Update on the name-collision saga: which fix actually moved ChatGPT

Leo H.
⚠️ Hallucinations & brand confusion

Follow-up to my earlier post about ChatGPT confusing me with a dead 2019 startup.

After 3 more weeks of testing, here's what specifically moved ChatGPT (not Perplexity, not Google AI Mode — they were easier):

The one thing that worked: an explicit disambiguation paragraph on the homepage.

Literal copy I added below the hero:

> '<My Brand> is the <my year> company building <category>, not to be confused with the unrelated 2019 social-analytics startup of the same name.'

Within 18 days, ChatGPT's brand-name probe answer stopped pulling stale data.

Why I think this worked:

LLMs reading my page now see explicit disambiguation language in the same context as the brand name. They don't have to disambiguate from training data alone.

Caveat:

This only worked because I have a clearly different category from the dead startup. If two live companies share a name in the same category — much harder.

The schema.org sameAs and disambiguatingDescription properties matter less than I thought. Plain English in the body matters more.

42

2 replies

  1. Ada K.

    The plain-English body text working better than schema is counterintuitive but it makes sense when you think about it — LLMs are reading the page as text, not parsing structured data first. The structured data helps at the entity-recognition layer but the actual description the model forms comes from the body copy.

  2. Jess Wright

    This is actually useful for a different problem I have. We changed our product name 14 months ago and ChatGPT still uses the old name. Going to try adding a 'formerly known as X' line directly in the hero body text. Did you place yours above the fold, or is it okay buried lower on the page?

Add a reply

Have a related question or experience?

Post a new question