Enterprise AI Search Optimization: The Complete 90-Day GEO Implementation Playbook

Most AI search advice assumes you have one domain, one brand, and a team that can publish a page this afternoon. That advice gets harder to use when procurement is involved.

At a large organisation, getting into AI answers means coordinating people who have other deadlines. Legal must approve each claim. Content can’t publish during a regulatory freeze. The platform team has two sprints available this quarter. You’re also managing forty domains acquired over time. The technical fix often isn’t the bottleneck. You need someone to own the decision, a clear definition of done, and someone who knows where the AI inclusion setting lives.

That’s what this playbook is for.

One piece of vocabulary first, because the acronym is ambiguous. Throughout this article, GEO means generative engine optimization - the practice of earning presence in AI-generated answers. Be aware that Google itself uses “GEO” in a different sense, meaning the generative experiences on Google Search and Discover; when a Google document says generative AI performance “on Google Search and Discover,” that’s their GEO, not the marketing category.

What 90 days is, and what it is not

Ninety days is a planning horizon, not a promise. It’s the shortest window in which an enterprise can move from “we have no idea how we appear in AI answers” to “we can attribute specific changes to specific outcomes and defend them in a steering committee.” It’s long enough to build instrumentation, ship a bounded set of changes, and observe them. It’s far too short to move consolidated authority from one domain to another, to rebuild a content supply chain, or to change how an enterprise manufactures credibility.

I wouldn’t promise rankings, and neither should a vendor. This playbook should help you answer the finance director’s question: what did we actually change, and how would we know if it hadn’t worked?

The foundation for everything below is Google’s own guidance. In the Guide to optimizing for generative AI features (last updated 10 July 2026), Google states plainly that its generative features are “rooted in our core Search ranking and quality systems,” and describes them as retrieval-augmented generation over the Search index plus query fan-out. Google also mythbusts the industry favourites: you don’t need “new machine readable files, AI text files, markup, or Markdown” for Google Search, there’s no chunking requirement, you don’t need to rewrite content for AI systems, chasing inauthentic mentions is counterproductive, and structured data isn’t required for generative AI search.

That’s a useful foundation for a change programme because it removes most of the tactics that consume budget without a defensible mechanism.

Why this is an organisational problem

The instinct inside a large company is to treat AI search as a content initiative or a technical SEO backlog. Both framings fail on contact with the organisation.

Content framing fails because the thing that determines whether you’re cited is rarely a new article. It’s whether your pages are crawlable and eligible, whether your entity is unambiguous, whether your claims survive scrutiny, and whether the surfaces where your details live are consistent. Microsoft’s Bing Webmaster Tools AI Performance dashboard (February 2026) reports citation counts at page level - which URL of yours gets cited, and how often - and names “reduce ambiguity across formats,” meaning aligning text, images and video so they consistently represent the same entities, products or concepts. Entity consistency is an architecture problem, not an editorial one.

Technical framing fails because the decisive configuration lives outside the codebase. Google’s Search generative AI control is a switch in Search Console, rolled out to all websites worldwide as of 31 August 2026, that determines whether your content appears in AI Overviews, AI Mode, and generative AI features in Discover. It inherits from parent properties. A property that nobody audited can quietly govern a whole portfolio.

Give each workstream an owner, a place in the teams’ calendars, and a written definition of done. A task in a backlog doesn’t settle those dependencies.

The six workstreams

The six workstreams are deliberately cut along lines of ownership, not lines of tactic, because ownership is what actually blocks an enterprise.

#	Workstream	What it settles	Primary owner
W1	Baseline & instrumentation	What we can actually see and measure today	Analytics / Insights lead
W2	Entity & knowledge graph	Which entity Google should associate with which domain	Brand or SEO lead
W3	Content & product surfaces	What we publish and where our structured facts live	Content lead
W4	Third-party corroboration	Who independently confirms us	Comms / PR lead
W5	Governance, risk & compliance	What we may switch on, and who signs it off	Legal & Compliance lead
W6	Measurement & reporting	Whether the programme should continue	Programme sponsor

Two design choices are worth defending.

