30 Claude Prompts for Amazon PPC That Use Live Account Data

Quick answer: The best Claude prompts for Amazon PPC do five things: name the account and date range, set a numeric threshold, ask for a fixed output, forbid changes, and demand the rows behind every total. This guide gives you 30 of them, grouped by job, each with a one line check so you can tell a good answer from a confident guess. They assume Claude is connected to your live ad account, so answers reflect the account today and not a file you exported last week.

Key Takeaways

• Thresholds beat adjectives. “Find wasted spend” gets a different answer every time. “Find terms with more than 3 divided by the conversion rate in clicks and zero orders” gets the same answer every time. This guide gives you the number for each prompt.

• Learn three numbers first. Break even ACOS, maximum CPC and the zero order click rule drive most of the prompts below. The worked example takes two minutes.

• Every prompt has a check. A two minute check of three totals against Campaign Manager catches most wrong answers before they reach a client or a bid.

• Prompts do not overlap. Each of the 30 answers a different question, so you can run them as a routine instead of a pile.

• Read first, write last. Twenty seven prompts only read. The last three draft and review changes, and none of them apply anything.

Live Data vs Pasted Exports: What Changes in Your Prompts

Most prompt lists for Amazon PPC start with “download your search term report and upload it to Claude.” That works, but it changes what a prompt can safely ask. A live connection changes four things.

The last row matters. With an export, the risk is old data. With a live connection, the risk moves to the question. A vague prompt on live data does not fail visibly. It answers fast and sounds right.

To connect an assistant, Amazon describes its own server as available globally in open beta to Amazon Ads partners with active API credentials, so most sellers and agencies reach it through a platform that is a partner. Anthropic’s help article on custom connectors using remote MCP explains how a connector is added in Claude. Our guide to using Amazon MCP with Claude walks through the setup step by step.

The Anatomy of a Prompt That Does Not Guess

Every prompt in this guide has the same five parts. Once you see them, you can write your own.

Figure 1: The five parts of a prompt that stays inside the data.

1. Scope. Account, marketplace and currency. “Brand X, US, USD.”

2. Window. Exact dates, not “recently.” “The last 30 full days, with the most recent 14 marked provisional.”

3. Threshold. A number with a reason. “At least 3 divided by the ASIN’s conversion rate in clicks.”

4. Output. A fixed shape. “A table with campaign, term, clicks, spend, orders and reason.”

5. Guardrail. What it must not do. “Never apply changes. If data is missing, say so and stop.”

The context block to paste once per conversation

Paste this at the start of a chat, fill the brackets once, and every prompt below inherits it.

Account: [brand or client]. Marketplace: [US]. Currency: [USD]. Break even ACOS: [25%]. Target ACOS: [20%]. Spend floor: [$50 per campaign in the period]. Stock gate: [14 days of cover]. Use live account data only. State the date range, marketplace and number of rows behind every answer. Label the most recent 14 days provisional because attributed sales keep updating. Never apply changes. If a number needs data you cannot see, say what is missing instead of estimating.

The 14 day provisional rule is our conservative default, not an Amazon rule. Shorten it if you have confirmed your account’s attribution settle time.

Three Numbers to Know Before You Run Any Prompt

Most weak PPC prompts fail because Claude has no break even number to compare with. These three fix that.

1. Break even ACOS = contribution margin before ads ÷ selling price

A product sells at $30. After cost of goods, Amazon fees and other per unit costs, $7.50 is left before ads. Break even ACOS is $7.50 ÷ $30 = 25%. An ACOS above 25% loses money on each ad order, unless you are buying rank on purpose.

2. Maximum CPC = price × conversion rate × break even ACOS

With a 10% conversion rate on that product: $30 × 0.10 × 0.25 = $0.75. A bid that averages above $0.75 on this product cannot be profitable at that conversion rate, whatever the keyword sounds like.

3. The zero order click rule: clicks needed = 3 ÷ conversion rate

Most guides say “negate terms with 10 or 20 clicks and no sales.” That cutoff is wrong for both a 3% and a 20% converting product. A better test asks how many clicks you would expect before getting three orders, then asks how likely it is that a term with that many clicks has zero orders by bad luck.

At a 10% conversion rate, 3 ÷ 0.10 = 30 clicks. The chance of zero orders in 30 clicks, if the term truly converted at 10%, is 0.90 to the power of 30, or about 4%. The same logic gives about 5% at 60 clicks for a 5% converting product and about 4% at 20 clicks for a 15% converting product. So the rule means roughly “less than a 5% chance this is just bad luck.”

