TL;DR
The ten Amazon Ads MCP prompts worth saving join Amazon ad data with something Amazon can't see, such as your margins, stock plan, brand-term list or budget plan, so each answer changes a spend decision. Four of the ten are read-only.
The prompts are plain language, so they run in Claude, ChatGPT or any assistant connected through MCP (Model Context Protocol). Replace every bracketed value, such as [account X] or $[30], before you paste. Write prompts also need the read permissions, and each permission is granted separately; see read-only vs read-write MCP.
Each answers a question an Amazon Sponsored Ads manager asks every week or month, and each is slow to answer in the Amazon Ads console alone. The test: the answer has to change a budget, a bid, a negative or a target ACOS, use your own economics such as break-even ACOS, and be safe to start, either read-only or a write with a preview.
Prompts 1, 2, 4 and 7 are read-only analysis. The other six each need one write permission to apply what they find: prompts 3, 6, 8 and 9 need Edit campaigns, prompt 5 needs Create and edit AI-managed groups, and prompt 10 needs Create campaigns. The permission names are Xnurta's; other MCP servers split access in similar ways.
Prompt 1: Which ASINs are profitable after ad spend? (read-only, Performance data)
For account [X], pull ad spend, ad-attributed sales and total sales by ASIN for the 30 days ending 14 days ago. Here's unit margin before ad spend for each SKU: [attach]. Calculate break-even ACOS per ASIN (equal to margin %) and ad profit as attributed sales x margin - ad spend. Flag any ASIN whose ACOS is above break-even, and rank ASINs by ad profit, lowest first.
Prompt 2: Why did this ASIN's sales drop? (read-only, Performance data, Ads configuration, Operation logs)
ASIN [B0XXXXXXXX] in account [X] sold [20]% fewer units in the last 14 days than the prior 14. Compare the two periods on sessions, Buy Box %, inventory, ad spend, ad sales, CPC and CVR, and list any bid, budget or status changes on its campaigns in both periods. Tell me which cause is most likely and what evidence supports it.
Prompt 3: Which search terms should move to exact match, and which should be negated? (Edit campaigns to apply)
In account [X], find search terms from the 30 days ending 14 days ago in auto, broad and phrase campaigns with at least [3] orders and ACOS below [break-even]%, that aren't yet an exact-match keyword anywhere in the account. Exclude any term on my brand list: [paste]. For each, name the exact-match campaign and ad group to add it to, a starting bid equal to its average CPC, and the source ad group where it should become a negative exact. Show me both lists and the old to new preview before applying anything.
Prompt 4: What changed before this campaign's performance moved? (read-only, Operation logs, Performance data)
Campaign [X] in account [X] went from [25]% to [41]% ACOS between [date] and [date]. List every change to its bids, budget, placements, targets and status in the 14 days before and during that period, with who or what made it, when, and the old and new values. Then show daily spend, CPC, CVR and ACOS for the same period and mark each change on the timeline. Here's our promo calendar: [attach]. Which change or event most likely explains the shift?
Prompt 5: Which AI-managed groups should get a new target? (Create and edit AI-managed groups to apply)
For account [X], list every AI-managed group with its goal, target ACOS, levers on and the last 30 days of actual ACOS, sales and spend. Here are our category margin targets: [paste]. For any group more than [5] points above the target its margin allows, propose a new target ACOS, list old and new values, and don't change anything yet.
Prompt 1 comes first because ACOS alone hides the truth: a product with a 60% margin is healthy at 35% ACOS, while a product with a 20% margin loses money at 25%. The window ends 14 days ago because attributed sales keep arriving for up to two weeks. The file-joining method is in how to combine Amazon Ads data with your own data. For prompt 2, if Buy Box percentage fell, fix the offer before adding budget. Prompt 3 negates at the source so your own campaigns don't bid against each other; Xnurta MCP can add the keywords and negatives itself, mainly for Sponsored Products. Prompt 4 uses operation logs whose entries hold old and new values; see who changed my Amazon campaign? Prompt 5 is the highest-leverage prompt here, because one target change steers every campaign in the group; see AI-managed group levers and personality.
Prompt 6: Which ASINs will run out of stock before new inventory lands? (Edit campaigns to apply)
For account [X], pull current inventory and the last 30 days' units sold by ASIN, and calculate days of cover. Here's our open purchase order sheet with arrival dates: [attach]. For ASINs with under [21] days of cover and no order arriving before they run out, list active campaigns and daily budgets and propose budget cuts. For ASINs with more than [90] days of cover, suggest where to add budget. Show me the old to new preview before applying.
Prompt 7: How do brand, non-brand and competitor terms compare? (read-only, Performance data)
For account [X], classify every search term from the 60 days ending 14 days ago as brand, competitor or non-brand, using these lists: brand terms [paste], competitor brands [paste]. For each segment, show spend, share of spend, sales, ACOS, CVR and CPC. Then list the top [10] non-brand terms by sales and the top [10] competitor terms by spend, with their ACOS.
Prompt 8: Which placement adjustments should you change? (Edit campaigns to apply)
For Sponsored Products campaigns in portfolio [X], compare the last 30 days by placement on clicks, CPC, CVR and ACOS, with current placement adjustments. Here's break-even ACOS by product: [paste]. Propose adjustments that keep each placement below break-even, including 0% where one runs above it. Show me the old to new preview before applying.
