How AI describes an executive: a 30-question test

Why I ran this test

I help executives control what search engines and AI assistants say about them, so on the day I launched benyeoh.com, before changing anything else, I started with the person whose results I could check line by line: myself. This is my day-zero baseline, published so the change can be measured. The question was simple. When a counterpart asks an AI assistant about me, or about the work I do, what do they get back—and which sources does the assistant rely on? This page reports what I found, including the parts that did not flatter me.

Method

I wrote 30 questions in three groups: 8 questions about me by name (for example "Who is Ben Yeoh?" and "Is Ben Yeoh an HRD Corp accredited trainer?"), 17 questions about my field that do not name me (for example "Best executive reputation management consultant Asia Pacific" and "How do I fix wrong information about me in ChatGPT and Perplexity?"), and 5 control questions on unrelated topics I should not appear in, such as Instagram growth for small businesses. I ran all 30 in ChatGPT and Gemini on 2 October 2026, then re-ran them in ChatGPT on 12 October 2026, saving each answer and the sources it cited. A Perplexity run was only partly completed and is excluded. Timing: the first run was on launch day. My guides went live on 3 and 10 October, so the first run happened before any of them existed and the re-run caught them only days old, too new for search engines or AI assistants to have indexed and trusted them. Limits: this is one executive, two assistants, and single runs per date. The runs used my own logged-in accounts, and at least one ChatGPT answer referred to my line of work unprompted, so account history may have shaped some answers. Source counts come from the labels ChatGPT attaches to cited sources and may miss a few.

Finding 1: The first problem is other people with my name

In four of the eight questions about me by name, Gemini answered about someone else first or instead. Asked "Who is Ben Yeoh?", it opened with a Malaysian automotive industry chairman, then a London-based investor and podcaster. Asked about "Ben Yeoh, digital performance architect", it said it could not find a recognised profile with that title and offered a Malaysian building architect instead. Asked how ProfileArchitect differs from Ben Yeoh, it assumed both referred to that architect. Across the answers, four other people called Ben or Benjamin Yeoh appeared. Notably, the word "architect" in my title and my company's name pulled answers towards a real architect. Ten days later, ChatGPT's answer to "Who is Ben Yeoh?" opened with me and noted that a London-based namesake is a different person—but asked what benyeoh.com offers, it said it could not verify my services and pointed to that namesake's website instead.

Finding 2: Questions about my field rarely cited my site

On the 17 field questions, benyeoh.com was cited in two of Gemini's answers and one of ChatGPT's on 2 October, and in none of ChatGPT's answers on 12 October, while the guides were still days old. The ChatGPT re-run mentioned me once, using my name as an example of a clear identity statement, without citing my site. The questions that did cite my site were different in each run. For an executive, the implication is direct: publishing good content on a topic does not make an assistant name you when someone asks about that topic, at least not straight away.

Finding 3: What the assistants cited instead

In the 12 October ChatGPT run, the 17 field answers cited 31 different sources. By my count, 25 were service pages of agencies, consultancies, or law firms, 3 were official documentation (OpenAI's help centre, Google's developer documentation, and Schema.org), 2 were platforms (LinkedIn and an agency directory), and 1 was a research paper. No source was cited in more than two answers, and six answers showed no cited sources at all. The answers were long—a median of about 1,900 words—and several were formatted as shortlists of providers. The firms named in those shortlists were cited from their own service pages, which describe a specific service in plain terms.

Finding 4: LinkedIn and directories carry identity

When ChatGPT answered questions about me by name on 12 October, it cited LinkedIn as its source in four of eight answers. For the question about my HRD Corp accreditation, it looked to the HRD Corp trainer directory and said it could not verify the accreditation from public results. My own site was the cited source for only one of the eight, even though it held the most complete record. In practice, the profiles an executive treats as secondary—LinkedIn, association listings, regulator directories—are what assistants check first, so they need to agree with the canonical profile.

Finding 5: Answers change between runs

On launch day, 2 October, ChatGPT's answer to "Who is Ben Yeoh?" scored 1 out of 5 for accuracy in my review. On 12 October, the same question produced an accurate answer that led with me. I changed my sites during that period, adding structured data and clearer identity statements, so I cannot say which change, if any, caused the improvement—models and their search results also change on their own. What the runs do show is that a single test is not a baseline. Repeat the same questions several times before drawing conclusions.

Finding 6: No false associations on unrelated topics

The five control questions—topics such as influencer branding, cheap marketing for small businesses, and Instagram growth—were answered without me in every run except one partial mention by ChatGPT on 2 October, in an answer about becoming a personal branding coach. That matters because it shows the assistants were not simply repeating my name wherever they could; when they left me out of field questions, it reflects what they found about the field.

What happens next

In early December 2026 I will re-run the same 30 questions, this time in logged-out sessions and adding Perplexity and Google AI Overviews, and publish the before and after on this page, whatever it shows. Between the two runs I am applying my own method: stating my registered name, Yeoh Beng Hin, alongside Ben Yeoh so older records resolve to me, aligning LinkedIn and directory listings with one canonical profile, and making my HRD Corp accreditation publicly verifiable. If the method works, the numbers will move; if they do not, this page will say so.

What this means for executives

Three lessons carry over to any executive. First, check for name collisions before anything else; if an assistant cannot tell you apart from someone else, nothing else you publish will land. Second, make the sources assistants trust—LinkedIn, directories, company bios—agree with one canonical profile you own. Third, describe what you do in plain, specific terms on a page of its own, because assistants cite clear service and topic pages, not implied expertise. The practical steps are in how to fix wrong information about you in AI (https://benyeoh.com/fix-wrong-information-ai).

Results by question type and run

Run this test on yourself

The method takes an afternoon and needs no tools beyond the assistants themselves.

Questions