How to Get the Best Results from Emma

The way you phrase your question has a significant impact on the quality of Emma's response. A well-structured prompt can be the difference between a confident but misleading answer and a precise, actionable one. This article gives you practical rules and examples to get reliable results from Emma.

Three rules for better results

These three principles apply to every question you ask Emma.

1. Be specific about direction — not evaluative

Say "highest ACOS" instead of "worst ACOS." Say "lowest revenue" instead of "weakest performers."

Why: Evaluative words like "worst" or "best" can be ambiguous — "worst ACOS" could mean the highest ACOS or, read differently, the best efficiency. Directional language removes the ambiguity completely, so it stays the recommended way to phrase these questions.

2. Always specify the time period

"Last month", "June 2026", "23–26 June 2026", "last 4 weeks" — never leave the time period open.

Why: Without a time period, Emma may pick a default that doesn't match your expectation. Two runs of the same prompt can return data from different timeframes.

Two facts about dates that often surprise people:

• Weeks run Sunday to Saturday. A "last week" question resolves to a Sunday–Saturday window, not Monday–Sunday.

• "Last 14 days" ends yesterday, not today. Today's data isn't fully loaded yet, so trailing windows always end the day before. If you compare Emma's numbers to another dashboard and see a one-day difference, this is usually why.

3. Tell Emma what NOT to do

The most effective way to improve results is to include explicit boundaries. Emma is designed to be helpful and will try to add value — guardrails channel that into the right places.

Useful guardrails you can include in any prompt:

• "Do not combine Vendor and Seller data in a single ranking."

• "Do not add sections I didn't ask for."

• "If any value looks unusually high or low, flag it and explain why."

• "State which data source each metric comes from."

Snapshot or trend? Emma has two ways to answer performance questions

This distinction is worth understanding before you ask any question involving time — it's the difference between getting a single number and getting the full picture.

• Snapshot gives you the metrics for one period (e.g. "June 2026"), together with an automatic comparison to the previous period and to the same period last year.

• Trend analyses multiple periods at once and gives you statistics across the whole window — the lowest period, the average, and the overall growth rate. This is what makes questions about seasonality, trajectory, or "is this improving" answerable at all.

Good phrasing for a trend question: "trend", "development over", "month by month", "how has X changed across", "seasonality". Good phrasing for a snapshot: "in June", "last quarter", "top 10".

Two things worth knowing about trend questions:

• A trend gives you one combined series — you can scope it to a marketplace, ASIN, category or tag, but not a separate line per ASIN in the same request. For "which of my thousands of ASINs are losing traffic," ask about your top 50–100 products or a specific category instead.

• If you don't name a marketplace, always name a target currency — otherwise EUR, GBP and USD would get added together into a meaningless number.

Prompt examples by use case

Below are ready-to-use prompt structures for common tasks. Copy and adapt them to your marketplace, data source, and time period.

Performance trend

Good for: Understanding how a metric has changed over time.

Show me the monthly revenue trend for my [DE] [Vendor] marketplace over the last 6 months.

For each month, show the value and the change vs. the prior month. Also give me the statistics across the whole window: lowest month, average, and overall growth rate. If any period shows a change greater than 100%, flag it and suggest a possible explanation. State which data source each value comes from.

Top and bottom products

Good for: Identifying which ASINs are driving growth and which are declining.

Show me two separate lists for [UK] [Vendor] only — do not mix Vendor and Seller data:

1. Top 10 ASINs by revenue change vs. last month
2. Bottom 10 ASINs by revenue change vs. last month

For each ASIN, include: revenue, units, CVR, and margin. Exclude products with zero revenue in both periods. If a product could belong on both lists, note that rather than placing it on one only. State which data source each metric comes from.

Note if advertising is in the list: per-ASIN advertising revenue can include "halo sales" pulled along from other products. Individual rows are fine to read — just don't add up this column to get a total.

Advertising analysis

Good for: Reviewing campaign efficiency. Note: always use "highest" and "lowest" — never "worst" or "best" — when asking about ACOS or ROAS.

