Creating a new website
One of the first considerations is the name, which can be descriptive or unique. Both online and offline, it is important that the name not be confused with those of other businesses or organizations. If the organization is to engage in commerce, it may need to register its name. This process generally involves checking county records to make certain the name is not already in use and to avoid confusion, although other steps may be required. A trademark can provide additional protection against confusion with other names across the US.
Owning Your Brand Intent & Site Architecture
Given human behavior, when people enter a query to find information about a brand, the brand's canonical page is expected to appear in the results. Knowledge panels were created for business entities, and search evolved to better address navigational search intent. In most cases, little needs to be done to optimize for brand searches. Visitors generally expect other pages on the site to link to the home page or canonical entity page. The site structure generally provides all the optimization needed. This basic structure is also the foundation of many hub or SEO silo structures: for example, five or more pages linked from a category page, with those pages also linking back to the category hub.
Exact Match Domains (EMD)
Over the years, many people practicing SEO have used EMDs as a shortcut to ranking for a search term by taking advantage of the perceived ranking weight of a matching domain name. The shift toward entity-based search has observationally reduced the effectiveness of EMDs by requiring them to compete as canonical entity identities rather than benefiting primarily from navigational search intent. Essentially, this is an exploration of the limits of search engines' ability to perform pseudo-subjective reasoning.
Pseudo-subjective reasoning is used here to describe decision-making processes that are sufficiently complex to simulate human judgment. While deterministic mathematical models cannot make truly random or subjective choices, high-dimensional rules and contextual evaluation allow them to enter the uncanny valley, appearing capable of judging where a generic keyword query ends and a distinct brand entity begins. Context from natural language processing is used to determine whether Apple refers to a company or a fruit.
The Search-friendly AI-aware Site/Domain Name page goes into more detail about site/domain names. It should now be considered part of Entity-Based SEO, having been published when the shift was becoming clearly observable.
Building a brand entity
For a local business seeking local search visibility, its address and phone number are essential. Many legacy SEO practitioners have suggested that the owner's name and photo are necessary to establish trust, but the supporting evidence is questionable. I would suggest that an image of the business provides clear value, not as a direct ranking factor, but as an emergent factor of Navboost. For example, someone looking for a local bike shop may continue searching if a business listing does not include a photo, potentially signaling that their intent has not been satisfied. A consultant, on the other hand, may benefit from having their own image on the site. Pseudo-subjective reasoning is based on signals derived from search intent, not the hard-coded algorithmic rules often proposed by legacy SEO.
Authorship byline: Authorship is semantically entangled with E-E-A-T, which involves subjective human evaluation of website content and quality. The concepts underlying these evaluations have also informed AI systems. In practical terms, the author is an entity, often without a website of their own. The existence of an author byline is no more inherently a ranking factor than the existence of a website. Apart from internal links that establish a hub structure or external links from the author's website, authorship does not independently change a site's ranking. It may provide additional visibility when someone searches for the author's name or a topic associated with the author's entity. However, authorship is not a magic bullet for SEO and is entirely optional. It adds particular value when the author is a consultant, influencer, or celebrity, but most blue-collar businesses do not need an authorship byline.
Sparse information is not your friend. Natural language processing is used to interpret website content that AI systems may use as grounding data to ensure their responses are based on current information. These systems are often described as narrative engines that predict the next word, but this is only one of many capabilities. Narrative generation alone cannot reliably establish correct assertions about a brand. Without grounding, probabilistic text generation produces plausible narratives rather than empirically supported facts. Nonfictional information must be clearly and explicitly published. Forcing AI to guess about a brand is problematic at best and closely related to publishing unreviewed AI content.
While many new websites try to put everything on their home page, there is a practical reason for this: the home page is often the first page created. However, everything about the business does not need to be forced onto the home page. The information can and should be published on relevant category pages. When an AI system finds that it has sparse information to ground its response, it looks for additional information, typically using a fan-out query to find published sources. The exact processes are, of course, subject to change.
For navigational query intent, the site's About page can be seen as the canonical source of information and is typically relevant to queries like, "Who is [brand]?" Category pages can be optimized with answers to queries like, "What services does [brand] offer?" By establishing these pages as authoritative first-party sources of information about the brand, the business provides AI systems with clearly attributable assertions that can be evaluated against other sources. This creates a potential moat for reputation management.
External mentions and links
As stated above, a website's existence is not a ranking factor. The general understanding is that search engines discover websites by following links pointing to them. While this may be incomplete, it is essentially correct. Websites and sitemaps can also be submitted directly to search engines. The legacy SEO practice of using nofollow links to shape search results is based on an incorrect understanding of how they work. Since domains can acquire links through registration or other normal processes, steps need to be taken when staging a website to prevent it from being discovered prematurely.
