Marketplace Ad Pros
Experiments Prompts
for Claude & ChatGPT
Run Marketplace Ad Pros experiments through your AI client. Review proposals, work the current action, log what you did, and wrap up with auto-generated results.
Click any prompt to copy
Full experiment lifecycle
MCP-aware prompts
New to Experiments?
Marketplace Ad Pros proposes optimization tests for your campaigns, you run them in Amazon, and our daily tracker measures the impact. Each experiment is one hypothesis — tied to a brand, a 14-day window, and a clear success metric — that you can review, start, work, and wrap up from your AI client.
1. Proposed
AI-generated test ideas waiting on your review. Start the ones worth running; dismiss the rest with a reason so future proposals improve.
2. Started
Running. You apply the plan in Amazon, the tracker captures a baseline and watches metrics daily, and your AI client drives the next step via the prompts below.
3. Complete
Done. The system writes up a results summary with conclusion, findings, before/after metrics, and suggested follow-up tests.
YouReview proposals, apply changes in Amazon, decide when to wrap up.
The platformProposes tests, captures the baseline, tracks impact daily, generates results.
Your AI client (MCP)Walks you through each step, pulls reports, logs what was done, completes the test on your go-ahead.
How to leverage them: Open Experiments daily and ask your AI client “what should I work on?” (prompt #01 below). The first 60 seconds tell you exactly what moves the needle that day. Experiments require a full-features plan (Two-Week Trial, Launch, Boost, or Dominion).
How to Use These Prompts
- Connect Marketplace Ad Pros to your AI client (Claude, ChatGPT, Claude Code, n8n, or any MCP-compatible tool) on the Integrations page.
- Install the Optimization Experiments Skill so your AI client knows the experiment lifecycle without re-explaining it each time.
- Click any prompt below to copy it, paste into your AI client, and replace
<EXPERIMENT_ID>with the experiment ID from your Experiments page when needed.
READ prompts only fetch data. WRITE prompts change experiment state — your AI client may ask you to approve each one before it runs.
Daily Check-In
Start every day with one of these. Pulls fresh `current_action` items the overnight tracker generated.
01
What's next
List my running experiments using get_experiments with state='started' and show_experiment_details=true. For each one, give me the name and the first sentence of current_action in one line. Then ask me which I want to work on.
READ One-screen daily summary of where every test stands.
02
Status of one experiment
Fetch experiment <EXPERIMENT_ID> with get_experiments(experiment_ids=['<EXPERIMENT_ID>'], show_experiment_details=true). Summarize where it stands: plan, current_action, most recent log entries, and any measurements. Then propose what we should do next.
READ Use when picking up a specific test mid-stream.
03
Brand-scoped check-in
List experiments for brand <BRAND_ID> using get_experiments(brand_id='<BRAND_ID>', show_proposals=true, show_experiment_details=true). Group them by state (proposed, started, complete) and summarize what's pending review, what's running, and what's wrapped up.
READ Multi-brand accounts: scope the daily summary to one client.
Review & Decide
Triage AI-generated proposals. Start the high-impact ones, dismiss the ones that don't fit.
04
Review pending proposals
Show me all proposed experiments with get_experiments(show_proposals=true, state='proposed', show_experiment_details=true). Sort by impact_score descending. For each one, give me the hypothesis and expected effort in one paragraph, then ask if I want to start it, dismiss it, or skip for now.
READ Weekly review of fresh AI proposals.
05
Start an experiment
Start experiment <EXPERIMENT_ID>. First, confirm I'm ready to apply the plan in Amazon (Marketplace Ad Pros doesn't push changes automatically). When I say yes, call start_experiment and then walk me through what I need to do in Amazon today.
WRITE proposed → started. Captures a baseline metric snapshot.
06
Start with a custom end date
Start experiment <EXPERIMENT_ID> with planned_end_date='<YYYY-MM-DD>'. Confirm with me first, then call start_experiment.
