How to run Amazon Sponsored Ads with Claude: seven workflows

What do you need before you run Sponsored Ads with Claude?

You need Claude connected to Xnurta. Xnurta MCP connects Claude, ChatGPT and other MCP clients to your Xnurta data. In Claude desktop, Cowork or claude.ai, you add https://mcp.xnurta.com/mcp as a custom connector and sign in with your own Xnurta login. On Claude Team or Enterprise, an Owner or Primary Owner adds the connector once and each member then connects individually. Claude then sees only the ad accounts and permissions that login carries, main accounts and sub-accounts included.

Adding a connector never installs skills, so in a new chat ask Claude to install them from the GitHub repository. Exact screens are in Connect Claude or ChatGPT to Xnurta MCP.

Then check the connection with one question: "List the ad accounts I'm authorized to access." If your accounts come back, the connection works.

Can Claude change your campaigns?

Only if write access is on for your account. Reading never changes anything. Writing comes in three kinds: AI managed groups, edits to ads and campaigns, and creating Sponsored Products campaigns. Each kind has its own permission scope, so a scope that edits ads does not necessarily create campaigns.

Changes to money, bids or archiving never run straight away. Claude shows a preview first, with the objects and the old value next to the new one, and submits only after you confirm. For any other change, ask for the same preview. Archiving cannot be undone and campaigns cannot be deleted, so create new campaigns paused.

The table shows what each workflow needs. Workflows 5 and 6 need no write access at all, and the other five have a read-only half.

How do you find keywords and negatives with Claude?

You start from your own search-term data, which shows what shoppers typed before they bought. Claude can then turn the winners into a paused campaign and the losers into negatives, all in one conversation.

Workflow 1 harvests: converting terms that no exact-match target covers yet. Creating the campaign is all or nothing, so a failed validation never leaves a half-built campaign. Negating the source terms in your auto campaign is a separate edit to ads.

Workflow 2 cleans up two mistakes: terms that spend with no orders, and negatives that block terms which convert elsewhere.

For ad account [account name], pull search terms from the last [30] days across auto, broad and phrase campaigns. Show terms with at least [2] orders and CVR above [10]%, with clicks, spend, ACOS and orders. Flag the ones that are not yet targeted as exact match anywhere in the account. Then group them by theme.

Create a paused Sponsored Products manual campaign for ASIN [ASIN] with one ad group, exact match on the terms above at a [0.90] bid and a [30] daily budget, and negate [term 1], [term 2] and [term 3]. Show me the full structure before you submit.

In ad account [account name], find search terms from the last [30] days that spent more than [amount] with zero orders and are not negated. Separately, list terms that are negated but produced orders elsewhere in the account.

Add the first list as negative exact at the campaign level. Show me a preview first with every term, campaign and match type, and wait for my confirmation.

How do you change bids and budgets with Claude?

You give Claude one rule per request, check the preview, then confirm. One rule keeps the preview short enough to read properly.

Workflow 3 handles bids. Ask Claude to show the target-bid formula (average order value times conversion rate times target ACOS) so every proposed bid stays checkable. Workflow 4 handles budgets and placements: find efficient campaigns that run out of money, raise their budgets, then adjust the top-of-search modifier. Hourly campaign data shows how spend builds through the day.

In ad group [ad group name], list keywords with ACOS above [40]% and at least [20] clicks in the last [14] days. For each, calculate a target bid as average order value times conversion rate times a [25]% target ACOS. Show current bid, proposed bid and percent change, and cap any single change at [30]%.

In ad account [account name], which SP campaigns hit their daily budget cap on [3] or more of the last [7] days while ACOS stayed below [30]%? Show budget, days capped, the hours the cap was reached and ACOS.

Raise each of those budgets by [20]% and show me the old and new budget for each campaign first. Then compare top-of-search and rest-of-search ACOS over [30] days for the same campaigns and suggest a top-of-search modifier. After I confirm, set it to [25]%.

How do you check true profit and who changed what?

For profit, you hand Claude your unit margins and it works out profit after ad spend, campaign by campaign. For history, you ask the operation log. Neither needs write access.

