Google confirmed on August 26, 2026 that it is rolling out new google.com/goto redirects in Search. For the person clicking a result, the practical effect is small. The destination still opens. For systems that extract Google’s search results at scale, however, the change can add work because the destination may sit behind an opaque Google-owned redirect that must be resolved before the final URL is known.
Derek Perkins of Nozzle reported seeing the implementation across nearly 100 percent of several residential IP providers and said the new links could not simply be decoded to reveal their destinations. That is one vendor’s observation from a traffic population particularly relevant to anti-scraping measures, not a Google-published global rollout percentage. Google confirmed the change but did not identify rank trackers, AI companies, or SERP scrapers as the specific target. Its explanation was broader: the company said it has a long history of deploying technical measures against evolving forms of abuse.
That is noteworthy, but it is not the important part of this story.
Google has used redirect and click-tracking mechanisms before. goto does not prove that search rankings are becoming irrelevant, that AI systems will suddenly prefer first-party websites, or that businesses need a new discipline. The more consequential development is one Google describes openly in its own documentation: when its AI systems answer a question, they do not necessarily operate from the same source set a marketer sees in the conventional results for that query.
Google calls one of the mechanisms behind this query fan-out.
One Question Can Become Many Searches
In its documentation for AI Overviews and AI Mode, Google says the systems may issue multiple related searches across subtopics and data sources while developing a response. Google separately says its models can identify additional supporting webpages while the response is being generated, allowing the AI features to show a wider and more diverse set of helpful links than a classic web search may show for the original query.
Google also explicitly cautions against treating AI Mode and AI Overviews as a single retrieval surface. The company says the two features may use different models and techniques, so the responses and links they show can vary. Source-set divergence, in other words, is not only something third-party studies have observed. Google acknowledges it in its own documentation.
Google’s newer generative AI optimization guide is equally important because it states what has not changed. Google’s generative Search features remain rooted in its core Search ranking and quality systems. Pages still need to be indexed and eligible for Search, and Google’s current guidance adds that a site must be included in Search generative AI features in Search Console to be eligible for display in those generative features. Conventional SEO fundamentals remain relevant, and Google says there is no special AI markup, AI text file, or schema required to appear in AI Overviews or AI Mode.
So this is not an argument that SEO stopped mattering. It is an argument that ranking for the original query is no longer a complete description of the source-selection problem.
SE Ranking tested 10,000 U.S. keywords in Google AI Mode, with its primary dataset collected on June 20, 2025. It reported an average exact-URL overlap of 14 percent between AI Mode citations and Google’s organic top 10, and a 21.9 percent overlap at the domain level. But the number needs context: 82.1 percent of the 9,721 queries for which the comparison was available still showed at least one overlapping URL with the organic top 10. In other words, organic rankings remained present in the source mix for most queries even though the overall sets were far from identical.
Ahrefs reported a similar pattern in March 2026 after analyzing 863,000 keyword SERPs and about four million AI Overview URLs. When Ahrefs limited the comparison to standard organic blue links, 37.1 percent of cited URLs ranked in the organic top 10 for the same query, 26.2 percent ranked between positions 11 and 100, and 36.7 percent did not rank in the organic top 100.
Ahrefs had published a much higher 76.1 percent top-10 figure in an earlier July 2025 study, but those two figures should not be treated as a clean time series. The earlier study examined the three most visible citations in each AI Overview, while the 2026 update analyzed a much larger set of citations. Ahrefs also changed its parsing methodology between studies. The newer result therefore supports source divergence, but it does not prove that Google’s overlap with organic results simply collapsed from 76 percent to 37 percent over time.
The defensible conclusion is narrower and stronger: organic ranking remains an important signal, but it is not a complete proxy for the sources from which an AI-assisted answer may be constructed.
For an evidence-dense business such as a luxury hotel, that creates a second management problem alongside visibility.
A Ranking Report Cannot Tell You Whether the Public Record Is Coherent
Luxury hospitality is unusually exposed to this problem because commercially important facts about a property are distributed among parties with different roles and different levels of authority.
The hotel controls its own website and much of its structured information. It can usually manage or influence listings, local profiles, OTA descriptions, and some review-platform information. But other important claims sit outside the hotel’s control: a Forbes Travel Guide rating, an AAA designation, a Michelin distinction, an editorial classification, a historical record, or a destination authority’s description.
Those sources are not interchangeable. The hotel is the logical authority for its current room count. Forbes Travel Guide is the authority for whether Forbes currently awards the property a particular star rating. Michelin is the authority for a Michelin distinction. An old travel article may be useful evidence for what was true at the time, but poor evidence for what is true now.
A conventional ranking report does not measure whether those claims agree with one another, whether the authoritative version is current, or whether different AI systems describe the hotel consistently. AGR’s own Luxury Hotel AI Visibility Index measures what systems return in hotel recommendation environments. That kind of measurement can expose the output. It does not, by itself, adjudicate the underlying public record.
