Convert MCP Server — Prompt Templates

Summarize all experiments in project [PROJECT_ID] from the last [TIME_PERIOD, e.g., "90 days" or "Q1 2026"].

For each experiment, show:
- Name
- Status (active, paused, completed)
- Type (A/B, MVT, Split URL)
- Duration (start → end or start → today)
- Primary goal and whether it reached statistical significance
- Winning variation (if applicable) and observed lift
- What action was taken as a result of the experiment being run

Then provide a brief overall summary: how many tests ran, how many reached significance, and what the overall win rate was.

Pull all completed (statistically significant) experiments in project [PROJECT_ID] from the last [TIME_PERIOD].

Group them by:
1. Page or funnel stage tested (e.g., homepage, product page, pricing page, checkout)
2. Type of change (copy, layout, CTA, social proof, pricing, navigation)
3. Outcome (significant win, significant loss, inconclusive)

Then identify:
- Are there patterns in what types of changes tend to win on specific pages?
- Are there patterns in what consistently fails?
- Which page or funnel stage has the highest density of inconclusive results (possible signal of insufficient traffic or weak hypotheses)?

Present findings as a pattern analysis, not just a data table.

Don't fill in gaps with guesses. Only rely on actual experiment results.

Show me all experiments in project [PROJECT_ID] from the last [TIME_PERIOD] that either:
- Lost (statistically significant negative result), or
- Were inconclusive (did not reach significance)

For each, show the experiment name, what was being tested, the primary goal, the observed effect (even if not significant), and how long the test ran.

Then analyze:
- Were any inconclusive tests underpowered (ran for too short a time or had too little traffic)?
- Do the losing tests share common themes (e.g., all involved removing content, all targeted mobile, all changed pricing)?
- Are there tests that showed a small positive directional trend but didn't reach significance — candidates for re-testing with more traffic or a refined hypothesis?
- Review segments available. Did any tests reverse in trend (results) for specific segments?

For experiment [EXPERIMENT_NAME_OR_ID] in project [PROJECT_ID]:

Break down results by the available audience segments. Show conversion rate, lift vs. control, and significance status for each segment.

Specifically, I want to know:
- Did mobile and desktop audiences respond differently?
- Were there segments where the variation won but others where it lost or was flat?
- Is the overall result being driven by one dominant segment, or is it consistent across segments?

If post-segmentation data is limited, tell me what's available and what isn't.

Pull all experiments from project [PROJECT_ID] over the last [TIME_PERIOD, e.g., "6 months"].

For each experiment, show:
- The experiment name and description (which typically contains the hypothesis)
- The primary goal
- The outcome (win/loss/inconclusive)
- The observed lift

Then assess:
- What percentage of experiments had clear, specific hypotheses in their descriptions vs. vague ones (e.g., "test new homepage" vs. "adding urgency messaging above the fold will increase add-to-cart rate by 5% for mobile users")?
- Is there a correlation between hypothesis specificity and test success rate?
- Which hypotheses led to the largest lifts — what do they have in common?

I need to compare experiment performance before and after [EVENT, e.g., "our site redesign on March 15, 2026"].

For project [PROJECT_ID]:
- Pull all completed experiments from [DATE_BEFORE_START] to [DATE_BEFORE_END] (the "before" period).
- Pull all completed experiments from [DATE_AFTER_START] to [DATE_AFTER_END] (the "after" period).

Compare:
- Win rate (% of tests reaching statistical significance with positive lift)
- Average observed lift in winning tests
- Average test duration to reach significance
- Types of tests being run (did the testing strategy shift after the change?)

Highlight any notable differences and flag if sample sizes make comparison unreliable.

Show me all currently active experiments in project [PROJECT_ID].

For each, show:
- Name, start date, days running
- Traffic allocation (and whether it looks balanced — flag any potential sample ratio mismatch)
- Conversion rate per variation so far
- Whether the test is trending toward significance or appears flat
- Any experiments that have been running for more than [THRESHOLD, e.g., "30 days"] without reaching significance

Flag any experiments I should consider pausing or extending, and explain why.