First, W5 is a workstream rather than a checkpoint. Legal review is usually modelled as a gate at the end, which means it arrives after the work and either rubber-stamps it or kills it. Treating it as a parallel workstream with its own owner means the constraints get designed in.

Second, W6 is owned by the programme sponsor, not the analytics team. Measurement framed as an analyst’s dashboard produces insight. Measurement framed as a funding instrument produces a decision. In an enterprise, those are different deliverables.

The Domain Control Map

Before the schedule, one piece of original analysis: the diagnostic I call the Domain Control Map.

Three independently documented Google behaviours combine into a trap that only bites at enterprise scale.

First, in the crawl budget guide (last updated 22 July 2026), Google states that its crawling infrastructure “defines a site as a unique hostname” and that https://www.example.com/ and https://code.example.com/ are “treated as separate sites and have separate crawl budgets.”

Second, in Site names in Google Search (last updated 10 December 2025), Google states that it “only supports one site name per site, where a site is defined by the domain or subdomain,” that it doesn’t support site names at the subdirectory level, and - critically for a group with a house style - that “our system generally won’t use the same site name for two different sites that are global in nature.”

Third, the Search generative AI control inherits through the Search Console property tree. A property “inherits the control from its closest parent that has changed its control to stop inheriting,” and by default inherits from its top-level domain property.

Compose those three and you get a structural constraint that no corporate positioning deck addresses:

• Every hostname is a separate crawl budget, so a sub-brand on its own domain doesn’t borrow credibility or crawl attention from the parent.
• Every domain or subdomain can carry exactly one site name, so a group that wants forty sub-brands to present as one entity is asking for something the system can’t do in a single field.
• Every hostname sits under an inherited AI-inclusion switch, so a single misconfiguration at the top level can remove an entire portfolio from AI answers, and a child property that overrides the parent creates an audit surface that the top-level owner is notified about.

The regional footprint compounds it, because the structure choice is itself a strategic decision with search consequences. Google’s multi-regional sites guide (last updated 10 December 2025) sets out the options and their costs plainly: country-specific domains give “clear geotargeting” but are “expensive,” “require more infrastructure” and “can only target a single country”; gTLD subdirectories are “low maintenance (same host)” but make “separation of sites harder”; and URL parameters are flatly “Not recommended.” An enterprise that acquired forty domains now has to decide which of them justify the infrastructure cost - and that decision, not the SEO tactic, is usually what takes ninety days.

Google’s own framing of the Organization structured data guide (last updated 8 September 2026) tightens this. It instructs you to “use the same name and alternateName that you’re using for your site name,” and documents identifiers including sameAs, legalName, iso6523Code (recommending the 0199 prefix for Legal Entity Identifier), naics, vatID, taxID and numberOfEmployees. Those fields exist to disambiguate one organisation from another. They don’t merge forty sub-brands into one name.

The deliverable: a one-page map, maintained as a living artefact, with one row per hostname containing: the hostname; which Search Console property owns it; the site name Google currently shows, verified in the UI rather than assumed; the AI-inclusion control value and whether it’s inherited or overridden; the entity IDs asserted in Organization markup; and the owner who can change it. Give this artefact an owner and a review cadence and it does more than any audit deck, because it converts an invisible inherited setting into an auditable line item.

This is a substantial change-management task. For most enterprises it’s the single most valuable thing produced in the first thirty days.

The 90-day sequence

Set up measurement before making changes. Check eligibility before optimising. Agree who can approve a switch before making it irreversible. Getting that order wrong causes most of the enterprise failures described here.

