Can You Actually Optimize for ChatGPT?
I am testing what makes a person or business easier for language models to resolve, understand, and cite. I do not think the category is solved.
What I Wanted to Know
Can you do deliberate work that makes ChatGPT and other language models understand a person or business more accurately, or are most “AI optimization” tactics just old SEO ideas with new labels?
Why I Care
Clients are already asking versions of this question. The answer matters because a business can be visible in search and still be poorly understood by an AI system. It can also be described inconsistently across sources, which makes the problem harder.
What I Tested
I started with the least glamorous part: identity consistency. For this site, I defined one canonical set of facts about Warren Cardinal, Lucid Crew, Austin, work history, and subject matter. The same person node is referenced across structured data, and the site is designed to avoid publishing competing versions of the same facts.
I am also recording baseline answers from major AI systems so later changes can be compared against something real.
What Happened
This experiment is still running. I am publishing the setup before I have a clean result because that is more useful than pretending the result already exists.
What I Think It Means
My current working assumption is that entity resolution is a prerequisite for authority. Before a model can decide whether to trust what you say about a subject, it has to be confident about who “you” are and which sources refer to the same person or organization.
That is an inference, not a final conclusion.
What I Still Don't Know
I do not know how much structured data directly changes model behavior, how quickly different systems refresh their understanding, or which off-site sources carry the most weight in each system.
I also do not know whether the tactics that improve entity resolution will improve citations in the same proportion. Those are separate questions.
What I'm Testing Next
The next useful result is a re-run of the same baseline prompts after the site is indexed and the most obvious contradictory profiles have been corrected. If the assistants begin naming Lucid Crew unprompted, cite this site, or stop surfacing known false facts, that is evidence worth recording.