WRITE Override the default 14-day window — useful for short pulse tests or longer seasonal ones.
07
Dismiss a proposal
Dismiss experiment <EXPERIMENT_ID>. Ask me for the reason first (even a one-liner) so future proposals can avoid similar suggestions, then call dismiss_experiment with that reason.
WRITE Reasons feed the proposal generator — don't skip them.
Work the Action
Execute the current_action on a running experiment. Your AI does the actual work — pulls reports, makes changes (with your approval), reports back.
08
Walk me through the current action
Load experiment <EXPERIMENT_ID> with show_experiment_details=true. Walk me through the current_action one step at a time. For each step, do the work yourself if you can (call ask_report_analyst, list_resources, etc.), tell me exactly what you did and what you found, and call add_experiment_log after each step. When you're blocked on something I need to do in Amazon, stop and ask.
READ WRITE The core working prompt — also auto-rendered on each running experiment's detail page.
09
Pull metrics for an experiment
For experiment <EXPERIMENT_ID>, run ask_report_analyst for the date range start_date..today on the campaigns/profiles involved. Compare against the experiment's baseline (in the log/measurements) and tell me whether the test is tracking toward its success criteria, with specific numbers.
READ Mid-test sanity check before deciding to keep going, adjust, or stop.
10
Plan the Amazon changes
Read the plan and current_action for experiment <EXPERIMENT_ID>. List the specific changes I need to make in Amazon Advertising console (campaign IDs, bid changes, budget changes, negatives to add) as a checklist I can work through. Don't make any changes yourself — I'll apply them and report back.
READ For users who prefer to apply changes manually in Amazon's UI.
Logging
Record what happened. The tracker reads the log on its next run and uses it to update current_action.
11
Log a change I made
I just made these changes in Amazon for experiment <EXPERIMENT_ID>: <LIST_OF_CHANGES>. Call add_experiment_log with a concise summary of what I did and which campaigns it affected, then tell me what the tracker should look for tomorrow.
WRITE Records human-applied changes so the tracker can attribute impact correctly.
12
Log an observation
For experiment <EXPERIMENT_ID>, log this observation with add_experiment_log: <OBSERVATION>. Include enough context that someone reviewing the log next week understands why the observation matters.
WRITE Capture qualitative signal (creative refresh, competitor entered category, ranking shifted) that pure metrics miss.
13
Log a decision to deviate
I'm deviating from the plan on experiment <EXPERIMENT_ID>. Original plan: <ORIGINAL>. What I'm doing instead: <CHANGE>. Why: <REASON>. Log this with add_experiment_log so the tracker and the results write-up will know the test changed mid-stream.
WRITE Critical for honest post-test analysis — don't let deviations vanish from the record.
Wrap Up
Complete an experiment so the AI results generator writes up the conclusion, key findings, and suggested next experiments.
14
Complete an experiment
Complete experiment <EXPERIMENT_ID>. Call complete_experiment, then share the generated results (conclusion, key findings, metrics summary, suggested next experiments) with me verbatim — don't paraphrase.
WRITE started → complete. Triggers AI results generation.
15
Pre-completion review
Before I complete experiment <EXPERIMENT_ID>, summarize the full log and the latest measurements. Tell me whether the success criteria were met and whether you'd recommend completing now or extending the run. Don't call complete_experiment yet.
READ Sanity check before triggering the results write-up.
16
What worked this quarter
List completed experiments from the last 90 days with get_experiments(state='complete', show_experiment_details=true). For each one, give me the conclusion and the single biggest learning. End with three themes worth doubling down on next quarter.
READ Quarterly retro across the whole experiment program.
Ready to Run These?
Experiments require a full-features plan (Two-Week Trial, Launch, Boost, or Dominion). Start the trial and Marketplace Ad Pros begins proposing tests within 24 hours.
See PlansTwo-week trial $20 · Paid plans from $99/mo · Cancel anytime
Copied to clipboard