Days 1–30 - Establish ground truth and remove blockers

W1, Baseline. Don’t start with a vendor tool. Start with what you already own and verify it. The Search generative AI control page tells you to use the Generative AI performance report to understand how changing your control may affect traffic - so confirm that report is reachable for your top-level property and that no child property has silently overridden the setting. Then audit property coverage: Search Console properties confirms that a Domain property aggregates all subdomains, protocols and subpaths, that a URL-prefix property captures only URLs beginning with that prefix including the protocol, and that accounts can hold up to 1,000 properties. Enterprises routinely have a verified domain property and forty unverified prefix properties that nobody looks at.

Check your analytics classification. This is where the incumbent advice is most likely to be wrong. The GA4 default channel group documentation defines a dedicated AI Assistant channel for arrivals “from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok,” and - this is the detail that changes reporting - states that it “excludes Google’s AI Overviews and AI Mode,” because Organic Search covers “non-ad links in organic-search results, including Google’s AI Overviews and AI Mode.” If your organisation bought a tool because AI referrals were “falling into Referral,” check that premise before renewing.

Exit criteria: Domain Control Map published and owned; every hostname has a named control owner; AI Assistant and Organic Search segments exist and have a baseline; Generative AI performance report reachable or its absence documented.

W5, Governance (starts here, not later). Three decisions need written sign-off before anything is switched:

1. Inclusion. Default is include. Exclusion means your content is prevented from being visible in Search generative AI features and you receive no traffic or impressions from them. Google’s page notes that content is typically excluded within one to two days of the control going live, with some content taking longer due to caching. Treat this as a business decision with a named approver, not a default.
2. Training. Google’s page is explicit that the Search generative AI control “doesn’t affect AI training,” and directs those who want to limit training to Google-Extended instead. Google also separates full removal via noindex. You now have a three-layer model - Search generative AI control for inclusion, Google-Extended for training, noindex for full removal - and the layers are routinely conflated in board briefings.
3. Claims. In the US, the FTC Policy Statement Regarding Advertising Substantiation (appended to 104 F.T.C. 648, 839 (1984)) reaffirms “that advertisers and ad agencies have a reasonable basis for advertising claims before they are disseminated,” and states that “a firm’s failure to possess and rely upon a reasonable basis for objective claims constitutes an unfair and deceptive act or practice in violation of Section 5.” 16 CFR 260.2 restates the operational duty: marketers “must ensure that all reasonable interpretations of their claims are truthful, not misleading, and supported by a reasonable basis before they make the claims.” The AI-specific sting is that a compressed, decontextualised restatement of a claim is still your claim being disseminated.

Exit criteria: A one-page AI inclusion policy with an approver; a claims-substantiation rule added to the content workflow; the three-layer control model explained to the steering committee.

W2, Entity. For each hostname, reconcile what Google shows against what the organisation wants shown, then assert the identifiers. Google’s Organisation documentation covers sameAs links to profiles elsewhere on the web, legalName, vatID, taxID, and standard identifiers. Where a group has multiple operating entities, be honest about it - the disambiguation fields exist precisely because a single undifferentiated brand name is ambiguous to the system.

Exit criteria: One approved site name and one set of alternateName values per hostname, documented; Organization markup deployed on home pages; current site names observed in the UI recorded against each row of the map.

Days 31–60 - Ship bounded changes through the approval path

W5 → W3 pipeline. Build the claims-substantiation gate into the editorial workflow now, while the volume is small. The Creating helpful, reliable, people-first content guide (last updated 5 October 2026) supplies an unusually usable review framework in its “Who, How, and Why” structure - who created the content, how it was produced including any automation, and why it was created - and it singles out YMYL topics where “our systems give even more weight to content that aligns with strong E-E-A-T.” For regulated or safety-relevant portfolios, that framework is the natural backbone of a substantiation checklist.

W3, Content. Be deliberately narrow. Don’t commission content per fan-out query: Google’s guide is explicit that creating separate content for each variation “primarily to manipulate rankings or generative AI responses in Google Search violates Google’s scaled content abuse spam policy.” Invest instead in depth on the topics where you have genuine expertise, which is exactly what Google’s “non-commodity content” framing rewards.