Workflow 5 answers a question the platform cannot, because your margins live in your own files. Break-even ACOS equals your margin before ads, so a product with a 30% margin breaks even at 30% ACOS. Xnurta MCP also returns TACOS, total sales, sessions and Featured Offer (Buy Box) percentage next to the ad metrics. To bring in more of your own data, see Cross-analyze Amazon Ads with your own data in one prompt.

Workflow 6 asks what changed. The log holds human and AI actions on campaigns, ad groups, targets, bids, budgets and managed groups. You can filter by operator: AI, human, automation rule or one person's email. The full filter list is in Who changed my Amazon campaign?.

Here is the unit margin before ad spend for each ASIN in product line [name]: [paste table]. Pull the last [30] days of ad spend and ad-attributed sales by campaign. Calculate profit after ad spend as attributed sales times margin minus ad spend. Which campaigns lose money, and should I cut, fix or scale each one?

List every bid, budget and status change on campaign [campaign name or ID] in the last [14] days, with the operator, the action type and the time. Did performance shift after any of them?

How do you steer a whole category with an AI managed group?

You set the goal for a group of campaigns and leave the day-to-day adjusting to the AI inside that group. This is Workflow 7, and it is the one with the most guardrails to set.

You control the levers. A managed group has an action space, a set of switches for placement bid adjustment, dayparting bid adjustment, performance-based bid adjustment, budget dynamics, negative targeting, target harvesting and structure optimization. Through MCP you can set the group's name, AI status, personality, goal, target ACOS, total budget, campaign membership, action space on or off, schedule and template. Brand, non-brand and competitor mode, and every word-list feature, stay in the Xnurta platform. Some changes may be skipped while AI is running on a group. Deleting a managed group archives it, and that cannot be undone.

Keeping one group per category lets you change one category's goal without touching the others. Read and change Xnurta's AI levers through MCP covers each lever.

Show every AI managed group in ad account [account name] with its target ACOS and AI status, and the AI-managed ACOS of its campaigns over the last [14] days. Which groups are furthest from target?

For managed group [group name], lower target ACOS from [30]% to [26]% and turn on target harvesting. Before you submit, show me the group, target ACOS, AI status, action space and campaign list you plan to change.

For managed group [group name], turn off budget dynamics and keep placement bid adjustment and dayparting bid adjustment on. Show me the action space before you submit, then confirm the AI status afterward.

How do you turn the seven workflows into a weekly routine?

Run them in the same order each week and review category goals once a month.

• Monday: Workflow 6 (what changed), then Workflow 5 (profit).
• Tuesday: Workflow 2 (negatives), then Workflow 1 (harvest).
• Wednesday: Workflow 3 (bids), then Workflow 4 (budgets).
• Once a month: Workflow 7 (category goals).

The official Weekly Ads Report skill gives you the weekly overview in one request: say "Give me last week's ads report" and you get week-over-week KPIs, 7-day trends, anomalies, AI and automation activity, top movers and next-week actions. You can also save your own routine as a skill, private or shared with your organization. Ten questions to ask your Amazon Ads data every Monday has a full Monday set.

What makes Claude's answers more reliable, and where are the limits?

Name the object, the time window and the output you want. Five habits do most of the work:

1. Name the ad account, campaign, ad group or managed group in the question.
2. Give a time window and a comparison, such as the last 14 days against the prior 14.
3. Say what you want back: a table, a top 10, a one-line takeaway or a CSV.
4. Ask follow-ups in the same chat. Start broad, then ask why, then drill in.
5. Confirm before any write, and read the old and new values before you say yes.

The limits are simple. Xnurta MCP holds about 15 months of history across 22 marketplaces. Performance is daily by default and hourly for SP, SB and SD campaigns and SP keywords, and the hourly data comes from Amazon Marketing Stream. Freshness follows the platform's update cadence, not the second. Keyword research starts from your own search terms, so a brand-new category with no history gives Claude little to work with. For what read-only access can and cannot do, see Read-only vs read-write MCP for Amazon advertisers.