Prepare a migration brief before changing your site's AI preferences. The useful deliverable this week is an agreed policy, a list of affected pages and a testable release plan. A provider announcement alone is not evidence that your chosen settings are already being applied.
This is Lemon SEO's recommended planning workflow for publishers and marketing teams. The examples are hypothetical, and the checkpoints are our recommendations rather than additional provider requirements. The source review is dated November 6, 2026; recheck provider guidance before implementation.
What changed and what has not
On October 30, Google described a shared training preference in robots.txt while retaining company-specific controls. It aims to support the mechanism in 2027. Treat that as a planned rollout, not a feature already confirmed across providers. Google's proposal and announced timing.
The mechanism uses the policy token __usage_ai-training with familiar robots.txt path rules. Its individual Internet-Draft is not an endorsed IETF standard. Draft syntax and document status.
Existing controls have their own history. Google introduced Google-Extended on September 28, 2023, to give publishers a choice about improving the generative products named in that announcement. Original announcement.
In his October 30 mailing-list message, draft author Kevin Kelley said Anthropic had started implementation, would need time to add support across services, and might adapt its interpretation as standards evolved. This is implementation-in-progress evidence. Author's message.
Separate the decisions before changing files
The illustration below shows two independently controlled channels leaving one set of pages. Use that model to discuss distinct content uses, rather than asking a stakeholder to approve an undefined “AI on” or “AI off” switch.
Microsoft's announcement separates training and grounding preferences while preserving eligibility for traditional Bing results, subject to normal crawling and indexing. Microsoft's control model.
Google says that, absent an explicit Google-Extended instruction, it plans to interpret the shared training opt-out as Google-Extended being disallowed. Do not assume the same precedence everywhere. Google's stated interpretation.
Our recommended policy worksheet asks the owner to answer:
- Which page groups should participate in each intended use?
- Which provider-specific exceptions are deliberate?
- Who can approve a change to those decisions?
- What evidence would justify revisiting the policy?
Build a migration inventory
Start with the site's current configuration and the person responsible for it. Use the earlier Google-Extended and crawler audit for background, then create a new migration record tied to the current announcements.
We recommend this sequence:
- Name the policy owner. Record who decides the intended content use and who implements the configuration. Do not treat a developer's ability to edit a file as approval of the business policy.
- Choose a representative URL sample. Include one page from each relevant template and any deliberately different section. Write down why each URL belongs in the sample so another reviewer can repeat the check.
- Capture the current output. Save the applicable robots.txt content and the sampled pages' response headers and rendered directives.
- Map the desired change. For each sample, state the intended preference, current setting, proposed setting and unanswered question. Keep an explicit “decision pending” entry wherever the owner has not chosen.
- Assign a review date. Ask an owner to recheck the linked provider guidance before implementation and after any announced change to support or syntax.
Put the announced dates on an owner calendar
For sites using NoArchive or NoCache on December 15, 2026, Microsoft opens temporary grounding preferences that day. Its announced milestones are:
- February 1, 2027: new preferences start taking effect as Microsoft crawls.
- Before March 1, 2027: eligible legacy users preserving grounding restrictions must publish page preferences and record the temporary Bing Webmaster Tools choice; training requires its own published preference.
- September 30, 2027: temporary grounding choices expire, unless replaced earlier when a page is recrawled and its new preference discovered.
Without the temporary grounding choice by March 1, legacy-tag pages default to allowing grounding until discovery of their new page preference. Transition scope and dates.
Use an evidence table to decide what can ship
| Audit finding | Recommended action | Evidence to retain |
|---|---|---|
| Policy owner has not chosen a preference | Hold the configuration change | Owner and unresolved decision |
| Desired policy differs across page groups | Prepare separate test cases | Approved URL sample for each group |
| Staging output differs from the approved diff | Repair before release | Response capture and expected output |
| Live output matches but provider discovery is unconfirmed | Record publication and keep discovery pending | Timestamped response and follow-up owner |
The comparison table below maps hypothetical findings to recommended next actions. It distinguishes a missing policy decision, a need for more test coverage, a failed local check and an unresolved provider observation. These are different reasons to hold or qualify a release report.
Consider a fictional publisher with public tutorials and a separate research section. The owner wants those groups reviewed independently. There is no client result, traffic forecast or recommended commercial policy in this example. Work through it as follows:
- Choose one tutorial URL and one research URL as initial test cases, documenting their template and section.
- Ask the owner to record the intended treatment for each group. Leave unresolved choices visible instead of letting the current setting silently become the decision.
- Compare the proposed configuration with those choices. If the sampled groups need different treatment, prepare separate expected outputs for review.
- Use the table to select the next action. A correct staging response can support a release decision; it cannot answer whether a provider has discovered the production setting.
Test a small release and preserve the result
Microsoft distinguishes publishing a setting from discovering it through crawling. IndexNow can signal a change but does not promise a recrawl date. Discovery limits.
Use a bounded release checklist, informed by the technical SEO audit checklist:
- Prepare the smallest diff that implements the approved policy for the intended scope.
- Compare staging output with the approved expected output for every sampled URL. Keep the before-and-after responses with the review record.
- Ask a second reviewer to check both the business intent and the actual change. Resolve disagreements before scheduling the release.
- Retain the previous configuration and agree who can restore it if the released output differs from the approved version.
- After release, capture the public responses again. Report “published and checked” only for the settings actually observed.
- Track provider discovery separately, using available evidence. If it remains unconfirmed, retain that status and a follow-up date rather than marking the whole migration complete.
Avoid these migration mistakes
The draft separates content-use preferences from crawling instructions; the policy token itself is not a crawl directive. Draft's crawl boundary.
Our review checklist therefore flags these risks:
- An unclear objective: an “AI opt-out” ticket with no specified use, provider or page scope. Return it to the policy owner for a concrete decision.
- An assumed rollout: a planned support date entered as proof that the setting is active. Keep a dated link to the announcement and a separate observation field.
- A missing exception review: a shared preference added without checking deliberate provider-specific choices. Include those choices in the diff review.
- An overstated outcome: a successful response check reported as a ranking, traffic or visibility result. Restrict the report to the evidence collected.
Turn the review into a bounded brief
If you need help mapping publisher controls to an evidence-led visibility review, explore AI search services. To start with the current settings and the scope of a possible audit, Request a free audit. Bring the existing configuration and the decisions still pending; the first objective is a clear, testable brief.