Then fix the structured surfaces, which are unglamorous and high-yield. Google’s Product structured data documentation (last updated 10 December 2025) distinguishes Product snippets from Merchant listings and notes that providing both on-page structured data and a Merchant Center feed “maximizes your eligibility to experiences.” Where you publish third-party-informed comparisons or reviews, Google’s guide to writing high-quality reviews (10 December 2025) sets the bar: quantitative measurements, original research, explicit benefits and drawbacks, and first-hand evidence when calling something best overall.

Bing’s October 2025 guidance on inclusion in AI search answers describes the same problem from the other side: assistants “break content down” into pieces that can be evaluated and assembled into answers, so self-contained phrasing, real headings, and answers that survive being lifted out of context matter more than raw volume. Microsoft is blunt about the limit of the advice - “there’s no secret sauce that guarantees selection in AI answers” - and its practical warnings are ones an enterprise already knows to worry about: long unbroken walls of text, and key answers hidden behind tabs that systems may not render.

Exit criteria: 20–40 priority pages revised under the new substantiation gate; structured data validated on a sample; no new claim published without a recorded basis.

W2 technical. Crawl hygiene is a prerequisite at scale. Google notes that soft 404s “will continue to be crawled, and waste your budget,” recommends consolidating duplicate content, returning proper 404 or 410 status codes, and warns against long redirect chains. Its audience thresholds are a useful self-assessment: large sites at one million or more unique pages with moderately frequent change, or medium-plus sites at ten thousand or more unique pages with daily change. If you clear either bar, the crawl budget guide is mandatory reading, not optional.

If your estate is built on client-side routing, budget for a second problem. Google’s JavaScript SEO basics (last updated 4 March 2026) explains that client-side routed single-page apps often can’t return meaningful HTTP status codes, which creates soft 404s - the same waste from the paragraph above - and that using URL fragments for navigation means “Googlebot can’t reliably resolve the URLs.” It recommends the History API and server-side or pre-rendering. This is the point at which a crawl-hygiene ticket becomes a platform programme rather than an SEO task, and it’s worth surfacing early rather than discovering it in week nine.

Exit criteria: Soft-404 and redirect-chain audit completed on the top 100 templates; crawl waste fixes merged; canonical conflicts in the multi-region footprint resolved.

Days 61–90 - Corroborate, measure, and decide

W4, Third-party corroboration. This is where AI search differs most sharply from classic SEO, and it requires a discipline most marketing teams lack. Google warns that “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.” The legitimate version of this workstream is earned coverage, analyst inclusion, standards participation, and accurate directory and marketplace profiles - all of which are slow.

Two enterprise-specific traps sit here. First, Google’s Business Profile guidelines require “only one profile per business” and impose strict name-consistency rules, permitting a distinct sub-brand name only where it “is consistently applied to all locations of that business.” Second, and more bluntly, the same document states that for products Google “do[es] not allow content related to regulated products and services, including alcohol, tobacco products, gambling, financial services, pharmaceuticals and unapproved supplements, or health or medical devices,” and warns that non-compliant submissions “may result in removal of the entire product catalog.” For a regulated enterprise, catalogue hygiene is a risk-control task, not a marketing task.

W6, Measurement. Set the reporting cadence and decide what would make you stop.

Exit criteria: A repeatable monthly report; a written go/no-go decision at day 90; a named owner for the next quarter.

RACI: who actually owns what

Responsible, Accountable, Consulted, Informed. One accountable role per row - this is the discipline that prevents the programme dying in committee.

