Ask in plain language about your rankings, ads, AI mentions, map coverage or search history. Data Chat writes the query, runs it on your own data, and answers in a few sentences with the table and the SQL underneath.
| Keyword | Was | Now | Change |
|---|---|---|---|
| running shoes berlin | #2 | #7 | -5 |
| best running shoes | #3 | #5 | -2 |
| trail running shoes | #3 | #4 | -1 |
| running shoes sale | #1 | #3 | -2 |
Four keywords dropped out of the top 3 in the last 30 days, and three of them went to the same competitor. "running shoes berlin" fell furthest, from #2 to #7; the rest moved one or two places, which is normal week-to-week noise.
Trusted by 10,000+ SEOs, agencies, and brand teams
Data Chat reads the same tables your trackers write to. Ask across tools, compare periods, rank competitors, or just find the number you need, without building a report first.
Your average position improved from 31.2 to 23.4 over the last 30 days, almost all of it on mobile.
Rank movement, winners and losers, per location and device.
Three advertisers hold 91% of the ad impressions on your keywords. runfaster.com is new this week.
Advertiser share of voice, new entrants, and how ad copy changes over time.
ChatGPT #2 Mentioned 48%
Gemini #3 Mentioned 41%
Perplexity #1 Mentioned 38%
Claude Missing 0%You are mentioned by 3 of 4 assistants. Claude never names you on comparison prompts.
Mentions, sentiment and position across ChatGPT, Gemini, Perplexity and Claude.
Coverage is weakest in the north-east of the grid: 8 of 25 points sit outside the top 3.
Local pack presence for every grid point and keyword you track.
Your team ran 104 searches this month; 40% of them from New York on mobile.
Every proxy and direct search your team ran, by location, device and language.
| Keyword | Your rank | Advertisers |
|---|---|---|
| running shoes sale | #3 | 4 |
| running shoes berlin | #2 | 3 |
| trail running shoes | #1 | 2 |
All three keywords you hold in the top 3 carry competitor ads. "running shoes sale" has four advertisers on it.
Joins between trackers, the kind of answer no single report shows.
No black box. You see the question, the query it wrote, the rows it got back, and what it made of them.
SELECT r.keyword,
min(r.position) AS best_position
FROM serp_position_results r
WHERE r.checked_at >= now() - interval '30 days'
GROUP BY r.keyword
ORDER BY best_position;| Period | Avg position | Keywords |
|---|---|---|
| Previous 30 days | 27.5 | 48 |
| Last 30 days | 23.4 | 52 |
Your average position improved from 27.5 to 23.4, mostly on mobile. The gain comes from six keywords entering the top 10, not from the rest moving.
Queries run through a read-only database role scoped to your organization. There is no path to another customer's rows, and nothing can be written, changed or deleted.
The model looks up the tables it needs, writes a single SELECT, and the statement is shown under the answer. Queries that fail the guard are shown too, with the reason.
The result table sits under every answer, and a chart is drawn when the question is about a trend, a comparison, or a share of something.
Two to four sentences: the answer to the question asked, then the one thing worth noticing. If the data cannot answer, it says so instead of guessing.
Threads belong to whoever asked. Open one up to the whole team, or keep it to yourself.
A new thread is visible only to you until you share it with the organization.
Shared threads show who asked what, so a colleague can pick up where you left off.
Each answer costs a few tokens from your balance, shown right under it. No seat fee, no separate subscription.
Threads keep their tables and charts, so an answer from last month still shows the data it was based on.
Only your organization's: position tracker, ad tracker, AI visibility and Maps campaigns with their runs and results, plus your proxy and direct search history and projects. It cannot see other customers, billing, or anything outside those tables.
No. Every query runs through a read-only database role inside a read-only transaction, and only single SELECT statements are accepted. Writes, schema changes and system tables are refused before they reach the database.
It says so. If the tables cannot answer a question, the reply explains what is missing instead of inventing a number. Every answer shows the SQL and rows it used, so you can check it yourself.
A few tokens from your organization's balance, charged per answer and shown under it. The charge follows the actual work: a short follow-up in an existing thread costs less than a cold question that needs several queries.
Not unless you share them. Threads are private by default; you can open one to the whole organization, and you can rename or delete your own at any time.
It is in beta. The questions it handles well are the ones on this page, and it will get better at the rest as we read the feedback on answers. The rules around your data are not beta: reads are read-only and scoped to your organization, and you only pay for answers you receive.
Your trackers already hold the answer. Ask for it in a sentence and get the number, the table and the chart in seconds.