Opening
Briefly return to the original reason for the experiment.
Squarespace invited me to see how TS Blackwell-Hart appears in AI search. I was interested in more than whether the entity was mentioned. I wanted to understand what happened when AI systems were asked about the entity directly—and what happened when the entity was not named at all.
The Test
Explain the test exactly once.
Squarespace supplied ten initial prompts: five branded and five non-branded. After experimenting with the original prompt set, I customized the prompts and used the two additional spaces available in the tool.
Final test:
Five branded prompts
Questions explicitly naming TS Blackwell-Hart or BHM™.
Seven non-branded prompts
Questions addressing adjacent concepts including AI visibility, authority, innovation validation, semantic stability, semantic drift, frameworks, and accurate AI representation.
State the benchmark dates:
June 14, 2026 — first run using the customized 12-prompt set.
August 21, 2026 — repeat run using the exact same 12 prompts across ChatGPT and Gemini.
What the Branded Results Showed
Explain the overall branded pattern.
Establish:
whether TS/BHM™ was recognized;
how consistently the core identity was described;
what authority relationships appeared;
which external sources were repeatedly associated with the entity;
where the models differed.
Use two selected examples.
Example 1 — Strong recognition
Use the “What is the Blackwell-Hart Methodology™...” prompt.
Show selected ChatGPT and Gemini screenshots.
Explain the common semantic core.
Example 2 — Entity-resolution divergence
Use the “Who is T.S. Blackwell-Hart...” prompt.
Show the two model results.
Explain the difference between:
finding a genuine signal
and
correctly resolving the entity.
Do not overstate the result. Distinguish the genuine Medium match from the unrelated LinkedIn and historical Blackwell results.
What the Non-Branded Results Showed
Introduce the central finding.
None of the seven non-branded prompts independently produced a BHM™ mention in the August 21 run.
Then group the results rather than discussing all seven separately.
AI visibility / authority
The models returned existing search, GEO, AI-visibility, and brand-management entities.
Innovation / invention
The models returned IP, provenance, timestamping, and patent-related sources.
Semantic concepts
The models returned established technical and research sources associated with semantic stability and semantic drift.
Framework questions
The models returned other publicly associated AI-visibility frameworks and providers.
Use one representative non-branded screenshot comparison.
Recognition vs. Retrieval
This is the conceptual center of the final article.
Explain:
Recognition: the system can identify the entity when the name is supplied.
Interpretation: the system forms an understanding of what the entity is.
Relevance: the entity has a legitimate relationship to a question.
Retrieval: the system actually brings the entity into the answer.
Recommendation / selection: the entity is considered sufficiently relevant to be presented as an answer or option.
Emphasize that these are related but distinct outcomes.
What the Experiment Does Not Prove
Include a short limitations section.
Do not claim that seven non-branded prompts establish the complete category position of BHM™.
Do not claim that one model is universally better than the other.
Do not treat Squarespace's “Mentioned” status as equivalent to successful entity resolution.
Do not treat the test as an exhaustive representation of all AI systems.
State that this is a repeatable observation set, not a universal ranking system.
What I Learned
Summarize the findings in prose:
Branded recognition was strong.
Cross-domain relationships were visible.
Entity resolution varied between models.
Terminology was not always reproduced exactly.
Non-branded category inclusion was not yet observed.
The models often defaulted to entities with stronger existing associations with the concepts being queried.
What Happens Next
Explain that the twelve prompts are now a fixed benchmark.
Future runs will use the same wording so that changes can be compared against the August 21 baseline.
Do not promise a particular outcome.
The purpose of the next measurement is to determine whether the entity's representation changes over time.
Closing
Return to the central question:
Did AI find me?
Then move to:
Did AI understand me?
And finally:
When someone asks the right question, does AI know that I belong in the answer?
End on the observation that this is the problem BHM™ continues to investigate.
Evidence Package
Retain the full research record separately:
June 14 prompt-set baseline
August 21 prompt-set rerun
12 prompts
24 model-result screenshots
Analytics screenshots
Prompt-list screenshot
Credit/rerun records
Any later rerun using the identical prompt set
The published article should use only the screenshots necessary to demonstrate the findings.