For my [UK] [Vendor] advertising, show me:

1. The 5 campaigns with the highest ACOS last month
2. The 5 campaigns with the lowest ACOS last month (excluding campaigns with zero spend)

For each campaign, include: spend, attributed sales, ACOS, impressions, and clicks. Calculate ACOS from the summed spend and summed sales for the period — do not average the ACOS values of individual campaigns. If ACOS exceeds 100%, flag it and explain what that means.

Good to know: campaign budget utilisation — how much of your own Amazon Ads Console budget a campaign has used — is available, so over/underspending questions can be answered directly. This is different from pacing against a target you've set inside emax digital, which isn't available yet.

Monthly Business Review (MBR)

Good for: A comprehensive monthly performance summary. This is the most complex prompt type — the guardrails below are essential for accurate results.

Create a Monthly Business Review for [June 2026] with year-over-year comparison, in EUR.

Show my largest Vendor marketplace and my largest Seller marketplace separately. Do not combine them in a single ranking.

For each marketplace: Revenue, Units, Traffic, Conversion Rate, Margin. Advertising: TACoS, Ad Sales Share, Ad Spend, ACOS, ROAS — broken down by ad type.

If data is not yet complete for the requested period, label values as "MTD" and compare against the same period last year. For each metric, state which data source it comes from. Calculate every percentage from the summed absolute values for the period — do not average percentages across marketplaces or ad types. Convert all monetary values into EUR and state that you've done so. Do not add sections beyond what I requested above. If any value looks unusually high or low, briefly explain the likely cause.

Good to know: the margin metric (Net PPM) only exists for Vendor accounts — Seller Central has no equivalent margin figure, so a Seller-only MBR won't include one.

Root cause analysis

Good for: Investigating why a product or marketplace is declining.

For [ASIN or product name], investigate why revenue is declining.

Check each of the following and state whether it's contributing to the decline, stable, or improving:
1. Revenue and units
2. Pricing and margin
3. Conversion rate
4. Advertising (spend, ACOS)
5. Inventory (availability, stockouts)
6. Content (listing quality)

If a data dimension is not available, say so explicitly rather than skipping it. Treat Vendor and Seller data separately.

Good to know: advertising can be tied back to individual products, but the cost side is more complete for Sponsored Products than for other ad types — see the limitations table below.

What to avoid

These patterns consistently produce worse results:

Too short, no guardrails
"Create me an MBR." — Without specifying the marketplace, time period, and boundaries, the output is likely to contain errors. Always use the full prompt structure shown above.

Combining too many topics in one question
"Show me inventory health, content issues, declining products, and ad efficiency for all marketplaces." — Asking for too many unrelated topics at once reduces reliability. Ask one topic per prompt. (A checklist against a single product or marketplace — like the root cause example above — is different and works well.)

Using "today" as a time reference
Today's data isn't fully loaded yet. Use an explicit date, "last week", or "last month" instead.

Averaging percentages
An account-wide ACOS is not the average of your individual campaigns' ACOS values — it has to be calculated from the summed totals, or the number is mathematically meaningless.

Summing per-ASIN advertising revenue for a total
This column includes halo sales from other products, so it won't add up to your real ad total. Read the rows individually.

Asking for a multi-marketplace total without naming a currency
Summing EUR, GBP and USD produces a meaningless number. Always name one target currency.

Current limitations

Emma is transparent about most of these when you ask, but it helps to know them upfront so you can plan your questions accordingly.

One thing worth clarifying: when Emma tells you a product is "missing" content (like A+ content), it means the product differs from your own master content saved in emax digital — not necessarily that nothing exists on the live Amazon page. If you want the absolute quality picture instead, ask for the Content Score.

These limitations reflect the current beta status. The data available to Emma will expand over the coming months — see "What Data Emma can Access" for the full roadmap.

Tips for Skills users

If you use Skills (prompt shortcuts), apply the same principles: include the marketplace, time period, data source, and guardrails in the skill text. A well-built skill is a prompt that already follows all the rules above — saved for instant re-use.

See the "Skills (Shortcuts)" article for how to create, manage, and schedule skills — including how to turn any skill into an automated, recurring report by email.

💡 Do you think the content on this page needs improvement? Let us know at support@emax-digital.com