Key answer
Keep one main name for the same object; synonyms are acceptable when aliases are explained. Evidence neither requires AI's wording nor establishes that synonym changes necessarily reduce quality.
Full explanation
The issue is not that changing words inherently “loses attention,” but whether the model recognizes the same referent and whether you change scope along with the name. AmbigQA examined event, entity and other ambiguities in 14,042 questions, showing that referents are not always unique. It did not measure a benefit from fixed terminology. Study WinoDict also shows that definitions help models use new names without guaranteeing consistent compliance. Study
Handle three cases separately:
· Aliases only: if “knowledge card” and “node” mean exactly the same thing, establish that once and retain one primary name.
· A more precise industry meaning: check that it matches the need before adopting it; retain “everyday name (technical name)” when useful.
· Different scopes: “AI search” and “web retrieval tool” are not automatically identical. Do not merge them just for consistency.
Terminology: a “node” is an independent knowledge topic in this project; a “card” is the interface displaying a node. They are related but different. Use this distinction and flag inconsistencies first.
Consistency makes a shared convention checkable; it does not give a word special prompting power. Reintroduce necessary definitions in long projects or new conversations.
References
1. Eisenschlos et al. (2023), WinoDict: Probing language models for in-context word acquisition, EACL
2. Min et al. (2020), AmbigQA: Answering Ambiguous Open-domain Questions, EMNLP