Create a new A/B test in account [ACCOUNT_ID], project [PROJECT_ID] with the following configuration:

- Name: [EXPERIMENT_NAME]
- Description: [HYPOTHESIS_OR_BRIEF]
- URL: [TARGET_PAGE_URL]
- Status: draft
- Traffic distribution: [PERCENTAGE, e.g., "100"]

Variations:
1. Control
   - Name: "Control"
   - Is baseline: true
   - Traffic distribution: [e.g., "50"]
   - Code: (no changes)

2. Variation 1
   - Name: "[VARIATION_NAME]"
   - Is baseline: false
   - Traffic distribution: [e.g., "50"]
   - JS: [PASTE_OR_REFERENCE_JS_CODE]
   - CSS: [PASTE_OR_REFERENCE_CSS_CODE]

Goals: [LIST_GOAL_IDS, e.g., "1002467, 1002468"]
Primary goal: [PRIMARY_GOAL_ID]
Audiences: [LIST_AUDIENCE_IDS]
Locations: [LIST_LOCATION_IDS]

For project [PROJECT_ID] in account [ACCOUNT_ID]:

1. List all available goals. Show each goal's ID, name, and type.
2. List all available audiences. Show each audience's ID, name, and targeting conditions.
3. List all available locations. Show each location's ID, name, and URL pattern.

I need these IDs to configure a new experiment.

Update experiment [EXPERIMENT_ID] in project [PROJECT_ID]:

Changes to make:
- [DESCRIBE CHANGES, e.g., "Change traffic split to 70/30 (control/variation)"]
- [e.g., "Replace the variation JS with the following code: [PASTE_CODE]"]
- [e.g., "Add goal ID [GOAL_ID] as a secondary goal"]
- [e.g., "Change status from draft to active"]

Confirm the changes before applying.

For project [PROJECT_ID], show me all experiments currently in "draft" status.

For each, show:
- Experiment name and ID
- Number of variations and their names
- Traffic split across variations
- Goals attached (list names and IDs)
- Audiences attached
- Locations (URL targeting)
- Whether global JS or global CSS is set

Flag any experiments that are missing:
- Goals
- Audiences
- Locations
- Or have only one variation (no control or no challenger)

For experiment [EXPERIMENT_ID] in project [PROJECT_ID]:

Show me the full configuration including:
- All variation names, IDs, and their JS and CSS code
- Global JS and global CSS (if any)
- Targeting conditions (audiences and locations)
- Goals and primary goal

I want to verify this matches what's in our local codebase.

In project [PROJECT_ID], find all active experiments that meet ANY of these criteria:

1. Running for more than [DAYS, e.g., "45"] days without reaching statistical significance
2. Showing a statistically significant negative lift on the primary goal
3. Have a sample ratio mismatch (traffic split is off by more than 2% from the configured allocation)

For each matching experiment, show me the name, days running, current results, and the issue detected.

Then ask me to confirm before pausing any of them.

Search the Convert knowledge base for: [YOUR QUESTION]

Examples:
- "How do I set up revenue tracking for a Shopify store?"
- "What's the difference between sequential testing and fixed-horizon in Convert?"
- "How do I configure cross-domain tracking?"
- "What are the body schema requirements for creating an experiment via the API?"

For project [PROJECT_ID], generate an experimentation velocity report for [TIME_PERIOD]:

Include:
- Total experiments created
- Total experiments that went live
- Total experiments completed
- Average time from creation to launch
- Average time from launch to conclusion (significance or manual stop)
- Experiments currently in draft (backlog size)
- Experiments currently active

Compare to the previous [TIME_PERIOD] if data is available. Are we testing more or less? Faster or slower?

Give me a full audit of project [PROJECT_ID] in account [ACCOUNT_ID]:

1. Project settings: name, domain(s), active integrations
2. Experiments: total count by status (draft, active, paused, completed, archived). List the 5 most recently active.
3. Goals: total count, list all with names and types
4. Audiences: total count, list all with names and brief targeting descriptions
5. Locations: total count, list all with URL patterns
6. Any feature flags configured

I'm getting oriented on this project — give me the lay of the land.

For project [PROJECT_ID] over the last [TIME_PERIOD]:

Give me the following metrics:
- Total experiments completed
- Win rate (% of tests with statistically significant positive lift)
- Average lift in winning experiments
- Average test duration (days)
- Most-tested page or funnel stage
- Most common experiment type (A/B vs. MVT vs. Split URL)

I'll use these as internal benchmarks to compare against industry data.