IMG 0116:9A model’s two sources of knowledge
An AI answer combines what the model “remembered” during training with what it finds in search right now. Search-enabled models such as Perplexity, ChatGPT with search and Alice rely on current search results. What appears when someone searches your name therefore shapes the answer directly.
IMG 0216:9Three typical errors
- Mixed identities: the model merges information about you and a namesake, especially when sources lack identifiers unique to you
- Skew towards the negative: when sources are scarce, a high-profile dispute or article becomes the defining fact
- Outdated information: a former role or a company you have sold, because there are no newer sources
IMG 034:3IMG 044:3Why you can’t simply “ask” a model to correct its answer
Contacting model developers about individual answers rarely works: a model does not keep a “profile card” on a person. The answer can only be changed through the sources it relies on.
IMG 0516:9What to do
- Ask the models 10–15 questions about yourself in Russian and English and save the answers
- For each statement, trace where it could have come from
- Create an official source with a biography, photo and unique identifiers
- Add sources that confirm your current role: directories, interviews, profiles
- Repeat the check once a month
IMG 0621:9Frequently asked questions
Does schema.org markup affect AI answers?
Indirectly. It helps search engines understand precisely who a page is about, and search-enabled models rely on that data.
How quickly do answers change?
For search-enabled models, within weeks of the search results changing. Answers drawn “from memory” change with model updates.
What to do next
If you recognise your situation, start with an audit: in five business days we will show you what a bank, a partner and an AI model see about you.
