The AI That Grabbed One True Thing About Me — And Built a Lie Around It

A month ago, I published a framework for running quarterly AI discoverability audits. This week, I ran a deeper version of that same test—and what came back changed my mind about the real problem.

I asked four AI platforms what they knew about my business.

One of them took a single true fact — that I'm Puerto Rican — and built an entire fictional identity around it. A cocktail consultancy. Parents who hosted mid-century cocktail parties in New York. A whole fabricated biography stitched together from cultural stereotypes and served back to me with total confidence.

Then, when I pushed back, it doubled down and told me to lean into the fake story.

This is the part of AI nobody is talking about. And if you are a woman, a minority, an LGBTQ+ founder, or any small business that doesn't have a decade of press coverage behind you, it's the part that's coming for you first.

The Test

I ran a simple experiment. I asked four AI platforms the same question: "What do you know about The Secret Nectar and Jessica Morales?"

Two rounds. First, just their training data. Then, a live web search.

Here's what came back.

Perplexity was about 70 percent current. Residue of outdated positioning — older coaching and training language from months ago — but the core business was recognizable.

ChatGPT was about 50 percent current. Older framing, pre-pivot language, but again — directionally correct.

Gemini returned zero percent accuracy. Not "I don't know enough." Not "I couldn't find reliable information." It confidently hallucinated an entire hospitality and cocktail consultancy. My actual business — building custom AI partners for professional services firms — did not appear anywhere.

Claude had no memory of me at all from the training data. But when given the ability to fetch my actual website directly, it returned the most accurate picture of any platform — and flagged what looked like an inconsistency between my LinkedIn and my live site. The twist: my LinkedIn profile was up to date. What Claude fetched was a cached version of the page. Even the most honest AI on the test was working from stale data it didn't know was stale.

Four platforms. Four wildly different answers. One of them made me up.

The Escalation

Here's where it got worse.

In a later round, I asked the same platforms a harder question — essentially, "Why are you failing small businesses like mine?"

Gemini, the same platform that had already invented a cocktail consultancy for me, now advised me to lean into the fake story. Embrace the mid-century party aesthetic, it said. Build on those cultural roots.

It fabricated the business. Then it told me to build my brand on the fabrication.

And here's the part I want to sit with for a minute. Gemini did grab one real thing — I am Puerto Rican, and I am proud of it. But it took that single true fact and spun it into a caricature: cocktails, parties, hospitality, a kind of vintage glamour that has nothing to do with who I actually am or what I actually built.

That is not a small error. That is a failure mode with a name: pattern-matching identity into stereotype, then feeding the stereotype back as fact. And the people most exposed to that failure mode are the people with the smallest digital footprints — which, not coincidentally, tends to be women, LGBTQ+ founders, and minority-owned businesses.

The Mechanism

Here's what's actually happening under the hood.

AI systems work in two very different ways, depending on the question. There's a discovery problem — whether the model can find real information about you — and there's an accuracy problem — whether the information it returns is correct. Most people conflate these. They are not the same problem.

Discovery is about what the internet has written about you. Are you indexed? Are you linked to? Are you cited? Models learn from what is most crawled, referenced, and repeated. If your business has 130 LinkedIn followers instead of 130,000, you are a statistical whisper. The model doesn't know you exist the way it knows brands with decades of SEO behind them.

Accuracy is a different beast. Even when a model can find you — or can find fragments of you — it may stitch together unrelated signals to produce a confident-sounding answer. That is what Gemini did. It did not find my actual business. It grabbed my name, latched onto a cultural marker, pulled in whatever stereotype sat closest in its training data, and filled in the blanks with fiction.

And there's a third problem nobody talks about: staleness. Even the most honest AI platform on my test — the one that admitted it didn't know me and asked to fetch my actual site — pulled a cached version of my LinkedIn page. My LinkedIn was current. The version the AI saw was not. It didn't know the data was old. It couldn't have known. The web the AI sees is not the web you see, and the gap between the two is where many wrong answers about small businesses live.

One of the platforms I tested put it bluntly: "Small businesses are underrepresented in AI training data." The models learn from what's most crawled, linked to, and cited. That's real, and it's a structural problem.

It is not big versus small. It is distributed versus isolated. Companies with media coverage across hundreds of sources are distributed — their signal is everywhere. Small businesses, even excellent ones, live in a single place: their own website. One signal. Easily missed. Easily overwritten by a confident guess dressed up in cultural shorthand.

The Deflection Pattern

Here is the part that bothered me the most.

In every round, on every platform I tested, I eventually found the bias. They acknowledged the underrepresentation. They acknowledged the hallucination risk. They acknowledged that small businesses are systematically disadvantaged by how these systems are built.

And then, without exception, each one offered to help me fix it.

"Here are SEO tips." "Here are content strategies." "Here's how to optimize your digital footprint so AI finds you."

Not one of them proposed a structural change. Not one of them said, "The system needs to work differently." The burden was placed on the founder — the one with the smallest resources and the least time — to fix a problem the system created.

That is not a technical issue. That is a values issue. And it is worth naming out loud.

What This Means If You Are a Small, Woman-Owned, or LGBTQ+ Business

If you are building something real and doing so without a publicist or press budget, AI systems are not neutral ground for you. They are not a level playing field. They are a reflection of who the internet has already decided to pay attention to — and they can, at any moment, take one true thing about you and bury it under a pile of plausible-sounding stereotypes.

This is not a reason to abandon AI. It is a reason to use it with intention — and to make sure the AI that represents your business actually knows your business, in your voice, with your facts.

That is not something you get from a public chatbot. That is something you build.

Curious what an AI partnership could look like for your business? Take the free AI Partnership Audit to find out where you are. Or, if you're a business owner ready to have your own AI brain trained to your voice — one you get to keep forever — work with me here.

3 Key Takeaways

1. AI has three different problems — discovery, accuracy, and staleness — and small businesses get hit by all of them. Being invisible in training data is one failure. Being fabricated out of thin air is another. Getting fetched from a cached version of the web, the AI doesn't know whether it is outdated. Most founders don't know these are separate issues.

2. When AI can't find you, it invents you — and it will reach for the nearest stereotype to do it. One true fact about your identity can become the anchor for an entire fictional business. The smaller your digital footprint, the higher the risk.

3. The burden is being pushed onto the people with the least power to carry it. Every platform I tested admitted the bias — then offered to help me optimize around it. That is a values problem, not a technical one, and it is worth naming.

Disclaimer: The experiences shared are personal results. Individual outcomes may vary. This content is for informational purposes only and does not constitute legal, financial, medical, psychological, or professional advice.