Activity	Analytics	SEO/Content	Web Platform	Legal	Comms	Sponsor
Domain Control Map	C	R	C	I	I	A
Search Console property audit	C	R	I	I	–	A
AI inclusion control decision	I	C	C	R	I	A
Training vs inclusion policy	I	C	I	R	I	A
Claims substantiation rule	I	C	–	R	C	A
Site name per hostname	C	R	I	I	C	A
Entity IDs (LEI/NAICS/VAT)	I	C	I	R	I	A
Structured data implementation	I	C	R	I	–	A
Content revision (substantiated)	I	R	I	C	C	A
Crawl waste remediation	C	C	R	–	–	A
Third-party corroboration	I	C	–	C	R	A
Directory / marketplace hygiene	I	C	I	C	R	A
Baseline report	R	C	I	I	I	A
Day-90 go/no-go	C	C	C	C	C	R/A

Two patterns matter. Legal is Responsible - not Consulted - for anything that changes an inclusion switch, defines a substantiation rule, or asserts a regulated identifier, because those are decisions with legal consequence. Web platform is Responsible for everything that requires a deploy, which is why capacity planning has to happen in days 1–30 or the pipeline stalls in days 61–90.

Dependency and risk register

#	Risk	Likelihood	Impact	Mitigation	Owner
R1	A child property overrides the AI inclusion control, silently excluding a sub-brand	Medium	High	Domain Control Map; monthly reconciliation of inherited vs overridden	SEO lead
R2	Developer capacity evaporates; no deploys land in the window	High	High	Book platform capacity in days 1–30; ship markup-only changes first, which are usually CMS-level	Sponsor
R3	Content freeze swallows the whole window	High	Medium	Schedule the freeze calendar before day 1; prioritise revising existing pages over publishing new ones	Content lead
R4	Legal review becomes a late-stage veto	Medium	High	Legal is a workstream from day 1 with its own deliverable, not a gate	Legal lead
R5	Brand architecture cannot be expressed in the site name field	High	High	Accept per-domain names; resolve ambiguity through Organization identifiers instead	Brand lead
R6	Fan-out content gets commissioned, tripping scaled content abuse	Medium	High	Written policy: no page exists to target a fan-out query; spam policy is in the training pack	SEO lead
R7	Regulated-product catalogue violation causes bulk removal	Low–Medium	Very high	Audit listings against Business Profile and Merchant policies before any expansion	Comms lead
R8	Bots blocked at the WAF or CDN, invisible to the crawl team	Medium	High	Audit allowlists explicitly; Google’s AI features guidance notes crawling must be allowed “by any CDN or hosting infrastructure”	Web platform
R9	robots.txt grows past what crawlers will parse	Low	High	RFC 9309 requires parsers to handle at least 500 KiB; with dozens of AI user-agent blocks plus long Disallow lists this gets close on enterprise estates. Measure, do not assume	Web platform
R10	A procurement process locks the team into an AI-visibility tool whose premise is already obsolete	Medium	Medium	Validate the GA4 AI Assistant channel first; only buy what your own baseline cannot answer	Analytics lead

R8 and R9 are worth dwelling on, because they’re the failure modes that survive every content audit. OpenAI documents four distinct user agents - OAI-SearchBot for search, GPTBot for training, ChatGPT-User for user-initiated fetches, and OAI-AdsBot - and states plainly that “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers,” while also noting that ChatGPT-User is “not used to determine whether content may appear in Search.” Anthropic, whose crawler documentation is dated 7 April 2026, similarly separates ClaudeBot, Claude-User and Claude-SearchBot, and asks sites to “do this for every subdomain that you wish to opt out from.” Perplexity documents PerplexityBot for search and Perplexity-User for user requests, and explicitly warns that WAFs “may need to explicitly whitelist” its bots. A single blanket firewall rule written in a hurry during a security incident can remove a company from AI search entirely, and no crawl report will necessarily flag it.

Crawler access as a change-control item

Changing AI crawler access changes how people can discover your business. Put it through the same change-control process as a deployment. Closing a ticket isn’t enough.

The asymmetry is what makes this dangerous. Allowing a training crawler while blocking a search crawler is a defensible editorial choice. Blocking the search crawler by accident during a WAF cleanup is an outage of a channel you may not have instrumented, and per OpenAI “it can take ~24 hours from a site’s robots.txt update for our systems to adjust,” with Perplexity citing a similar up-to-24-hour propagation window. Budget for propagation delay in your measurement window, not in your incident response.