Figure 2: The zero order click rule. Waiting for 3 ÷ conversion rate clicks leaves about a 5% chance of a false alarm.

Use your own conversion rate for the ASIN being advertised, not an account average. A $90 product and a $9 product should not share a cutoff.

The 30 Prompts at a Glance

Figure 3: Seven groups, 30 prompts. Blue groups need ad data only. The green group needs retail signals such as stock and Buy Box.

Group E needs more than ad data. If your connection only sees campaigns and reports, Claude will either say so or guess, which is exactly why each prompt below tells it to stop and say what is missing.

Group A: Diagnose What Changed (Prompts 1 to 5)

1. Weekly account pulse

Compare the last 7 full days with the previous 7 and with the same 7 days four weeks earlier. Show spend, ad sales, ACOS, TACOS, orders, CPC and conversion rate. List the three largest movements and tag each as ads, stock, price, Buy Box or seasonality. Mark the most recent 14 days provisional.

Check: Spend and ad sales match Campaign Manager for the same dates.

2. ACOS driver split

ACOS moved from [last period] to [this period]. Split the change into the effect of CPC, conversion rate and average order value. Rank the three largest drivers and show the numbers behind each.

Check: CPC divided by conversion rate, divided by average order value, reproduces the ACOS shown.

3. Sales drop root cause

Total sales fell [x%] between [period A] and [period B]. Compare ad spend, ad sales, organic sales, price, stock, Buy Box and page views. Tell me which metrics did not change so I can ignore them, and which two changed most.

Check: The cause named should be visible in at least one source you can open, such as the Business Report.

4. TACOS against ACOS

For the top 10 ASINs by ad spend, show ACOS and TACOS by week for the last 12 weeks. Flag ASINs where ACOS is flat but TACOS is rising, and ASINs where TACOS is falling while ad spend rises.

Check: Pick one ASIN and recompute TACOS as ad spend divided by total sales. Our guide to TACoS explains how to read the gap.

5. Break even gap

For the top 10 ASINs by ad spend, show ACOS next to break even ACOS and mark each as above or below. Say where each break even number came from. If cost data is missing for an ASIN, mark it unknown instead of estimating.

Check: Break even ACOS equals your own margin sheet, not a round guess.

Group B: Find Wasted Spend (Prompts 6 to 10)

6. Zero order terms by the click rule

For each advertised ASIN, take its conversion rate over the last 30 days. List search terms with more than 3 divided by that conversion rate in clicks and zero orders. Show campaign, term, clicks, spend and the click cutoff used for that ASIN.

Check: Open one listed term in the search term report and confirm clicks and zero orders.

7. Safe negatives

From the terms in prompt 6, mark which are safe to add as exact negatives. Exclude terms that contain my brand, my own ASINs or a close variant, terms that had orders in the previous 90 days, and terms whose intent matches the product. Give a reason for each exclusion. Do not apply anything.

Check: No brand term or recent converter is on the negate list.

8. Overlapping search terms

Find search terms that received spend in more than one campaign in the last 30 days. For each, show spend, orders and ACOS per campaign and recommend which campaign should keep it.

Check: One overlapping term, traced by hand, shows the same split.

9. Auto campaign match types

For each auto campaign, split spend, orders and ACOS by close match, loose match, substitutes and complements. Mark each match type as above break even, between target and break even, or below target.

Check: The four match types add up to the auto campaign total.

10. Cost of waste

Total the spend on the terms found in prompt 6. Show it as a share of total ad spend for the same 30 days and as a monthly figure. List the five biggest line items.

Check: The five items plus the rest equal the total from prompt 6.

Group C: Keywords and Targets (Prompts 11 to 15)

11. Harvest candidates

List search terms from auto and broad campaigns with at least 2 orders in the last 30 days and ACOS below my target, that are not already exact match keywords anywhere in the account. Show orders, clicks, ACOS and the campaign each came from.

Check: Confirm one term is not already an exact keyword in another campaign.

12. Harvest without cannibalizing

For each term from prompt 11, say which exact match campaign it should join and which negative to add in the source campaign so the two do not compete. Draft the changes as a table. Do not apply anything.

Check: Every harvested term has a matching negative in its source campaign.

13. Keywords where you already rank

If my connection includes organic rank, list keywords where the ASIN ranks in the top 3 organically and still receives ad spend above the spend floor. Show spend, orders and the share of orders that came through ads. If organic rank is not available, say so and stop.

Check: One keyword, searched on Amazon, shows the rank Claude reported.

14. Starved targets

List targets with impressions in the bottom 20% of their ad group, a bid below the suggested range, and ACOS below target where they did convert. These may be efficient targets that are not getting enough exposure.

