A useful Chat starts with one clear job and the information needed to complete it. This article walks through a practical analysis job, from adding data through improving the first answer.
Use a connected Meta account to compare campaign and ad performance for a
specific date range without exporting the data first.
Analyze a CSV
Attach an ad-platform export, sales report, survey, or other data file and
ask a focused question about the rows inside it.
Review an ad or screenshot
Paste or attach an ad image, dashboard screenshot, or page section and ask
StefanBrain to explain what it sees and what should change.
Turn analysis into the next deliverable
Continue in the same Chat to turn one finding into a test plan, creative
brief, revised ad, or deeper investigation.
You normally should not manually select a tool. Tell StefanBrain the job you want completed and let it choose or combine the right capabilities. Select a tool only when you need that tool’s rigid controls or output.
A Chat is one saved conversation and decision trail. Use it for a focused job and the follow-up work that depends on the same goal and evidence.Context can reach the Chat in several ways:
Start inside a Project when the work should stay with that Project’s Chats, instructions, files, and linked Products.
Type @ in the composer to choose an available Product or Project by name.
Use Add files or photos, paste from your clipboard, or drag in a CSV, image, screenshot, brief, or other one-off source.
Enable a connected account when the job needs current data from that service. Browse connector guidance.
Mentioning a Project in a regular Chat can add that Project’s context to the request, but it does not move the Chat into the Project. Start from the Project when organization matters.
You do not need to repeat stable Product or Project context in every prompt. Once the correct context is attached, focus your request on the goal, analysis method, and deliverable.
Select New chat for account-level work. For work tied to an ongoing campaign, open its Project and use New chat in [Project name].
2
Add the analysis source
For direct Meta data, open Connectors and enable the intended account. This lets StefanBrain read the available account data without a manual export.If the data is not connected, use Add files or photos to attach a CSV. You can also paste or attach an ad image or dashboard screenshot when the visual itself is the evidence.Wait for an attachment to finish before sending. Name what StefanBrain should inspect instead of adding a source without a question.
3
Ask StefanBrain to complete the job
Leave the tool picker unselected unless you need a specific rigid workflow.
Then ask for the result in natural language.
Analyze the enabled Meta account from [start date] to [end date]and explain why [metric or business result] changed.Compare campaign and ad performance. Show the three biggest driversin a table with the supporting numbers, what each one likely means,and what I should check next.
If you attached a CSV, replace the first sentence with: Analyze the attached CSV from [start date] through [end date].The Product and Project already attached to the Chat provide their existing context. Restate only the details that are new or specific to this analysis.
This focused Chat names the CSV, connects the request to the CPA goal, and returns the first-pass evidence in the requested review format.
4
Review the first answer
Check whether the answer used the intended account or file, date range, and
metrics. Then ask:
Did it answer the business question you asked?
Does it show the numbers behind each finding?
Do the proposed next checks help answer the remaining question?
5
Explain what should change
Avoid a follow-up such as This is bad. Try again. StefanBrain needs to know why the answer missed your goal and what a useful revision should do differently.
This focuses on CTR, but I need to understand the CPA increase.Rework it around spend, CTR, CPC, conversion rate, and CPA.Show which numbers support each conclusion and the three checksI should run next.
The follow-up identifies the CTR-versus-CPA mismatch, then the revised answer reorders the findings around CPA evidence and remaining investigations.
This follow-up identifies the miss, redirects the analysis, and defines the revised deliverable without repeating the full Project or Product context.
6
Continue or start a clean Chat
Keep the same Chat when the next request depends on this analysis. Start a new Chat when the client, Product, campaign, data set, or main decision changes.
When a response misses the mark, answer four questions before sending the follow-up:
What part of the result missed your goal?
Why does that miss matter to the decision you are making?
What should StefanBrain analyze, preserve, or change?
What exact result should come back next?
For a visual review, turn a vague reaction into visible criteria:
Reanalyze the attached ad for hook clarity, visual hierarchy,product visibility, and CTA clarity.For each issue, point to the visual element behind itand suggest one focused change.
The goal is not a longer prompt. It is enough direction for StefanBrain to understand what success looks like.
