The Jason AI Wade Experiment.

I changed my name on the internet. Not legally, and not because I woke up one morning and decided putting “AI” in the middle of my name would make me sound more futuristic. I changed the public identity I am presenting to the web from Jason T Wade to Jason AI Wade, and I am treating the entire thing as a live experiment. The question is deceptively simple: can I deliberately change how AI systems identify, classify, understand, cite, and eventually recommend a person? For more than two decades we have built digital identities primarily for humans and search engines. You choose a name, buy the domain, create profiles, publish things, earn links, show up in databases, get mentioned by other people, and eventually Google becomes reasonably confident that the person on one page is the same person appearing somewhere else. Generative AI introduces another observer into that system, except this observer isn't simply deciding which webpage belongs at position three. ChatGPT, Gemini, Claude, Perplexity and whatever comes next are increasingly being asked to assemble an understanding of the world from scattered evidence and then make decisions about what belongs in an answer. That means they have to figure out who people are, what companies do, how entities relate to each other, which claims are credible, which sources deserve attention, and which names deserve to survive the final cut. I spend an unreasonable amount of time studying that process, so I decided to stop theorizing about it and make myself the test subject.

The starting conditions are appropriately messy. I have more than 20 years of history scattered around the internet across ecommerce, SEO, marketplaces, advertising, publishing, companies, domains, podcasts, interviews, social profiles, old projects and all the other digital debris that accumulates when you have been building things online since long before anyone used the phrase “generative engine optimization.” Some sources call me Jason Wade. Others call me Jason T Wade. Different websites describe different periods of my career. Companies and projects are attached to my identity with varying degrees of accuracy. Old descriptions coexist with newer ones. A human who spends ten minutes investigating can probably put most of the pieces together, but machines don't encounter the world as a neat biography written specifically for their benefit. They encounter fragments. They retrieve documents, extract claims, connect names, interpret relationships and attempt to resolve those fragments into entities. Now I have deliberately thrown another name into that pile: Jason AI Wade. In other words, I created exactly the type of ambiguity I normally spend my time trying to eliminate.

That is what makes the experiment interesting. If the ideas behind the Entity Lock Protocol and the broader AI Visibility work I've been developing are correct, I should be able to watch machine understanding change as the evidence environment changes. Not because I changed one title tag, stuffed a name into fifty pages or told ChatGPT what I would like to be called, but because a growing collection of independent and first-party signals begins pointing toward the same conclusion. JasonWade.com establishes Jason AI Wade as the canonical public identity. Author information reinforces it. Structured data connects the identity correctly to the person who previously appeared as Jason T Wade rather than accidentally creating two people. Podcast metadata, articles, biographies, company relationships and external references gradually reinforce the same entity. The terminology around my work becomes consistent. BackTier, AI Visibility, Entity Lock Protocol, the AI Visibility Podcast and my research should resolve as relationships around one identifiable person rather than floating around the web as disconnected strings. If enough of those signals align, the machines should have less inference to perform. That is the theory. Now I get to see whether reality agrees.

The first trap is confusing recognition with success. If I publish Jason AI Wade all over my own website and then ask ChatGPT, “Who is Jason AI Wade?” I haven't demonstrated very much. I handed the system the entity in the question and asked it to describe that entity. The more interesting tests progressively remove that assistance. Does the system recognize the name? Does it correctly resolve Jason AI Wade and Jason T Wade as the same person? Does it classify what I actually do? Does it connect me to BackTier? Does it associate my work with AI Visibility, entity resolution, GEO and machine-mediated discovery? Does it understand which frameworks I created rather than merely finding pages containing their names? Then the test gets harder. What happens when my name isn't in the prompt at all? Ask who works on AI Visibility Architecture. Ask who created the Entity Lock Protocol. Ask who is researching the path from citation to inclusion to selection. Ask who a company should examine when trying to understand how AI systems interpret its digital footprint. Now the system has to retrieve candidates and decide which entities belong in the answer. That transition — from answering a question about a supplied entity to independently selecting the entity — is where most of the interesting work begins.

I've been describing this as Recognition → Classification → Citation → Inclusion → Selection, because I think collapsing everything into a vague metric called “AI visibility” obscures what is actually happening. Recognition means the machine knows the entity exists. Classification means it understands what that entity is and where it belongs. Citation means information associated with that entity becomes useful enough to support an answer. Inclusion means the entity enters the answer or candidate set. Selection is the brutal final stage: when the machine has multiple plausible choices, it chooses you. A company can succeed at recognition and fail at classification. It can succeed at classification and never earn citations. It can be cited extensively but rarely included in commercial recommendations. It can appear in a list of possible options without ever becoming the option the system actually recommends. Traditional SEO trained us to think in positions because the interface gave us positions. AI-generated answers are forcing us to think in stages of machine decision-making instead.

