Comparisons over time
“Which keywords dropped out of the top 10 since last month?"
What to ask Data Chat, how to narrow a question, and how follow-ups and sharing work.
Ask the question you actually want answered. You do not need to know the table names, and you do not need to write SQL.
Comparisons over time
“Which keywords dropped out of the top 10 since last month?"
Rankings and movement
Competitors
“Which advertisers appeared most often against my brand keywords?"
AI visibility
Local coverage
Housekeeping
A vague question makes the model guess, and a guess is where wrong answers come from. Three things make the biggest difference:
A conversation keeps its context, so the second question can be short:
You: Which keywords lost top-3 positions last month? You: Just the ones in Germany. You: Now show those as a chart.
Suggested follow-ups appear under each answer — the questions the data itself raises. They are a good way to find the question you should have asked.
Long threads are compacted automatically as they grow, so a conversation stays affordable rather than re-sending its whole history every turn.
Each conversation has one of two visibilities:
| Visibility | Who can read it |
|---|---|
| Organization | anyone in your workspace |
| Restricted | you, plus the people you name |
Sharing is inside your workspace only — a conversation has no public link, and switching a conversation to restricted does not hide it from workspace admins.
An unsent question is kept as you type, so navigating away and coming back does not cost you the sentence you were composing.