Why Does AI Get My Company's Founder or Leadership Wrong?

Why do assistants guess about founders?

Research on why language models hallucinate gives a useful way to think about it. Kalai and colleagues argue that facts which appear only once in training data are the ones models are most likely to get wrong. Their example: if 20% of birthday facts appear exactly once, base models can be expected to hallucinate on at least 20% of birthday facts. The founder of a large company is repeated thousands of times. The founder of a 30-person company may be named on a handful of pages. When asked, a model that has seen the company but not a clear founder statement may produce a plausible name rather than say it does not know.

Search-based answers help only if a page states the fact plainly. "Our story" pages written as narrative, with the founders mentioned by first name in the third paragraph, are easy to misread.

What are the common patterns?

Why does AI know a business but get its founder or category wrong?

Because recognising a business and knowing facts about it are different things. An assistant may have seen your company name often enough to recognise it. Its founder and category, though, may appear only in passing, and often inconsistently. Several things push the category off course:

• The founder's earlier work. If the founder previously ran an agency, answers may describe the new product company as an agency.
• Words in the name. A name containing "Labs", "Studio" or "Partners" suggests a category that may not fit.
• Early positioning. A company that pivoted is often still described by its launch category, because that is what the coverage said.
• Thin category statements on your own site. If the home page says "We help teams work better" and never names a category, assistants infer one.

The fix for both is the same: state the facts plainly and repeat them where assistants look. How assistants misclassify a service business is covered in why AI describes your service business as a different kind of provider.

What should your own site say?

A leadership page with plain sentences. "[Company] was founded in 2019 by [Name] and [Name]." Then list each leader: full name, title, the year they took the role, and a short background. Keep former leaders off the page, or in a clearly labelled history section with their dates.

An about page that names the category. "[Company] is a [category] for [customer]." Put it in the first paragraph.

A company history with dates for founding, pivots, acquisitions and leadership changes. Dated milestones stop earlier facts being presented as current.

Structured data. Google's organisation markup documentation lists foundingDate, legalName and sameAs, among others. Schema.org also offers founder, which can name the founders as Person entries, consistent with the visible text.

Which external sources matter?

• LinkedIn: each leader's current title and the company page. Former leaders' profiles should show an end date.
• Business databases: founders, leadership and founding dates on company and startup databases. Request updates where they are wrong.
• Wikidata and Wikipedia: "founded by", "chief executive officer" and key people fields, updated through the sites' own processes with sources.
• Press releases: announce leadership changes with dates, and say who the person succeeds.
• Speaker bios and author pages that describe leaders' current roles.

How an inaccurate Wikipedia infobox feeds answers is covered in whether an inaccurate Wikipedia infobox affects ChatGPT or Gemini answers about your company.

What should I fix when AI search merges my founder's name with our company name?

When a company is named after its founder, such as "Morgan Hale Design" founded by Morgan Hale, assistants blend the two. The founder's personal history becomes the company's, company reviews are read as reviews of the person, and "Morgan Hale" questions return company details. Fix it by giving each a distinct, consistent identity:

• Separate pages: a company about page and a personal page for the founder, each stating the relationship. "Morgan Hale Design is an interior design studio founded by Morgan Hale in 2016." "Morgan Hale is an interior designer and the founder of Morgan Hale Design."
• Use the full company name every time, with its descriptor, rather than the founder's name alone.
• Separate profiles: a company page and a personal profile on LinkedIn, and separate business and personal social accounts, each linking to the other.
• Different markup: Organization on the company page, Person on the founder's page, and sameAs links pointing to each entity's own profiles.
• Knowledge panels: if Google shows a panel for either, the subject or an official representative can get verified and suggest changes.
• Reviews: ask clients to review the company's profile under the company's name.

How assistants handle name variations more broadly is covered in why Gemini describes your business under a different name from your website.

How do you test what assistants say?

Ask ChatGPT, Gemini, Perplexity and Google's AI features, several times each:

• "Who founded [company]?"
• "Who is the CEO of [company]?"
• "Who runs [company]?"
• "What kind of company is [company]?"
• "Who is [founder name]?"

Record whether each answer is correct, which person or category it gives when wrong, and which sources it cites. The cited source is usually the fix. Retest monthly, and after any leadership change.

What might an illustrative case look like?

Consider an illustrative analytics software company founded by two people, with a new CEO appointed a year earlier. Assistants named only one founder and described that founder as CEO. Some answers also described the company as "a data consultancy", which was the founder's previous business. Its "our story" page mentioned the founders by first name only, and its home page never named a category. The company rewrote its about page to start with its category and founding sentence. It added a dated leadership page and published a short release on the CEO appointment. It updated three business databases and asked the former CEO to update their LinkedIn title. Within two months, most answers named both founders, the current CEO and the right category.

What do people ask about founders and AI answers?

Why does AI invent a founder's name?

When no clear source states it, models can produce a plausible guess. A plain founding sentence on your site gives search-based assistants something to quote.

Should we remove former leaders from our site?

Remove them from current leadership pages. A dated history section can keep them, clearly marked as past roles.

Does our founder need a personal website?

Not necessarily, but a personal page on your site or a consistent profile helps assistants tell the person and the company apart.