AI Data Chat Overview

Data Chat answers questions about the data your campaigns have already collected — rankings, ads, AI mentions, Maps grids, saved searches — without you building a report or exporting anything.

Ask “which keywords did I lose top-3 positions for last month?” and you get a sentence, the rows it was derived from, the SQL that produced them, and a chart where one helps.

Evidence, not just an answer

The answer is the least trustworthy part of anything an AI tells you, so it is never the only thing you get. Every answer carries:

The SQL it ran

Every query, in order. If the answer looks wrong, the query tells you why.

The rows it read

The actual result set, not a summary of one. Named tables, real values.

A chart, when it helps

Drawn from those same rows — never from numbers invented in the prose.

Follow-up prompts

The next questions the data suggests, so a thread goes somewhere.

Check the query, not the wording

If a number surprises you, read the SQL before you believe or dismiss it. The most common cause of a surprising answer is a question that was ambiguous — “last month” and “the last 30 days” are different, and the query says which one it used.

What it can see

Your organization’s own data across every tool, and nothing else:

Area	What it covers
SERP Position Tracker	campaigns, runs, checks, ranked results
Ads Position Tracker	campaigns, runs, checks, ad results and creatives
Maps Visibility	campaigns, runs, checks, grid results, places
AI Visibility	campaigns, runs, checks, mentions, ad creatives
Searches	proxy searches with their pages and items, direct searches
Workspace	projects, alerts, schedules

What it cannot do

Being clear about this is part of trusting it:

• It reads. It never writes. No campaign is created, changed, paused or deleted by asking. The connection it uses is read-only at the database level, not merely instructed to behave.
• It cannot leave your organization. Row-level security scopes every query to your own data, so a question can never reach another workspace — whatever the question says.
• It cannot see everything in a row. A few columns are withheld from it entirely, alert webhook settings among them.
• It only knows what your campaigns collected. It cannot look up a ranking you never tracked. If nothing ran, there is nothing to ask about.

Or bring your own agent

Data Chat and the MCP server are the same capability with a different model driving it. Both read the same catalogue of tables, through the same read-only database role, under the same row-level security.

	Data Chat	MCP server
Where you ask	in the app	Claude, Cursor, or any MCP client
Whose model answers	ours	yours
Reads	the same 25 tables	the same 25 tables
Isolation	read-only role + RLS	read-only role + RLS
Billing	per answer, by measured cost	your own model’s cost; our reads are free

Use Data Chat when you want an answer without leaving the app, or without paying for your own model. Use MCP when you already work in an agent, want your own model, or want to combine this data with something else you have connected.

They are not alternatives you have to choose between

Both are available at once. The same conversation you have in the app can be continued from an agent, because the data underneath is identical.

Where to go next

Asking good questionsWhat to ask, how to narrow it, and how follow-ups work.

Reading an answerThe SQL, the rows, the chart — and how to tell a good answer from a confident one.

What it costsData Chat bills by what the question actually cost, not a flat price.

Open in iSearchFrom

Open Data Chat in the app.