On the other side, if freshness matters to you, IndexNow is cheap to adopt and worth understanding for enterprise reasons: search engines participating in the protocol “agree that submitted URLs will be automatically shared with all other participating search engines,” so a single submission fans out across the network. It accepts up to 10,000 URLs per request. Bing’s AI Performance announcement explicitly ties IndexNow to keeping content current for citation. Note the caveat: IndexNow notifies participating engines of changes; it isn’t a ranking mechanism, and adoption is by voluntary participation rather than guaranteed coverage.

Making the budget case

The most common failure is a pitch that promises outcomes nobody can guarantee. A better pitch is structural.

Build it on four arguments that survive scrutiny:

The switching cost of doing nothing is a discovery risk, not a traffic risk. If your category is discovered through AI assistants and you’re structurally absent, the loss compounds and isn’t visible in your own analytics, because absence generates no data. This is precisely why the GA4 AI Assistant channel matters as a baseline even though it reports only arrivals - it can’t tell you about the demand you were never in the set for.

The control surface is a risk surface. One setting, inherited across a property tree, can remove your entire portfolio from AI answers. That’s a business continuity item and belongs on the risk register regardless of whether you invest in anything.

Fragmented ownership is the real cost. Forty domains with no owner, forty site names no one has verified, and an inclusion switch nobody audited are three material misconfigurations waiting to happen. The Domain Control Map costs analyst-days and removes a category of risk entirely.

You’re building a capability, not buying a ranking. The instrumentation, the claim substantiation discipline, and the entity model compound. They also transfer: the same entity work improves how you’re understood everywhere, not only in AI answers.

Break the costs out in your proposal. Content revision is the largest line and depends on editorial capacity. Platform work may take a few days but sit in a queue you don’t control. Legal review is a fixed cost that helps the work last.

Measurement: what you can and cannot claim

Google’s own documentation has shifted, and getting this right protects you from an embarrassing board conversation. The AI features page (last updated 10 December 2025) states that AI feature traffic is reported “in the Performance report, within the ‘Web’ search type.” The newer guide (10 July 2026) adds that a site must be included in Search generative AI features in Search Console to be eligible, and points to the Generative AI performance report as the way to measure generative AI performance on Google Search and Discover. Read together, the correct statement is: AI traffic is surfaced within standard Search reporting, and a separate generative AI report now exists. Any blanket claim that Google breaks out no AI traffic is out of date; any claim that the older page is wrong about AI features requiring nothing beyond standard eligibility is also out of date. Reconcile them rather than choosing one.

For Bing, the AI Performance dashboard provides Total Citations, Average Cited Pages, Grounding queries, page-level citation activity and visibility trends. Microsoft is careful in its own framing - Average Cited Pages “does not indicate ranking, authority, or the role of any page within an individual answer” - and that caution should be mirrored in your reporting.

For everything else, be honest about the limits. If you run prompt-based mention tracking, understand what it measures: a sample of your prompts on a given day in a given model version, not a census. Report it as a directional indicator with a stated method, never as a share of market. And if you pull query-level data from the Search Console API, respect the constraints Google documents: the performance data guide (last updated 28 August 2025) states the Search Analytics method “exposes a maximum of 50K rows of data per day per search type,” returns at most 25,000 rows per response, and notes data is typically available after two to three days. On a large estate, query-level completeness isn’t achievable; plan the aggregation before you promise precision.

Similarly unresolved: the llms.txt question. Google’s 10 July 2026 guide says plainly that you don’t need such files for Google Search. Meanwhile the llms.txt proposal site now describes a v2 specification and asserts “thousands of sites publish an llms.txt file, documentation platforms generate one automatically, and Chrome’s Lighthouse audits sites for one.” These are directly in tension, and the llms.txt claim is the proposal’s own. If a stakeholder raises llms.txt, the honest answer is that adoption claims come from the proposal’s authors, Google’s own guidance says Google Search doesn’t use these files, and for a large enterprise the real cost is maintaining a file that will drift out of date. Record the decision and the reason; don’t relitigate it quarterly.

