-level grounding permissions. It says these preferences do not change eligibility for traditional Bing results, while restricting grounding reduces visibility in AI experiences.

That separation matters when a client wants to restrict model training but still wants content used in AI answers. Those goals should be documented independently rather than compressed into a general instruction to block AI.

It also makes the scope of the review clearer. A robots.txt inventory alone is not enough for the work described in the brief: marketers should also account for page-level tags and HTTP headers when reviewing current restrictions and planned changes.

Microsoft has a separate transition requirement for existing NoArchive or NoCache users. It directs those users to transition before March 1, 2027. Maintaining grounding restrictions requires publishing page-level preferences and recording a temporary preference in Bing Webmaster Tools.

Treat that as an assigned migration task, not a note buried in a general AI strategy document. Identify affected sites, record the intended grounding policy, and name the person responsible for completing the required steps. Microsoft’s February start date applies as it crawls; it should not be described as an instant, simultaneous change across every site.

What marketers should do this week

The most useful immediate work is an inventory and approval process. The announcement gives teams a reason to reconcile settings with business intent before making technical changes.

Inventory the current controls. Collect robots.txt rules, AI crawler settings, page-level tags and HTTP headers. Record who approved each restriction. If the original decision-maker or rationale is unclear, flag that gap instead of assuming the setting still reflects the business’s wishes.

Separate training from answer visibility. Ask whether the business permits model training and whether it wants content used to inform AI answers. Document the answers separately, along with any provider-specific exceptions. Do not promise that the same preference produces identical effects everywhere.

Review clients individually. Agencies should avoid a blanket rollout across their portfolio. Freelancers should obtain approval before changing permissions. An in-house marketer should bring the proposed policy and the current configuration into the same review so that business decisions are not left implicit.

Assign Microsoft’s transition work. For any site using NoArchive or NoCache, identify an owner and schedule the required transition before March 1, 2027. Include both the page-level preferences and the temporary Bing Webmaster Tools preference when maintaining grounding restrictions.

Get technical review before deployment. Have a developer review proposed changes. A technical SEO audit is a relevant place to include this configuration review, but the permission decision still needs business approval. If content rights or licensing obligations drive the policy, check with a lawyer.

OnlyaClick tip: Keep one approval record that lists each provider, the intended training permission, the intended grounding permission and the person authorized to approve changes.

Keep client reporting equally precise. Describe planned support as planned, preserve existing controls during the pending implementation, and avoid claiming that a future shared preference is already being honored everywhere.

The bottom line

The shared signal could simplify how websites communicate training preferences to participating providers. It does not remove the need to understand Google-Extended, Microsoft’s separate grounding controls or the implementation schedule for each company.

For marketers, the next move is policy before deployment: inventory current restrictions, separate training decisions from AI-answer visibility, obtain approval and prepare Microsoft’s transition where relevant. A common signal can reduce duplicated instructions without making every provider’s rules the same.