Ask Acumin is a chat that knows your brand. Which makes it more useful than a general assistant for some questions and no better for others. Knowing which is which is most of getting value out of it.
Three things that define it
It's grounded in your brand. Your content, your analysis, your competitors. It's not answering from general knowledge about your industry; it's answering from your data.
It's read-only. It's advisory. It doesn't change your brand data, generate briefs, or take actions on your behalf. It reads and it argues.
It can't invent a statistic about you. This is the important one, and it's enforced rather than encouraged.
The number guard
There's a specific mechanism worth understanding, because it changes how you should read the answers.
Any figure in a sentence that talks about your brand's own performance, and that looks like a statistic, has to trace back to a real tool result. If it can't, it doesn't survive.
General numbers (dates, specifications, public facts, arithmetic) aren't touched. The guard is aimed at exactly one failure mode: a fluent, confident sentence containing a plausible number about you that nothing actually computed.
That's the same doctrine as everywhere else in the product: numbers from code, words from the model. The practical consequence is that the assistant will sometimes tell you it doesn't have something rather than produce a number that would have fitted nicely.
That's the feature. An assistant that always has an answer is an assistant you can't use for anything that matters.
Ask it things only it can answer
The highest-value questions are the ones that require your data:
- "Which of my formats has the widest gap between views and engagement?"
- "What do my top five posts have in common that my bottom five don't?"
- "How does my cadence compare to the competitors I'm tracking?"
- "Which of my briefs did I generate and never open?"
Compare those to questions any general assistant answers just as well:
- "What's a good hook structure for short-form?"
- "How long should a brand video be?"
Not wrong to ask, just not where the advantage is. The library is usually better for those, because it's edited.
Asking well
Be specific about scope. "How am I doing?" produces a summary. "Which platform has declined most over the last three months, and is it cadence or reception?" produces an answer.
Name the decision. "I'm choosing between two formats for next quarter. What does my data say about each?" gives it something to argue towards. Analysis without a decision attached produces description.
Ask for the evidence. "What's that based on, and how many posts?" is a good follow-up to anything. Sample size is the difference between a finding and a direction, and asking makes it explicit.
Push back. If an answer feels too clean, say so. It's a conversation, and the second answer is often better than the first because you've narrowed what you're actually asking.
Start a new chat for a new topic. Threads carry context, which is helpful within a subject and noise across subjects.
What it won't do
Worth knowing so you don't spend three messages discovering it:
It won't generate your briefs. Signal Briefs and Creative Briefs are their own surfaces, metered, and deliberately deliberate.
It won't change anything. Read-only advisory. Nothing you say to it edits your brand.
It won't work on demo brands. Blocked there: no API spend on the showcase account.
It only knows what's connected. If a platform isn't connected, it has no first-party data for it and will say so rather than estimating. See connecting your accounts.
Use it as a reading companion
The pattern that works best in practice: open your Snapshot or Content Analysis, then ask the assistant about the specific thing you don't understand.
"The headline says my engagement per view is strong on Reels. How many posts is that based on, and how does it compare to my long-form?"
You've given it a precise question anchored to something real, and the answer will either strengthen or deflate a conclusion you were about to act on. Both outcomes are worth the thirty seconds.
Most surfaces carry a contextual prompt for exactly this: a seeded question that fits the page you're on.
How to use this tomorrow
Open your Snapshot, find the claim you're least sure about, and ask the assistant what it rests on and how many posts are behind it.
If the sample turns out to be small, you've just avoided acting on an early signal as though it were a finding, which is the most common and most expensive content mistake there is.
Related: Measured vs reported is the doctrine behind the number guard. Reading your first Snapshot is the best thing to have open while you ask.