How to Get Better Results from Emma, Your Amazon AI Agent

Written by emax digital | Jul 16, 2026 3:14:09 PM

AI chatbots have gotten remarkably good at answering questions, but they're only as good as the question you give them. Ask something vague, and you'll get an answer that sounds confident but might miss the mark entirely. Ask something precise, and the same chatbot suddenly feels like a sharp analyst who's read every report you have.

This isn't unique to any one tool. It's true of ChatGPT, Claude, and Copilot, and it's true of Emma, the AI assistant built into emax digital. Below are the habits that make the biggest difference, with examples of what they look like in practice inside Emma.

Have you met Emma?

Emma is emax digital's AI chat assistant, currently available to selected users in beta. Instead of navigating dashboards and filters to find an answer, you can just ask Emma directly in plain language and get back a formatted, structured response in seconds.

Emma is aware of what's already on your screen and can also pull specific figures from your connected Amazon accounts when asked, including numbers you can't get directly from Amazon's own consoles. If you're curious how that works in more detail, let's talk.

The goal is simple: turn "digging through a dashboard" into "asking a question," so you spend less time finding data and more time acting on it. Getting good at how you ask is what makes that goal actually pay off, which is what the rest of this post is about.

1. Say the direction, not the judgment

Words like "best," "worst," or "weakest" feel clear to us, but they're genuinely ambiguous to a chatbot. "Worst ACOS" could mean the highest ACOS (bad) or the lowest ACOS (technically the most efficient). The model can't always guess which one you meant, and it will pick one silently rather than ask.

The fix: replace evaluative words with directional ones.

Instead of this Try this
"Show me my worst-performing campaigns" "Show me the 5 campaigns with the highest ACOS"
"Which are my best products?" "Which are my top 10 ASINs by revenue?"
"What's my weakest marketplace?" "Which marketplace has the lowest revenue this month?"

In Emma: this matters constantly, since so many Amazon metrics (ACOS, margin, conversion rate) have a "lower is better" logic that isn't obvious from the word "worst" alone. 

2. Always give a time frame

Always give a specific time frame, such as "last month," "June 2026," or "last 4 weeks." Vague time frames get vague, or inconsistent, answers. Most assistants will either default to something arbitrary or ask you to clarify.

In Emma: a prompt like "show me my revenue trend" can return a different window depending on the day you ask it. "Show me my monthly revenue trend for the last 6 months" doesn't leave that to chance.

3. Tell the chatbot what not to do

This is probably the single biggest upgrade you can make to any prompt. Most chatbots are trained to be helpful, which means they'll often add context, sections, or comparisons you didn't ask for. A boundary or two keeps the answer focused on exactly what you need.

A few guardrails that work in almost any prompt:

  • "Don't mix [category A] and [category B] data together."
  • "Don't add sections I didn't ask for."
  • "If a number looks unusually high or low, flag it and explain why."

In Emma specifically: "Don't mix Vendor and Seller data in the same ranking" is one of the most useful boundaries you can add, since combining the two can produce numbers that look right but aren't comparable.

4. One topic per question

Asking about inventory, content quality, declining products, and ad efficiency all in a single prompt sounds efficient, but it usually produces a shallower, less reliable answer than four focused questions asked one at a time. Chatbots, like people, do their best thinking with a narrow brief.

Putting it together: a well-built prompt

A good prompt for any AI assistant usually has four parts: a subject, a direction, a time frame, and a boundary. Here's what that looks like inside Emma:

Show me the top 10 and bottom 10 ASINs by revenue change vs. last month for [UK] only. Include revenue, units, and margin for each. Don't mix Vendor and Seller data.

Compare that to "how are my products doing?" It's the same underlying question, with very different odds of getting something useful back.

Common mistakes worth avoiding

  • Too short, no guardrails. "Create me a business review" will technically work, but without a scope, time frame, and a few boundaries, expect a generic answer.
  • Too many topics at once. Split it into separate questions instead.
  • Assuming "today" means today. Most AI assistants working with business data (Emma included) don't have same-day figures. "Last week" or "last month" is a safer bet.

Save your best prompts, don't rewrite them

Once you've found a prompt that reliably gets you what you need, save it rather than retyping it. In Emma, this is what Skills are for: a shortcut you trigger with a /command instead of writing the whole prompt out again. A good skill is really just a good prompt (subject, direction, time frame, boundaries) saved for one-click reuse.

The gap between a mediocre AI answer and a genuinely useful one is rarely the tool. It's the prompt. Get specific, and the chatbot you already have starts feeling a lot smarter.

Want to see this in action?

Emma is currently rolling out to emax digital users in beta. If you'd like early access, or just want a quick walkthrough of what it can do for your Amazon business, get in touch with us at!