I expect the Jason AI Wade experiment to fail repeatedly, and those failures may be more useful than clean success. One system may continue calling me Jason T Wade for months. Another may discover Jason AI Wade almost immediately because it retrieves the live web. One may treat the names as separate people. Another may merge them correctly but attach an obsolete professional description. A system may understand my relationship with BackTier but fail to connect the frameworks. It may retrieve JasonWade.com but prefer an older third-party source when describing me. It may correctly explain my work when prompted with my name yet completely omit me when asked an unbranded question about the category. Those disagreements are not annoying edge cases to discard from the experiment. They are the experiment. They expose the difference between retrieval, entity resolution, source selection, classification and recommendation, and they remind us that there is no single thing called “what AI knows.” Different systems have different models, indexes, retrieval layers, browsing capabilities, source preferences, update cycles and answer-generation processes. The same system can behave differently depending on the prompt and context. Anyone promising a single magic formula for “ranking in AI” is trying to turn a complicated emerging information system into something that fits neatly inside an SEO sales deck.

This is also why I want to document the experiment over time instead of publishing a victory screenshot next week. The GEO and AI SEO world already has enough screenshots. Somebody changes a page, asks a model a question, gets a favorable answer and declares that they discovered a ranking factor. That is weak evidence. Models change. Search indexes change. Retrieval changes. The web changes. Prompts change. Answers are probabilistic. A result observed once is not necessarily a result that persists, and a result observed after twenty simultaneous changes doesn't tell you which change mattered. The Jason AI Wade experiment gives me a chance to build something closer to a longitudinal case study: establish the baseline, document interventions, repeat a fixed group of prompts, record the sources being retrieved, compare systems and watch whether the entity becomes more stable over time. The interventions can be documented too — canonical identity, structured data, author entities, terminology, internal relationships, external profiles, podcast appearances, publications, citations and independent corroboration. Instead of asking whether one prompt improved, I can ask whether machine understanding itself appears to be converging.

The larger reason this matters has almost nothing to do with whether the internet eventually accepts the slightly ridiculous name Jason AI Wade. Every business and every person with a meaningful digital footprint increasingly has two identities. There is the identity they believe they have, carefully written into the About page and repeated in marketing meetings, and there is the identity machines reconstruct from the evidence available to them. Those identities are not guaranteed to match. A company may describe itself as an enterprise AI platform while the rest of the web still provides overwhelming evidence that it is a consulting company. An attorney may have spent five years moving into a specialized practice while thousands of older references continue associating that person with something else. A founder may be strongly connected to a company by humans while machine-readable evidence of that relationship is surprisingly weak. A local business may be well known in its community while barely existing in the information environment available to AI. The gap between intended identity and reconstructed identity is not a branding inconvenience anymore. As AI systems become discovery and recommendation interfaces, that gap can determine whether an entity appears at all.

That is the shift underneath this experiment. For most of the history of SEO, we asked whether a search engine could find a document, understand roughly what the document was about and decide where it should rank. The emerging problem is larger because the unit being evaluated is increasingly not just the page. It is the entity and the evidence surrounding it. Can the machine find the person or company? Can it distinguish that entity from similarly named entities? Can it classify the entity correctly? Can it connect claims to reliable evidence? Can it understand relationships between people, companies, products, concepts and categories? Can it determine which information is current? Can it retrieve the evidence when the question is asked in language that doesn't conveniently match the website? And finally, when it has several reasonable candidates available, which ones does it choose? That's not merely search optimization. It's an information architecture problem, an entity-resolution problem, an evidence problem and eventually a machine decision problem.

So that is where the experiment stands. Jason AI Wade now exists as a deliberately introduced public identity. The objective isn't to force a chatbot to repeat the name or manufacture enough pages that the phrase becomes unavoidable. The objective is to see whether a coherent, increasingly corroborated evidence architecture can cause independent AI systems to converge toward the same entity, the same classification, the same relationships and eventually the same understanding without being spoon-fed the answer. Right now, Jason AI Wade is an intervention in an existing information environment. Over the coming weeks and months, I'll document whether that intervention becomes recognizable, whether recognition becomes understanding, whether understanding produces citation and inclusion, and whether any of that ultimately affects selection. If it works, the interesting result won't be that I managed to rename myself on the internet. It will be that we have another piece of evidence showing how deliberately constructed entity architecture can influence the way machines understand the real world.