Over the last few months, I noticed Squarespace began emailing me to see how TS Blackwell-Hart appears in AI search. I decided to take them up on the offer—but I wasn't particularly interested in a visibility score. I wanted to know what happened after the system found me.
Squarespace provides an AI Visibility tool that tests how a brand appears in AI-generated responses. Squarespace supplies an initial set of ten prompts: five branded and five non-branded. After experimenting with those prompts, I refined the set and used the two additional prompt spaces Squarespace provides to expand the non-branded testing. The final twelve-prompt set was designed to examine something more specific: whether AI systems could recognize TS Blackwell-Hart and the Blackwell-Hart Methodology™ when they were named directly, and whether they would independently connect either one to relevant questions when they were not.
The final test used twelve prompts. Five were branded and explicitly referenced TS Blackwell-Hart or BHM™. Seven were non-branded and asked broader questions about AI visibility, authority, semantic stability, semantic drift, innovation validation, and accurate representation in AI-driven search.
I first ran that customized set on June 14, 2026. On August 21, I ran the exact same twelve prompts again through both ChatGPT and Gemini.
The intention was not to determine which system was "better." It was to observe how differently two AI systems interpreted the same entity when given the same questions.
The first five prompts produced a fairly strong result.
When TS Blackwell-Hart or BHM™ was named, both systems generally recognized the entity, although one prompt exposed a significant difference in entity resolution between the two models. Overall, BHM™ was connected with AI-assisted discovery, authority infrastructure, entity interpretation, consistent representation, and the structuring of information across digital environments.
The systems also demonstrated that the TS Blackwell-Hart entity does not exist solely on its own website. The Inventors Association of Australia (Victoria) appeared repeatedly as an external source associated with TS and BHM™. That relationship is legitimate: I am a committee member and site sponsor, I provided the organization with a BHM™ license, and the IAA-Vic site maintains an inventor page for TS Blackwell-Hart.
That part of the experiment was encouraging.
It also revealed something more complicated.
ChatGPT and Gemini did not always construct the same representation from the same information. When asked who T.S. Blackwell-Hart is and what his professional background is, Gemini successfully connected the entity to the IAA-Vic relationship and produced a substantial professional profile. ChatGPT found my genuine Medium account under the TS Blackwell-Hart identity, but it also introduced an unrelated LinkedIn profile and an unrelated historical Blackwell as possible matches.
The difference matters.
Finding a name is not the same as resolving an entity. A system can encounter a genuine signal and still fail to connect that signal to the correct person.
The branded prompts also showed another form of variation. The models sometimes selected different supporting sources and reconstructed different portions of the surrounding information environment. One model might rely primarily on TS Blackwell-Hart, while another incorporated IAA-Vic or The Hartful Company into its answer.
Again, this was not necessarily a failure. It was evidence that the same entity can be reconstructed through different paths.
Then came the non-branded prompts.
This was where the experiment became much more interesting.
None of the seven non-branded prompts produced a BHM™ mention in either model.
When asked how businesses can establish authority and consistency across AI search platforms, the systems instead returned established search and AI-visibility companies and resources. When asked how organizations can maintain consistent brand interpretation across AI platforms, they moved into the existing world of AI brand governance and brand-management systems.
When asked how independent creators can validate and document original innovations, the models returned patent, provenance, timestamping, and intellectual-property services.
When asked about semantic stability and semantic drift, they returned established technical and research sources.
And when asked what frameworks can improve AI visibility and representation, they surfaced other frameworks and providers already associated with that category.
The important point is not that BHM™ failed to appear.
The important point is what the systems did instead.
They already have established conceptual neighborhoods for these questions. When the name "Blackwell-Hart Methodology™" is placed in front of them, they can recognize it. When the name is removed and the question becomes generic, they return the entities and sources that already have stronger associations with the concept being requested.
That creates a distinction I think is easy to miss when we talk about AI visibility.
An entity can be recognized without being selected.
A system can know who you are when your name is presented to it without independently considering you when someone asks a relevant question without your name attached.
That is a very different problem from simply being "visible."
It also changes the way I think about AI search measurement.
A percentage showing whether a brand was mentioned is useful, but it doesn't tell the whole story. I want to know whether the system resolved the right entity, how it described that entity, which sources it connected to it, which category it associated with it, and whether it considered the entity when the question was asked without the entity's name.
Those are different measurements.
In this test, branded recognition was strong. Cross-domain relationships were visible. Entity resolution varied between models. Canonical terminology was not always reproduced exactly. And category-level inclusion in non-branded questions was not yet established.
That is a much more useful description of the result than simply saying that TS Blackwell-Hart "appears in AI search."
It also reinforces something I have been saying about BHM™ from the beginning.
The challenge is not simply being found.
It is being understood.
And even that may not be enough.
A system can understand what an entity is and still choose someone else when answering the question.
That is the next problem.
The experiment is ongoing. The twelve prompts are now a fixed test set, giving me a consistent benchmark for future comparisons rather than a constantly changing set of questions.
I am less interested in finding a magic visibility score than I am in watching how the representation changes over time.
Because ultimately, the question is not:
"Did AI find me?"
It is:
"When someone asks the right question, does AI know that I belong in the answer?"
