hello, that's a common question : IA are not "search engine" like google. It's totally natural to have different answer for the same question : AI are not deterministic by design.
it use "RAG" retrieval augmented generation to get information and use GEN IA to create answer.
The RAG is not a google search engine : it provide "few references" (1, 3, some time 5, could be 10 in some case) but never all result. It use the most probabl result based on the graph of the user (doc use frequency, prefered deducted document etc) and it's not searching for a word but a "chunk" / part of a content containint something that it's probably what you want. It's not searching "this word contained in this page/document" it's the opposite of a search engine or a database query language.
So it's not an issue, it's just not the purpose and design of actual IA are doing.
An alternative exist for dataverse : use the dataverse MCP connector in the tool section (not knowledge) it will vibe SQL query on database (and you can instruct it how to make it) : this way it will not RAG but use SQL query with things like "this line where "xxx" is contain in this label" so it will give answer (limited to 20 line at max). (just be carefull, there a 3 dataverse MCP connector in the tool : read description, 1 is depracated, 1 is preview, middle one is prod <- use this one).
this way you can be deterministic on the query and content retrieved. The answer will be created with genIA so it need a strong prompt to be nearly always the same :)
For sharepoint list : a SharePoint MCP tool is under "frontier preview" and will be release soon we all hope and it will probably work the same.
For document : Copilot studio and low code AI are black box, so ATM it's complexe to change behaviour. It could need some microsoft foundry agent tweak to try something like this. but i m not confident at 100% : AI is not search engine.
I hope this help you to understand and you could use some of my idea to solve the problem :)
If yes please click on the this is the good answer and mark question green, it's important for the community and :D search engine ;)
