First prompts for your ad accounts
Tempio6 min read
A freshly connected agent is most useful in the first hour, when you still remember which numbers you were unsure about. This guide is a list of questions worth asking in that hour, what each one actually returns, and the two or three ways an answer can be read wrongly. Everything here works with a read-only key.
Start with the connections, not the numbers
The first question is the one that tells you whether any other answer means anything.
Which ad platforms are connected, and when did each last sync?
Each row carries a status and a last sync time, and it also carries a capabilities object describing what that platform can express: which change operations it supports, where budget attaches, which objectives it accepts, what a deletion means there, and whether a change can be dry-run against the platform before it is applied. Reading that once saves a week of assuming every platform behaves like the one you know best.
A connection that has never finished a sync is the single most common cause of a confusing first session. It answers every reading question with an empty set, which is indistinguishable from a broken client. Settle that before you draw any conclusions.
The three questions that frame everything else
Blended spend and results for the last thirty days, with the per-channel breakdown.
The blended figures come back with each channel underneath them. Two conventions apply and both matter. Money is an integer in minor units with its currency beside it — never a formatted string, so nothing is rounded twice. Ratios are computed at the moment they are read and come back empty when the denominator is zero.
That second one is the rule worth internalising on day one. An empty ratio means "there is nothing to divide by", never "zero". An account with no conversions has no cost per conversion; reading the blank as a nought asserts that conversions were free, which is the most common way a report ends up lying to the person holding it.
Which campaigns carried the spend last week?
Rows come back highest spend first, each carrying a marker that says what the ordering means: whether the figures were natively comparable, converted in order to compare them, or left raw. A raw order across currencies is not a ranking, and the marker exists so that nobody reads one as though it were.
Give me the brief for that ad account.
The account brief is the long answer: what the last ninety days covered, the totals, a comparison against the previous twenty-eight days, the mix of objectives, how concentrated the spend is, the rankings by return and by cost per acquisition, and the rhythm across weekdays. It is computed rather than live, carries its own age, and is flagged once it is more than seven days old. Say when it was computed before quoting it.
Structure questions, for the shape of the account
List every campaign on that account, including the ones that are not delivering.
The ones not delivering are the interesting half. Paused campaigns with budgets still attached, campaigns that ended without being cleaned up, and duplicates from a migration all live there.
What sits under that campaign — ad groups, ads, creatives, keywords?
Structure reads down through each level, newest first, each row carrying its own identifiers and an absolute link into the app. The link is for a person to open; it is not a grant, and the page behind it checks membership regardless.
Show me the media library.
The images and videos the workspace holds and has handed to the platforms, newest first. Useful early, because a library that does not match what you thought was live is a sign that somebody has been working outside the system.
Questions that are really about hygiene
Which campaigns have the highest spend and no conversions over the last fortnight?
Where is spend most concentrated, and what happens to the account's numbers if I remove the top campaign from the total?
Which ad accounts have not been touched since the connection was made?
These are the questions people mean when they say they want an audit. They are all reads, they are all cheap, and the answers tend to change what you were going to do next. If the numbers argue for a change, the arithmetic behind the argument is worth doing yourself: the ROAS calculator and the CPA calculator run on your own figures with no account attached.
Asking well, rather than asking often
One habit is worth building immediately, because it is the difference between an agent that is pleasant and one that is slow.
Ask for the rows in one call. A single question covering thirty days across every connected account returns one result. An agent walking campaigns one at a time makes one call per campaign, which is both slower and the fastest way to meet the rate limit — sixty requests a minute per credential after authentication, a hundred and twenty a minute per address before it.
Two habits behind that one. Name the window explicitly rather than saying recently, because a range ending today is reporting a day that is not over and comparing a partial day against a whole one invents a bad week. And ask what the answer covers when it matters: every result carries a report stating its currency, its time zone, its date range, its entity level and whether the scope is the rows returned or all of them.
What happens if you ask for a change
With a read-only key, nothing does — the drafting capabilities are not in the client's tool list at all, so the agent will tell you it cannot rather than trying.
With a draft-scoped key, asking for a change produces a batch of change requests and a link to the gate, not a change. That is the whole arrangement and it is worth stating plainly before you rely on it: your connections read your accounts, a change is written as a draft on the record with the state it would change from and the checks it passed, applying it is gated on approval — by you, or by a rule you wrote — and no platform write reaches a customer account until that platform has granted write access. Applying a cleared change to a platform is in private beta.
There is a good reason to keep the first session read-only anyway. The point of the first hour is to find out whether the numbers are what you believed they were. Finding out they are not is much better done before anything has been drafted against them.
Where to go next
If more than one workspace is involved, the boundary rules change and Run one agent across several ad accounts covers them. For the full inventory of what a connected client can and cannot see, What an agent can reach in your account is the honest list. And once the reading is settled and there is a quarter to shape, the planner is where that happens: How to write a media plan.
Read next
Agent setup guides
How to point your own AI at your ad accounts: the transport, the key, the scopes it carries, and what it can and cannot do once connected.
Connect Claude Code to your ad accounts
Mint a scoped key, add the MCP server to Claude Code, read the account in one prompt, and see what a drafted change looks like at the gate.
What an agent can reach in your account
An honest inventory: what a connected client reads, what it can write, what it can never touch, and the four limits that hold whatever you ask it to do.
Run one agent across several accounts
One workspace can hold many accounts across many connections, and one key reads all of them. Where the boundary actually falls, and how to work across it.
Questions
Put frontier AI to work on every ad account you run.
Tempio connects your ad accounts into one workspace and hands your own AI the numbers and the tools. Every change it writes is a draft until you approve it, or until a rule you wrote does.
Request early access- Read and draft by default. Approve is a separate permission you grant.