Google Generative-AI Search Control: Record Scope, Baseline and Rollback

Document eligibility, property inheritance, policy, baseline metrics, verification, propagation, observation, and rollback before changing Google’s AI Search control.

Sonar braces a master lever beside three paused search channels while holding a baseline sheet and keeping the rollback cord within reach.

Direct answer: Google Search Console’s generative-AI control lets eligible property owners include or exclude their site’s links and content from AI Overviews, AI Mode, and generative-AI features in Discover. Exclusion prevents traffic and impressions from those features. Before changing it, record the property scope, inherited state, policy objective, performance baseline, approver, observation window, and rollback rule.

The control is still rolling out to a subset of website owners. It is not a general noindex switch, does not control AI training, and is not used as a ranking or inclusion signal for other parts of Search. If the setting or generative-AI performance report is absent, document ineligibility instead of inventing implementation steps.

Confirm that the control exists for your property

Open the exact verified Search Console property and look under Settings for the Search generative AI control. Save the property identifier, property type, verified owner, access role, current label, inherited or manually configured state, and timestamp. A screenshot from another site or market is not evidence that your property is eligible.

Google’s current help page says the rollout covers a subset of website owners for testing before expanding further. The earlier June 2026 announcement described a subset of UK sites. The current product page—not an old screenshot—should govern your operational decision.

Record the three in-scope surfaces

Documented effect and the inference to avoid
Surface or systemEffect of exclusionDo not infer
AI OverviewsSite links and content cannot appear or help ground responsesOrdinary Web ranking is switched off
AI ModeSite links and content cannot appear or help ground responsesEvery search impression or referral disappears
Generative-AI features in DiscoverSite links and content cannot appear in the named AI featuresAll Discover eligibility ends
Other parts of SearchGoogle says the control is not a ranking or inclusion signalOther Search systems can never change
AI trainingNot controlled by this settingExclusion is equivalent to Google-Extended

Google says excluded content may still help power Search as a whole, such as language understanding. It points site owners to Google-Extended for certain training controls and to noindex for blocking a page from Google Search completely. Keep those controls distinct in the decision record.

Check property inheritance before changing anything

A Search Console child property can inherit the closest configured parent property’s choice. A domain property, subdomain property, or URL-prefix property can therefore affect nested properties. Record the parent chain and identify the closest manually configured control before assuming the visible property is independent.

For a deliberate exception, document why the child property should stop inheriting and who owns the boundary. Include protocol, hostname, and path exactly. Confusing http with https, or www with a bare domain, can produce a decision record for the wrong property.

Separate policy from performance

A publisher may exclude content because the proposed use conflicts with licensing, consent, editorial independence, or a collective policy. Another publisher may include content because discoverability and attributed visits are central to its mission. Neither choice has to be disguised as a traffic experiment.

Write the governing principle first. Then state which measurements could inform the decision without overriding it. This prevents a short-term impression change from silently replacing a rights or audience policy approved by the organization.

Two decision layers that should not be collapsed
LayerExamplesDecision ownerEvidence
Governing principleLicensing, consent, mission, competitive posture, user expectationsPublisher, legal, editorial, product leadershipPolicy, contracts, risk assessment
Observed performanceAI impressions, referrals, subscriptions, conversions, direct audienceAnalytics and product ownersDefined reports and comparable windows

Save a baseline before the change

Export the longest comparable pre-change window available. Where the generative-AI performance report is present, save page, country, device, and date dimensions plus the report’s property-level total. Also preserve ordinary Web Search, Discover, analytics referrals, conversions, subscriptions, and direct-audience measures under their own definitions.

The current generative-AI Search report covers impressions from AI Overviews and AI Mode. Google directs Discover measurement to a separate report. It can be unavailable because rollout is incomplete or the property has insufficient impressions. Record that missingness rather than substituting the general Performance report as though it were the same dataset.

Use the generative-AI reporting guide for field definitions and the AI visibility measurement crosswalk to prevent impression, citation, referral, and conversion metrics from being merged.