That distinction matters. The first problem is visibility: can the page or brand be found? The second is evidence integrity: when a machine tries to understand the entity, what public evidence is available, which source is authoritative for each fact, where do those sources conflict, and how stable is the resulting representation?
The two problems overlap. They are not the same.
The Source Set Is Also Volatile
The same SE Ranking study contains another finding that complicates the picture. When the company ran the same keyword set three times on the same day, the strict three-way comparison produced only 9.2 percent exact-URL overlap across all three result sets. For 21.2 percent of the queries, there was no URL shared across all three tests.
That is the most severe way to measure the volatility, and SE Ranking itself cautioned that requiring a URL to appear in all three sets makes overlap look lower. In pairwise comparisons, exact-URL overlap ranged from 18.5 to 19 percent and domain overlap ranged from 26.6 to 27 percent. The system was still volatile, but the broader range is the more accurate representation of what the study found.
That volatility is not evidence that Knowledge Formation Optimization works. In fact, it creates a serious measurement problem for anyone claiming to optimize AI answers.
If the source set changes naturally from run to run, a single before-and-after test is weak evidence. A different citation after an intervention may reflect the intervention, normal retrieval variance, a model change, a ranking change, or simply which source happened to be selected on that run.
But volatility also exposes the weakness of optimizing for one specific citation. If a system may choose a somewhat different set of credible sources each time, the durable objective cannot reasonably be “make the model cite this one URL.” A more defensible objective is to improve the quality of the evidence environment across the sources a system might plausibly encounter.
That is where the idea of evidence convergence enters, but it has to be treated as a hypothesis, not a proven mechanism. The proposition is straightforward: when authoritative sources relevant to a fact are accurate, current, and mutually consistent, machine representations of that fact should, in theory, become more accurate and more stable.
That proposition has to be tested rather than assumed.
A Serious Counterargument: Brands May Control More of This Than It Appears
Any argument about distributed evidence has to confront a significant counterexample.
Yext analyzed 6.8 million citations from more than 1.6 million queries across Gemini, OpenAI, and Perplexity. It classified 44 percent of citations as coming from websites, 42 percent from listings, 8 percent from reviews and social, and only 6 percent from news, forums, and other sources it categorized as uncontrollable. Yext’s conclusion was that 86 percent of AI citations came from sources marketers could directly manage or strongly influence.
If that result generalized cleanly to every category and query type, much of the evidence problem would already have established owners: websites, listings management, local SEO, reviews, structured data, and data syndication.
But the scope matters. The industry breakouts Yext published were retail, finance, healthcare, and food service. Its examples center on local consumer intent, such as a nearby dentist, a local bank, or “best pourover coffee near me.” Hospitality was not one of the industry breakouts presented in the report, and comparative luxury-credential questions are materially different from local operating-information queries.
That difference is not cosmetic.
“What time does this hotel check in?” is largely a first-party or listings fact.
“Which Charleston hotels have the strongest independently verified luxury credentials?” requires evidence controlled by parties such as Forbes Travel Guide, AAA, Michelin, and other independent sources.
A listings platform can distribute a hotel’s approved address, phone number, amenities, or business description to systems within its network. It cannot edit Forbes Travel Guide’s inspection record. It cannot change Michelin’s classification. It cannot rewrite an independent publication or force an authoritative third party to adopt the hotel’s preferred version of a disputed fact.
That is the narrower white space that matters for luxury hospitality.
KFO Is Not a New Collection of Tactics
Modern SEO already covers far more than rankings and keywords. Good practitioners work on crawlability, entities, structured data, content quality, internal architecture, digital PR, authoritative mentions, local listings, and information consistency. AGR’s own GEO for Hotels framework treats retrieval optimization as necessary work, not something KFO replaces.
Listings platforms such as Yext already synchronize controlled facts. Knowledge graphs already organize entities and attributes. Digital PR already creates third-party coverage. Reputation management already corrects inaccurate public information. GEO and AEO increasingly measure citations and presence in generative systems.
AGR’s canonical KFO framework defines Knowledge Formation Optimization as the discipline of structuring, sequencing, distributing, corroborating, and correcting frameworks and entity definitions across the public information environment, then measuring whether AI systems reproduce them accurately over relevant queries and time. It does not edit model parameters or claim control over proprietary retrieval systems.
Operationally, that makes KFO less a replacement discipline than a cross-functional governance and measurement framework applied to the public evidence record of an entity.
The closest structural analogue may be Master Data Management and data governance. MDM establishes canonical internal records, determines which system is authoritative for which field, reconciles conflicts, and assigns stewardship. The difference here is that the record is public and only partly controllable. A hotel can govern its process for identifying the authoritative fact, documenting contradictions, correcting sources where legitimate, publishing evidence where missing, and monitoring machine representations, but it cannot command an independent authority to change its record.
That distinction is particularly relevant in hospitality because the evidence that establishes category, quality, and comparative position is often held by outsiders.
What KFO Actually Owns
The proposed unit of analysis is the entity’s public evidence record across sources the company controls, sources it can influence, and authoritative sources it cannot directly control.