When 90 days is the wrong plan

Commit to the horizon honestly. Ninety days fits when you have a bounded number of domains, an existing editorial function, a working analytics stack, and no active platform freeze.

It’s the wrong frame when any of these are true: you’re consolidating or migrating domains; you’re mid-way through a replatform; a regulatory freeze covers the window; or the organisation has never established baseline measurement. In those cases, the honest first 90 days is instrumentation plus the Domain Control Map - and that’s a legitimate deliverable, not a failure.

The clearest signal that the horizon was wrong is if you exit day 90 unable to answer whether your inclusion control is set the way your business intends. If you can answer that, with a map, a policy and a report, the programme is real regardless of what the citation numbers did.

The first ten days

1. Produce the Domain Control Map for every hostname and give it an owner.
2. Check the Search generative AI control on every Search Console property and record inherited versus overridden.
3. Pull the GA4 AI Assistant and Organic Search segments and write down what they actually show.
4. Open the Generative AI performance report for your top-level property, or record why you can’t.
5. Circulate a one-page AI inclusion policy with a named approver.
6. Add “basis of claim” as a required field in your editorial workflow.
7. Ask legal the three-layer question: inclusion, training, or full removal - and which applies to which property.
8. Get platform capacity booked in the calendar, or accept that days 61–90 ship nothing.
9. Publish your freeze calendar so sequencing can respect it.
10. Agree the day-90 decision in writing now, while nobody is invested in the answer.

That last one matters more than it looks. Programmes that end in an ambiguous review are the ones that never continue.

Sources and References

• Google’s Guide to Optimizing for Generative AI Features on Google Search - Google Search Central, last updated 2026-07-10.
• AI Features and Your Website - Google Search Central, last updated 2025-12-10.
• Search generative AI control - Search Console Help; control rolled out worldwide as of 2026-08-31.
• Add a website or platform property to Search Console - Search Console Help.
• Optimize your crawl budget - Google Search Central, last updated 2026-07-22.
• Site Names in Google Search - Google Search Central, last updated 2025-12-10.
• Organization (Organization) structured data - Google Search Central, last updated 2026-09-08.
• Creating helpful, reliable, people-first content - Google Search Central, last updated 2026-10-05.
• Spam policies for Google web search - Google Search Central.
• Understand JavaScript SEO basics - Google Search Central, last updated 2026-03-04.
• Intro to Product structured data - Google Search Central, last updated 2025-12-10.
• Write high quality reviews - Google Search Central, last updated 2025-12-10.
• Managing multi-regional and multilingual sites - Google Search Central, last updated 2025-12-10.
• Getting your performance data - Search Console API, last updated 2025-08-28.
• Default channel group - Google Analytics Help.
• Guidelines for representing your business on Google - Google Business Profile Help.
• Universal Commerce Protocol overview - Google for Developers, last updated 2026-09-30.
• Introducing AI Performance in Bing Webmaster Tools Public Preview - Bing Webmaster Blog, February 2026.
• Optimizing Your Content for Inclusion in AI Search Answers - Microsoft Advertising blog, October 2025.
• Overview of OpenAI Crawlers - OpenAI platform documentation.
• Perplexity Crawlers - Perplexity documentation.
• Does Anthropic crawl data from the web, and how can site owners block the crawler? - Anthropic Help Center, 2026-04-07.
• IndexNow Documentation - IndexNow.org.
• RFC 9309: Robots Exclusion Protocol - IETF, 2022.
• FTC Policy Statement Regarding Advertising Substantiation - US Federal Trade Commission, 104 F.T.C. 839 (1984).
• 16 CFR 260.2 - Interpretation and substantiation of environmental marketing claims - Electronic Code of Federal Regulations.
• The /llms.txt file, v2 - llms-txt proposal site (cited to document a conflict with Google’s guidance, not as an endorsement).