How to Govern Search Console AI Controls Across Site Sections

How should teams govern AI search choices across site sections?

Map the affected pages and their inherited settings before changing any AI search choice. A narrow decision can have a wider effect if the team overlooks a parent property. The safe starting point is a scope check, not a site-wide switch.

Search Console is Google's tool for website owners. A property is the site or part of a site managed within that tool. A domain property covers a broad set of addresses. A URL-prefix property covers addresses that begin with a specific prefix.

A child property can receive its setting from a parent. This is inheritance: the parent choice passes down unless a closer manual choice applies. Record that relationship alongside the page list. A folder name alone does not explain the effective setting.

Google's Search generative AI control guide describes inclusion, exclusion and inheritance choices. The control covers listed AI features in Search. It is separate from training choices and from removing a page from Search as a whole.

Consider a company with a newsroom, an English service section and a client portal. Those areas serve different people and purposes. One broad instruction to block AI may hide the actual concern. Ask what content should be restricted, where and why.

This guide focuses on that decision. It does not recommend the same setting for every publisher. It provides a way to choose a limited scope, test it and retain a safe route back.

Which choices must remain separate?

Keep search display, model training and private access in separate decision records. Each answers a different question. Combining them can lead a team to use the wrong control.

A public expert guide may be intended for broad discovery. A licensed research report may have different terms. A client file should not be public at all. Start with the content's purpose and rights, then choose the matching control.

• Search participation: decide whether public content can appear in the covered AI search features.
• Training preferences: review the relevant platform's separate rules for model use.
• Search removal: assess whether a page should be absent from the search index.
• Private access: require a valid identity and permission before serving a protected file.

Robots.txt is a file that states crawling preferences. It is not a lock on a document. A public address may still be known or opened by someone else. Use real access checks for private material.

Google's Google-Extended documentation describes a separate token for specified Gemini training and grounding uses. Grounding means supplying source content to help a model produce a factual answer. Google states that this token does not affect inclusion or ranking in Search.

OpenAI's guidance on search and training bots also separates its search and training bots. Do not copy a choice from one platform to another without checking its scope. Similar labels can govern different uses.

If a private file is exposed, deal with that access problem first. Do not wait for a marketing test to finish. Once access is safe, the team can review which public pages should support discovery.

What belongs in the property and ownership map?

The map should connect each property to its covered addresses, current choice and accountable owner. Keep it short enough to maintain. It should answer who may change what, not merely list account names.

Begin with the broadest domain record. Add subdomains and URL-prefix records underneath it. Check the effective parent choice for each section. Note any manual setting that breaks the chain.

Then mark the language variants. Turkish and English pages may share a template but sit under different paths. They may also have different business owners. Do not treat them as one row just because they describe the same service.

• Record the exact property name and address scope.
• Save the current setting and the parent it follows.
• Verify the owner who may change the setting.
• Name the editor and technical contact for that section.
• List important pages that must stay outside the test.

A sample export business illustrates the problem. Its press team manages announcements from Türkiye. Its UK team manages service pages for local buyers. A change requested by either team needs an address-level check before it can affect the other.

FL's guide to location-based communications puts the target market at the centre of planning. That is a useful first-party signal for this map. A shared brand does not remove differences in audience, language or publication purpose.

Generative engine optimisation, also called GEO, concerns visibility in AI answers. Location-based GEO is a different use of the same acronym. Neither label tells you which site section an account setting will affect.

How do you run a limited and reversible pilot?

Change one setting for one well-defined section, after saving its current state. Keep content edits and technical releases outside the test where possible. Multiple changes at once make the result harder to interpret.

Choose a normal baseline period. Mark major campaigns, outages or other unusual events. Those notes help explain differences between the test and the baseline. A neat chart without that context can mislead the next reviewer.

Select a few pages that represent the section. Include a current guide and an older record if both are in scope. Check that each page loads, uses the right language and contains valid sources. Confirm that the chosen property covers each address.

Have an authorised owner make the change. Record the time, reason and previous choice. Share the note with the editor and web team. Do not leave the only record inside one person's account or notebook.

A two-week pilot can be a useful starting plan. It is not a platform promise or a statistically valid result by itself. Extend the period when data is sparse. Allow for the control change to take effect before judging a trend.

Define reversal triggers before the test starts. A wrong section entering the scope calls for correction. An unexpected decline calls for a scope and access check. A single answer on one screen is not enough to settle the business decision.

Name a backup owner as well. The main contact may leave or be away when a problem appears. A shared record of the old choice makes recovery faster. It also reduces dependence on a particular agency.

At the end, write what changed, what stayed stable and what remains unknown. Keep the underlying exports with the note. This gives the next team a usable record rather than a screenshot without context.

How should the results be measured?

Measure display, visits and source references separately before making a business claim. A link shown in an answer is not a visit. A visit is not proof of a sale or a gain in trust.

Google's generative AI performance report guide defines impressions for supported Search features. It describes page, country, device and date views. Missing report data does not establish that a site has no visibility across all AI products.

Keep the report's scope at the top of the file. State which features it covers and which it excludes. Record the date range and time zone. Check whether table totals and chart totals use the same grouping rules.

Next, examine visits to the relevant pages in the site's analytics. Note any measurement limits, including consent settings. A missing tracked visit is not always proof that no person arrived. Avoid blending those counts with impression totals.

A fixed question set can add context. Record each question, language, target market, system and test date. Distinguish a brand mention from a link to the brand's page. Treat this as a sample, not a census of all answers.

FL's guide to a shared PR and SEO strategy provides a second first-party signal: content and technical work need a common purpose. Use that principle when reading the results. Ask whether the right material still supports the intended reader's decision.

A page may receive more attention while an answer uses an old job title. The access test may pass, but the representation still needs work. Correct the source information. Do not let a rising total hide a factual error.

What should a supplier deliver before sign-off?

Require a scope map, decision record, test evidence and reversal plan before signing off the work. A broad promise to manage AI bots is not enough. The deliverable should show exactly what changed.

• Scope: properties, languages and affected addresses.
• Decision: business purpose, old setting and accountable owner.
• Preflight: access, language, source links and search status.
• Test: baseline, change time and comparison method.
• Handover: findings, unknowns and reversal responsibility.

Ask why the supplier needs each permission. Reading a report may not require the ability to change a setting. Grant access for the actual task. Record who retains that access after handover.

If the scope grows, revisit the plan. A new country or language may need a different sample. A result from one section is not proof for the entire site. Check the conditions before applying the same decision elsewhere.

The final note should choose among keeping the setting, reversing it or extending the test. Give a reason for the choice. State the uncertainty in plain words. That turns a technical switch into an accountable publication decision.

The FL PR expert insights archive offers related communications guidance. Use any guide alongside your actual property map. An example workflow is not a decision already made for your site.

Frequently Asked Questions

Clear scope and ownership prevent a small control change from becoming a wider publication problem.