Record confounders before observing the result

Seasonality, news cycles, product launches, demand, ranking changes, migrations, consent changes, analytics releases, reporting incidents, and other editorial releases can move the same metrics. List known confounders and their dates in advance. Add new ones during the observation window without rewriting the original hypothesis.

Google’s reports have aggregation and row-limit rules. Property totals can differ from page tables, and newest data can be preliminary. The site’s generative-AI logging incident analysis shows why report reliability belongs in the decision record.

Write the decision before the outcome is known

Name the selected control state, exact property, parent inheritance, business objective, governing principle, expected benefit, accepted cost, accountable owner, approver, change window, observation period, review method, rollback trigger, and next review date. Attach the source snapshot and baseline exports.

Use a falsifiable expectation where performance matters. “We expect a change” is too vague. A better record names the metric, direction, comparison window, minimum material threshold, and conditions that invalidate the comparison. It can still conclude that the policy choice holds regardless of the measured outcome.

Apply the control with two-person verification

  1. Confirm the exact property and current inherited or manual state.
  2. Confirm the approver and change ticket match the decision record.
  3. Capture a timestamped before screenshot or export.
  4. Choose include, exclude, or inherit as approved.
  5. Capture the resulting label and the operator identity.
  6. Have a second authorized reviewer verify the property and state.
  7. Annotate analytics and the release ledger.

Google says an exclusion change generally takes a few days to take effect and content is typically excluded within one to two days after the control goes live, with some longer propagation because of caching. Do not treat the click time as an instantaneous exposure boundary.

Observe without claiming causation

Minimum comparison record
FieldRecordReason
Control effective windowChange time plus documented propagation allowanceAvoid mixing transition days with stable state
Pre/post windowsDates and exclusion rulesMake the comparison reproducible
Report definitionsSurface, dimension, aggregation, timezonePrevent incompatible totals from being joined
ConfoundersDated releases, incidents, demand changesBound causal language
Decision resultKeep, reverse, extend, or inconclusiveConnect evidence to action

Describe the result as an observation for this property and period. A pre/post change cannot isolate the control from every concurrent system. If Google changes the feature, report definition, eligibility, or inheritance behavior, close the old observation and open a new one.

Keep other Google controls separate

The Search generative-AI control is one layer. Google Search Central still points to Googlebot access and snippet controls such as nosnippet, data-nosnippet, max-snippet, and noindex for other inclusion or presentation goals. Google-Extended addresses certain training and grounding uses outside Search.

Do not apply noindex to solve a narrower generative-AI policy question unless removal from Google Search is truly intended. The Google AI preview controls guide explains the snippet-control boundary. Record every control independently so later reviewers can reconstruct the full state.

Download the control decision record

The CSV uses one row per property decision or review. It captures eligibility, inheritance, surface scope, governing principle, baseline locations, confounders, approved state, verification, propagation allowance, observation result, and rollback. Remove the EXAMPLE-REMOVE sample row before use.

Download the Google generative-AI Search control decision record (CSV)

Limits and re-audit triggers

This framework cannot predict the business effect of inclusion or exclusion. The control and reports are still rolling out, AI features change, and Search Console data is aggregated. A site’s audience, rights position, market, and revenue model can justify a different choice.

Re-audit when Google expands eligibility, changes the covered surfaces, modifies inheritance, revises the performance report, changes propagation guidance, or updates the relationship to other Search controls. Also review after a property migration, ownership change, major policy change, or measurement incident.

Primary sources

Source check: August 29, 2026. Verify the live Search Console help pages before changing a property setting.

Keep learning

Continue this topic

Community discussion

Discuss: Google Generative-AI Search Control: Record Scope, Baseline and Rollback

Have a question, a useful example, or a different perspective? Join the discussion, share evidence, and help other readers reach a better answer.

0 replies Moderated
No replies yet.

Be the first to ask a focused question, share a practical example, or add useful evidence.

Ask a question or join the discussion

Share evidence, a useful example, or a clear question. Be specific, stay on topic, and challenge ideas without attacking people. First-time replies may be held for moderation.