Prompt 9: How should you move budget without raising the total? (Edit campaigns to apply)
For account [X], find campaigns that ran out of budget on [3] or more of the last 14 days with ACOS below break-even, and campaigns that spent under [60]% of budget or ran above [45]% ACOS. Here's our monthly budget plan by portfolio: [attach]. Propose moving daily budget from the weak group to the capped group within each portfolio, keeping each portfolio's total within plan. Show me the old to new preview before applying, with portfolio totals before and after.
Prompt 10: Can you turn a launch brief into a paused campaign? (Create campaigns to apply)
Here's the launch brief for [product]: [attach]. Check which of its target keywords we already bid on in account [X], with current bids, CPC and CVR. Then draft a paused Sponsored Products launch: one auto campaign and one exact-match campaign for the top [10] keywords, $[30] daily budget each. Show me the full structure and wait for my OK before creating it.
Stock-aware pacing (prompt 6) stops you paying for clicks on products that can't ship. Brand, non-brand and competitor traffic (prompt 7) have different economics, so judge them separately; if branded terms take a large share of spend at a very low ACOS, test a lower brand bid on a few terms first. Placement adjustments (prompt 8) matter because a campaign's average ACOS can look healthy while one placement loses money on every click, and Amazon allows adjustments of up to 900%. For prompt 9, a campaign that runs out of budget can't show again until the daily budget resets, so capped winners lose sales every afternoon; keep hourly questions to about a one-week window. Plan ad groups up front for prompt 10, because ad groups can't be added to an existing campaign later; see create an Amazon campaign from a brief. Pause rather than archive any campaign you stop, because archived campaigns can't be reinstated.
Five habits: name the object (SP campaigns in portfolio [X], account [X], not my campaigns), set a settled window (the 30 days ending 14 days ago vs the prior 30, not recently), ask for the math (show the formula, date range and attribution window you used), spell out the output (a table sorted by ad profit, one verdict per row), and gate every change (show me the old to new preview before applying). Give the AI a short brief on each account's goals and tolerances so it doesn't guess.
When the decision is already made, repeats on a fixed schedule or needs a reaction faster than a conversation. The same budget raise every Prime Day is a job for Amazon's budget rules. Bid changes auction by auction belong to Amazon dynamic bidding or an optimization engine. Hundreds of edits you've already decided are faster as a bulk sheet; see bulk edit Amazon campaigns with AI. A team new to AI on client accounts can stay read-only, since prompts 1, 2, 4 and 7 still pay off.
It runs all ten on the same metric engine as the Xnurta platform and InsightAgent, Xnurta's in-platform AI analyst, so answers match the dashboard your team reports from. It offers about 15 months of history across 22 marketplaces, up to 90 days per request. Hourly reads cover SP, SB and SD campaigns and SP keywords for about a one-week window. Logs are labeled human, AI or rule, plus Amazon console, API connection or a named teammate, with old and new values, and record the tool call and ad account rather than the typed question. It also returns account TACOS, sessions, Buy Box and inventory; margin, costs and purchase orders come from your files. Write permissions are granted separately, with a server-enforced old to new preview for money, bids and archiving. New Sponsored Products campaigns are created all-or-nothing and can start paused. Generally available from 27 Oct for Sponsored Ads. Write access is turned on per account by your CSM; to keep one client read-only, give its connections read permissions only.
Start with read permissions only and run prompts 1 and 7 this week: prompt 1 gives you break-even ACOS per ASIN, which feeds prompts 5 and 8, and prompt 7 shows how much of your spend is branded. Check one ACOS figure from at least 14 days ago against the console; if it matches, add a write prompt next week, starting with prompt 9. When the wording settles, a team can save it as a Skill; see how to write an SOP and your first Skill. See the Xnurta MCP page or book a demo.
Next question: Which prompts cut wasted spend and grow your keywords?
FAQ
Yes, the prompts are plain language, so they work in any assistant connected to an Amazon Ads MCP, and read prompts need only read permissions. In ChatGPT, full MCP with write actions is in beta on Business, Enterprise and Edu plans as of October 2026, so check your plan first.
Yes, on read-only connections. Give each junior one connection per client ad account with only the three read permissions, so prompts 1, 2, 4 and 7 run in full and the rest return proposals. A senior who holds write on that client reviews the proposal and applies it.
Your files hold numbers Amazon doesn't have, and margin, purchase orders, budget plans and brand-term lists decide whether a result is good or bad. Without them, the assistant guesses a target ACOS or days of cover, and a guessed break-even point can make a profitable product look wasteful.
Run the harvest, budget and stock prompts (3, 6 and 9) weekly, since search terms and stock shift fast. Run the profit, brand-split and placement prompts (1, 7 and 8) monthly or after a price change. Run prompts 2 and 4 whenever a number moves unexpectedly.
Partly. Sponsored Display performance reads like any other ad type, so the profit, diagnosis, change-history, stock and budget prompts (1, 2, 4, 6 and 9) cover it. Sponsored Display has no search terms, so the harvest and brand-split prompts don't apply.
The assistant should say what's missing, and you supply it. Xnurta MCP covers about 15 months of ad and business data across 22 marketplaces, but not your costs, forecasts or market-wide search volume. Attach those as files, or narrow the question to what the data can answer.
Xnurta MCP
Xnurta MCP is for Xnurta customers only. Create your MCP connection directly in the Xnurta platform. New to Xnurta? Book a demo and we'll walk through your account with you.