Check: Impressions and bid match the target view in Campaign Manager.

15. Product targeting by ASIN

For product targeting, list targets above the spend floor with ACOS over break even, and targets under target ACOS with fewer than [500] impressions. Show whether each targeted ASIN is currently in stock if that data is visible.

Check: Targeting a competitor ASIN that is out of stock should show as a flag, not a win.

Group D: Bids, Placements and Budgets (Prompts 16 to 20)

16. Bids above the profit ceiling

For each top ASIN, calculate maximum CPC as price times conversion rate times break even ACOS. List keywords and targets whose average CPC is above that ceiling, with the gap in dollars and the clicks over the last 30 days.

Check: Redo one calculation by hand using the numbers in Claude’s table.

17. Small bid steps

Draft bid changes of at most 10% for targets that are above break even ACOS with enough clicks to judge, and at most 10% upward for targets below target ACOS that lose impressions. Show current bid, proposed bid, and reason. Do not apply anything.

Check: No proposed change is larger than 10%.

18. Placement skew

For each campaign above the spend floor, split spend, orders and ACOS by top of search, rest of search and product pages. Flag campaigns where one placement explains most of the waste or most of the profit, and say how it distorts campaign ACOS.

Check: Placement rows sum to the campaign total.

19. Budget caps

Which campaigns with ACOS below target ran out of budget on at least 3 of the last 7 days? If hourly data is visible, show the hour budgets ran out. Estimate missed orders as the campaign’s recent orders per hour times the hours it was capped, and label it an estimate.

Check: The campaign shows a budget cap notice in Campaign Manager for those days.

20. Budget reallocation, flat total

Propose moving budget from campaigns above break even ACOS to campaigns below target ACOS, keeping the total budget unchanged. Show current budget, proposed budget and the ACOS that justifies each move. Do not apply anything.

Check: Total proposed budget equals total current budget.

Group E: Retail Aware Checks (Prompts 21 to 25)

These prompts need stock, price or Buy Box data. Our guide to what Amazon MCP cannot see explains why a connection with ad data only will get them wrong.

21. Stock gate

Which advertised ASINs have fewer than [14] days of cover at their last 14 day sales rate? Show units available, units inbound, daily units and days of cover, plus the ad spend in the last 7 days. If inventory is not visible to you, say so and stop.

Check: Units available match Seller Central.

22. Buy Box loss

Which advertised ASINs did not hold the Buy Box on at least one of the last 7 days? Show the days affected and the ad spend and clicks on those days. If Buy Box status is not visible to you, say so and stop.

Check: Spot check one ASIN on the product page.

23. Price change and conversion

Which advertised ASINs changed price in the last 14 days? Compare conversion rate for 7 days before and 7 days after each change, and say whether ad spend or clicks changed. Tag any ASIN where conversion moved by more than 20% in either direction.

Check: Price history matches your own records.

24. Effective cover for a promotion

For the event starting on [date], effective cover is units available divided by daily units times expected uplift of [2]. List ASINs whose effective cover is shorter than the event plus [7] days of lead time buffer.

Check: Recalculate one ASIN by hand.

25. Suppressed or inactive listings

If listing status is visible to you, list advertised ASINs that are suppressed or inactive, with spend over the last 7 days. If it is not visible, say so and stop.

Check: Open one ASIN in Seller Central.

Group F: Reporting and Client Updates (Prompts 26 to 28)

26. Five block client report

For [client], marketplace [US], last full week against the four week median, draft a report in five blocks. One: headline numbers against target. Two: what changed and why, each movement tagged as ads, stock, price, Buy Box or seasonality. Three: what we did this week. Four: what we propose, marked awaiting approval. Five: data notes, including provisional days and anything you could not verify.

Check: Block five lists at least the provisional days.

27. Portfolio exception scan

Across all accounts I can see, flag any account whose weekly ACOS differs from its own four week median by 20% or more and whose weekly spend is above [spend floor]. Rank by spend. Show the median, this week and the change, and list accounts below the floor separately.

Check: Each flagged account is visible in its own console and the change is real.

28. One paragraph plan

Write one paragraph for [client] covering the biggest win, the biggest risk and the one change we propose next week, using only numbers already shown in this conversation. Do not introduce new figures.

Check: Every number in the paragraph appears earlier in the chat.

Group G: Draft and Review Changes Safely (Prompts 29 and 30)

29. Change list with a way back

Draft the changes from this conversation as a table with entity, current value, proposed value, reason, expected effect, risk and the value to restore if it fails. Group by campaign. Do not apply anything.