It should also be noted that link graphs such as PageRank involve diminishing returns, meaning that 20 links do not necessarily double the ranking value of 10 links. On a base-10 logarithmic scale, the orders of magnitude are: 1–9 are 1, 10–99 are 2, and 100–999 are 3. People who build links are not necessarily being deceptive, but some may overstate the ranking benefits of link acquisition. A site does need links to compete for rankings, and some topics require considerably more than others. The link graph itself can be understood through a Markov chain model, which represents the probability of a user navigating from one page to another. An in-depth look at PageRank SEO Strategy: The Markov Chain Model was published August 28, 2026 by The SEO Heretic.
The Search-Everywhere Framework: Driving Real Referral Traffic
Before throwing the baby out with the bathwater, consider how linking was supposed to work before sites with zero traffic were created primarily to simplify link building. The places where potential clients are already visiting, and where they are likely to follow a link to the brand's website, are where the brand needs to be mentioned or linked. These pages can generate referral traffic regardless of nofollow attributes or search-engine rankings. They include social media platforms and user-generated content sites.
Assertion confidence or trust? I dislike using the term trust because of its semantic inferences. AI systems need to interpret the information they encounter on the internet, but much of that content is ambiguous to some degree. A phone number or address on a web page may belong to the business, another business, or something else entirely. When the same number and address appear on another site in a context that identifies the brand, the association becomes less ambiguous.
Likewise, mistakes can be made in schema. Having the same information in the visible page content provides another representation against which the schema assertion can be evaluated, increasing confidence when the information is consistent. Trust suggests that if Amazon.com has a typo, the site itself is not trusted, rather than the system simply having less confidence in that particular piece of information. Factual AI responses are more likely to rely on statements that can be interpreted with confidence. Ambiguous content can contribute to hallucinations when information is misinterpreted or attributed to the wrong entity.
The term pseudo-subjective trust can apply to a page. If the content on the page contains errors and critical omissions, that can subjectively indicate that the author does not know what they are talking about. However, the evaluation is based on the system's interpretation of the page, which may overlook other details. The practical implementation is to avoid using the page as a source or to have less confidence in it.
However, subjects are rarely black and white. Confidence that an assertion has been made does not require treating it as the only valid assertion, while an all-or-nothing interpretation of trust can force a choice that may be wrong. Two entities can look at a coin and discuss different sides of the same coin. A confidence score allows an AI system to retain different assertions from different authors, publishers, or entities without necessarily treating them as mutually exclusive.
In everyday language, we typically describe the sky as blue. Scientific and photographic descriptions may use more precise color distinctions, including cyan, depending on the conditions and context. A nonfictional AI chat can use the user's intent and the context window to determine which description is relevant. This is an example of pseudo-subjective reasoning applied to the interpretation of an assertion.
Unique Content or Gain of Knowledge
These terms may provoke an emotional response, similar to the anxiety associated with taking a test. When that occurs, it typically stems from a misunderstanding of what they mean in an information retrieval context. A new business establishing a physical location creates a potentially unique entity association, and that specific location data represents a gain of knowledge. The evidence supporting an AI citation can rely on this unique gain of knowledge, often obtained through real-time grounding data. In practical terms, unique information gain refers to information that is not already embedded within an AI model's parametric training weights.
This does not imply that pre-existing training data can never generate a citation. Years ago, I noted that an 80286 processor had a 16 MB memory limit. That statement was correct in terms of the processor's directly addressable physical memory. However, third-party memory expansion and memory-management solutions made additional memory usable through techniques such as bank switching. The practical limitation had changed, or the original assertion needed a more precise qualifier: an 80286 can directly address 16 MB of physical memory. This illustrates how new information can qualify or supersede an existing assertion.
At a foundational level, LLM training involves compressing information into model parameters. A deeper analysis of these mechanics is available in LLMs, data compression, the effect on AEO and SEO. Source URLs are not generally retained as reliable, retrievable references within compressed model weights. This is one of the limitations that retrieval architectures help address. Nevertheless, brand entities can still be represented and mentioned, while retrieval can provide the source information needed to support citations and hyperlinks in generated answers.
Returning to the gain of knowledge, which is not a test, a press release announcing that a brand has developed a 64 MB memory card for the 80286 processor, along with a Windows driver, is the kind of information that can contribute to both SEO and AEO. When first published, it would provide new grounding data. If subsequently incorporated into model training data, that information could become part of the model's existing knowledge. This is not parasitic SEO; it is the flow of information, which rarely consists of simply copying and pasting existing content.
In marketing, there is an expression: Vive la différence ! The value lies in what makes something different, and that difference can represent a gain of knowledge. The effort is explaining the business entity, product, service, or concept accurately so it is understood, not the amount of time that was used to create the content.
Published
by Wayne Smith – The SEO Heretic
Wayne Smith has worked in online marketing, search, and web development for several decades. The role of an SEO heretic is to question dogma and tradition when empirical evidence shows it is incorrect.