For a hotel, that means establishing a canonical fact record; identifying which source is authoritative for each material fact; auditing contradictions, gaps, and stale claims; observing how multiple machine systems currently represent the property; documenting visible citations where available; identifying plausible supporting evidence across the public record; coordinating legitimate corrections or new evidence; and measuring whether machine representation changes afterward.
Some of that execution will be SEO. Some will be listings management. Some will be structured data. Some will be digital PR. Some will be source correction. Some may be nothing more glamorous than documenting that an authoritative third party is wrong and escalating the discrepancy with evidence.
The discipline performing the individual correction matters less than whether somebody owns the integrity of the record as a whole.
This is also why AGR has argued that measurement tools cannot, by themselves, solve the source problem they reveal. A dashboard can show that a hotel was omitted, misclassified, or cited through an intermediary. It cannot automatically determine which underlying assertion is authoritative or persuade an independent source to correct an error.
The Standard Has to Be Causal Evidence, Not Anecdotes
There is still a major gap between identifying the problem and proving that KFO improves the outcome.
A credible framework cannot say, “We corrected several sources and the AI answer looked better six weeks later.” AI systems change. Retrieval changes. Rankings change. Third-party pages update independently. Different runs of the same query can produce different citations.
A valid test therefore requires more than before-and-after screenshots. The fact set and authority hierarchy should be declared before testing begins. The same defined questions should be run repeatedly across the same systems. Model or product versions should be logged where observable. Outcomes should be numerical: factual accuracy, contradiction rate, completeness, citation presence, and answer stability.
For intervention testing, comparable control properties are necessary. One group would receive evidence remediation while matched properties would not during the same observation window. Search rankings should be measured alongside machine-answer accuracy so that an ordinary SEO improvement is not mistaken for a separate evidence-governance effect.
An even cleaner first test is cross-sectional. Take a meaningful sample of comparable luxury hotels, define a fixed set of commercially material facts, pre-declare the sources that count as authoritative for each fact, and score the quality of each property’s public evidence record. Then test the same questions across multiple AI systems, with repeated runs in the same time window, and independently score factual accuracy.
If high-integrity public records consistently correspond with more accurate machine representations, the hypothesis gains support. If high-integrity properties are represented badly just as often as low-integrity properties are represented well, evidence convergence is not the operative variable.
That result should be publishable either way.
A framework that only reports wins is marketing. A framework willing to publish null results has a chance to become measurement.
So What Does google.com/goto Have to Do With Any of This?
Not very much causally.
Google’s redirect change tells us that Google controls the mechanics of Google’s own results and can make those results more difficult for automated systems to extract. It does not prove that AI retrieval is moving away from ranking, that independent sources are becoming universally more important, or that KFO works.
The evidence for the broader argument comes from somewhere else: Google’s own description of query fan-out and broader source discovery, independent measurements showing incomplete overlap between AI citations and conventional organic rankings, and repeated tests showing substantial citation volatility.
At the same time, Google’s documentation and the vendor studies make the counterpoint equally important: SEO remains foundational, organic rankings frequently appear in AI source sets, Google Business Profile data remains important for local information, and a large share of cited sources in some categories are already controlled or influenced by brands.
Both things can be true.
A hotel should still care intensely about search visibility. It should also recognize that a ranking report cannot tell it whether the facts, credentials, classifications, relationships, and independent evidence surrounding the property form a coherent public record across the systems now answering travelers’ questions.
That is the narrower argument for KFO.
Not that search has been replaced. Not that SEO has failed. Not that every AI citation can be engineered.
The claim is that an evidence-dense business now has something additional to govern and measure: the public source environment from which machines may construct an understanding of the entity.
Whether improving that environment reliably improves machine understanding remains an empirical question.
That is exactly why it is worth measuring.
Sources
Google Search / google.com/goto rollout and Nozzle observations: Search Engine Roundtable, “Confirmed: Google Search Rolling Out google.com/goto Tracking Parameters,” August 26, 2026. Source.
Google confirmation and anti-abuse statement: Search Engine Land, “Google confirms deploying goto URL redirects to search results links,” August 26, 2026. Source.
Google AI features and query fan-out: Google Search Central, “AI features and your website.” Source.
Google guidance for generative AI features: Google Search Central, “Optimizing your website for generative AI features on Google Search.” Source.
AI Mode source overlap and volatility: SE Ranking, “AI Mode Research: Sources, Volatility, & Differences between AIO and Organic Search.” Source.
AI Overview citation overlap, 2026: Ahrefs, “Update: 38% of AI Overview Citations Pull From The Top 10,” March 2, 2026. Source.
Earlier AI Overview citation study, 2025: Ahrefs, “76% of AI Overview Citations Pull From the Top 10,” July 21, 2025. Source.
Brand-managed AI citation study: Yext, “AI Doesn’t Rank, It Cites. And 86% of Its Sources Are Brand-Managed,” October 9, 2025. Source.
