Skip to main content
You don’t need to learn commands — describe what you want. These examples show the range, and are good starting points to adapt. Everything falls into three levels:

​
Understand performance

Ask “why” as a follow-up. The assistant can drill from a headline number into segments, channels, and cohorts without you rebuilding the question.

​
Build audiences

Ask for a size first — it’s instant and changes nothing. Save it only when the number looks right.
When you exclude people who didn’t engage, ask Claude to also require that they were actually messaged — otherwise the audience fills up with people who never received anything in the first place.
Contacts can carry email, phone, first and last name, plus any extra columns you want to keep as attributes.
A saved audience is a rule, not a frozen list — it is recounted against today’s data every time you ask. An imported list is the exception: it reports the size it was imported with, and says so.

​
Prepare campaigns

Creative generation runs in the background — asking whether it finished is a real question, and you’ll get the status rather than an empty list.
Templates can be searched by what they say, not just by name. Reading one back gives you the copy a recipient would see, so you can judge tone and brand fit without opening the editor.
A shopping campaign’s template must contain a product-feed block — that’s what gets filled with products at send time. Ask for feed-carrying templates and you only see the ones that qualify.
A draft is correctable, not disposable — you don’t have to recreate it to change your mind. Only what you name changes; everything else stays put.
Once a campaign is running, only its name, description, budget, schedule and sending limits can change, and you’ll be asked to confirm first — moving a schedule or raising a budget on a live campaign takes effect immediately. A finished or cancelled campaign can’t be edited at all.
This is the step that produces the preview: resolved audience, cost per channel, wallet balance, any blockers — an unverified sending domain or missing creatives, for example — and how many days delivery will take at your daily sending cap.
The days matter more than people expect. A 100,000-person audience on a channel capped at 5,000/day is a 20-day send, not a Tuesday morning. Nothing is wrong with it — but it changes what “launch on Monday” means.
Launching a campaign, sending a blast, adding people to a running campaign, or firing a scheduled run immediately all send real messages and spend real budget. Each one requires a fresh preview plus your explicit approval.

​
When something looks wrong

Before concluding a campaign underperformed, check that it ran at all.
A campaign that failed outright looks exactly like a campaign with low volume — both show small numbers — and a campaign that never started looks the same again. The difference only appears when you ask for its errors and its run history, so ask before you rewrite the creative.

​
Guided methods

The assistant carries a set of growth playbooks. Ask for one by name when you want a rigorous pass rather than a quick answer:

​
Tips for better answers

  • Give a time range. “Last 90 days” beats “recently”.
  • Name the campaign or audience. Ask Claude to list them first if you’re unsure.
  • Ask for the size before saving an audience — it costs nothing and avoids surprises.
  • Ask why. Follow-ups are cheap; the assistant keeps the context of what it just showed you.
  • Say when to stop. “Draft it but don’t send” makes the boundary explicit — though the confirmation step protects you either way.
  • Check who a number counts. Every answer states its population — buyers, contacts we’ve messaged, tracked visitors. Two correct numbers from two questions can still be about different people, so read that line before comparing them.
Under the hood the assistant has a fixed set of capabilities. You never call these directly, but this is the complete boundary of what it can do:Reading — account and campaign listings; a campaign’s configuration, its errors and its run history; funnel, attribution, per-channel, variant, lift, spend/ROAS and creative reports; creative generation status; data overview and event catalog; cohort catalog and cohort matrices; retention buckets; growth equation; LTV and AOV; product coverage and product search; converter profile; touch fatigue; audience sizing (saved, by rule, or ad hoc); segment listings with name search; template listings searchable by content, and reading one template’s copy back; campaign schedules; send previews and cost estimates.Creating and editing drafts — campaigns (autonomous, shopping, triggered), AI creatives, shopping subject lines, audiences (rule-based or imported) and templates; editing an autonomous or shopping campaign; pausing, resuming and cancelling campaigns; updating, pausing and deleting schedules.Sending — confirmation required — launching a campaign, sending a one-shot blast, dispatching a shopping campaign, adding recipients to a running campaign, and firing a scheduled run immediately.Not included: moving money, changing billing or account settings, or sending raw messages outside a campaign.