Check: Every row has a restore value.

30. Review after the change

Compare 7 days before and 7 days after the changes made on [date]. Separate the effect of the change from seasonality by comparing the same weekdays a year ago, if available. For each change recommend keep, adjust or roll back, and say how sure you are.

Check: The date range excludes provisional days or labels them.

How to Check Any Answer in Two Minutes

A confident answer is not a correct one. Run this check on any output before you act on it.

1. Echo. Does the answer state the account, marketplace, currency and date range? If not, ask it to.

2. Three totals. Pick spend, orders and ACOS. Compare them to Campaign Manager for the same dates.

3. One row. Open one listed term, campaign or ASIN and confirm it.

4. The rows behind it. Ask for the rows behind any total, and confirm they add up.

Figure 4: The answer check loop. Nothing reaches a bid or a client until it passes.

If any step fails, do not fix the prompt first. Check the connection, the date range and the client. Most failures there come from scope, not wording.

A Weekly Rhythm for the 30 Prompts

You do not run all 30 at once. A realistic rhythm looks like this.

Prompts That Look Smart but Fail

• “Optimize my account.” There is no threshold, window or output. You will get plausible advice that cannot be checked.

• “What is my profit?” Claude cannot see your cost of goods unless you provide it. Prompt 5 forces it to say where the break even number came from. The same blind spot applies to margin and break even.

• “Show me what my competitors spend.” Amazon does not publish competitor ad spend, so any number is invented.

• “Fix everything you find.” Applying a batch of changes in one step turns a small misread into an account wide incident. Use prompts 17, 20 and 29 to draft, then approve one change group at a time.

• “Use the last few days.” The most recent days are still changing, and a short window gives noisy conversion rates. Name the dates.

Keep Writes Behind an Approval Step

Twenty seven of these prompts only read. If your connection can write, treat that as a separate permission. Anthropic’s help article says to review tool approval requests carefully, use the tool menu to disable tools you do not want invoked, and be careful with “Allow always.” It also says that in Research mode Claude can call connector tools without further approval, and advises disabling tools that can take write actions. For an ad account, that means turning off write tools before running any broad research prompt.

A simple order works well. Start read only. Add drafted changes next, with prompts 17, 20 and 29. Allow writes last, one action at a time, with a log of old and new values. Amazon’s Agent Policy took effect on March 4, 2026, and trade coverage says it is unclear whether the policy reaches Amazon Ads APIs or MCP, so confirm the current text with your provider.

How Atom11 Helps With Live Account Prompts

These prompts work best when the connection reads live from the account, sees retail signals next to ads and keeps clients apart. Atom11 is an Amazon Ads Partner platform, and its MCP is built for exactly that.

• Live, not exported. The Atom11 MCP reads live from the account, so answers reflect the account when you ask. Setup takes about five minutes through the connector settings of your assistant, with no terminal and no data pipeline.

• Your existing permissions. Users sign in with their Atom11 credentials, and an AI tool can access only what that user can already access in the platform. Keys can be revoked from settings.

• Retail signals beside ads. Prompts 21 to 25 need stock, price and Buy Box. Atom11 brings ads, inventory, pricing and Buy Box into one place, which is why those prompts can come back with answers instead of a request for data.

• The logic behind Neo. Neo compares periods across more than 15 ad and retail signals to explain why sales moved, and the MCP exposes the same logic. That is the engine behind prompts 1 and 3.

• Writes with confirmation. Atom11 MCP offers write access for adjusting bids and budgets with a confirmation step on the same endpoint, which fits the approval rule above.

• Agencies. Whether you manage one brand or a portfolio of clients, the connection is the same, so prompt 27 can scan every account you can see.

Atom11 PPC pricing is tiered by monthly ad spend, starting at $499 a month under $50K, and the MCP is included in every tier. Product costs matter for prompts 5 and 16, so ask how cost data enters any system, Atom11 included, before you rely on a margin answer.

Want to run these prompts on your own account? Book an Atom11 demo and we will walk through them live.

Conclusion

The difference between a useful Claude answer and a confident guess is almost never the model. It is the prompt: a clear account, exact dates, a threshold with a reason, a fixed output and a rule against changing anything. Learn the three numbers, break-even ACoS, maximum CPC, and the zero-order click rule, and most of the 30 prompts above write themselves.

Start small. Paste the context block, run prompts 1, 21 and 22 on Monday, check three totals against Campaign Manager, and only then move on to waste and bids. When the answers hold up for a month, add drafted changes. When those hold up, consider writes with an approval step. A live connection gives Claude the data, and these prompts give it the discipline to use that data well.