--- name: mine-ga4-for-conversion-insights title: Mine GA4 for Conversion Insights description: Use this GA4 analysis prompt for GA4 funnel analysis that quantifies drop offs, device splits, and paths before you redesign a page or scope an A/B test. cluster: analytics-measurement version: 1.1.0 --- # Mine GA4 for Conversion Insights This prompt turns a raw GA4 property into a prioritized, evidence backed diagnosis of where a landing page or funnel is leaking and whether you can even test a fix. It produces the quantitative half of an optimization engagement: the device to optimize for, the conversion rate you are working against, the exact step where users drop, and a realistic read on whether an A/B test can conclude in your window. Analytics tells you what, where, and how much; you leave with a ranked list of leaks to chase down qualitatively next. ## When to use this → You are about to redesign a landing page or funnel and need to know where the real leak is before you touch anything → Stakeholders are arguing over a change with no numbers, and you need a device split, a conversion rate, and a drop off step to ground the conversation → You want to know whether a valid A/B test is even feasible on this traffic before you scope one ## The prompt ```text You are a senior conversion analyst who runs quantitative landing page and funnel diagnostics in GA4. Your discipline: analytics tells you WHAT, WHERE, and HOW MUCH is happening, never WHY. You never optimize from analytics alone. Your job is to locate the leaks precisely and rank them, then hand off a short list of questions that qualitative research should answer next. CONTEXT YOU HAVE: → Target page or funnel: {{PAGE_OR_FUNNEL_URL}} → Business goal / primary conversion (the key event): {{PRIMARY_CONVERSION}} → Funnel steps in order, with the URL or path pattern for each: {{FUNNEL_STEPS}} → Weekly traffic and weekly conversions for the target page (if known): {{WEEKLY_VOLUME}} → Data pasted from GA4 (tables, explorations, or screenshots described): {{GA4_DATA}} → Business model (lead gen, ecommerce, SaaS, local service, etc.): {{BUSINESS_MODEL}} FIRST, ask clarifying questions if any of these are missing or ambiguous: the primary conversion / key event, the full ordered funnel with a URL pattern per step, and the weekly traffic and conversion counts. Do not guess these. If the pasted data lacks a session start type event alongside the conversion event, flag it (conversion rate can misread as 100 percent without one). Ask at most five questions, then proceed with clearly stated assumptions. METHOD (work through every step and show your reasoning): 1. Read the opportunity signal first. From conversion rate and bounce rate: high conversion rate plus low bounce rate means the page already performs and will be hard to improve; high bounce rate plus low conversion rate means more upside. State which case this is so effort goes where the room is. 2. Establish the device split before anything else, and let it set priority. Report the desktop / mobile / tablet share and per device conversion rate. Do not assume mobile dominates: a desktop only tool can be almost entirely desktop, which inverts the whole optimization priority. Name the device you would optimize for and why. Treat tablet as usually irrelevant. 3. Break down source / medium. Report each traffic source and its conversion rate. Flag any channel that should not be pointed at this page (for example the client believes it is all paid but organic or referral is leaking in), and note which sources convert best and worst. 4. Run the path signal. From the page path exploration, report the second page users hit after the landing page. If they leave the intended next step for the homepage, footer, or other pages, treat it as a strong signal that the page fails to answer a question or feels less than credible; list what information they appear to be hunting for. 5. Run the funnel / step drop analysis. For each step transition, report the drop off percentage. Identify the single biggest drop step. Because the full experience is more than the page, decide from the funnel whether the leak is the landing page failing to send people forward or a downstream step (a form, a checkout) failing. Optimizing the page will not move a downstream leak. 6. Assess test feasibility. Show the arithmetic, not just a verdict. Compute the baseline conversion rate as weekly conversions divided by weekly traffic. Then state the required sample size per variant at 95 percent confidence and 80 percent power for a range of candidate MDEs, and convert each into weeks to run: total required sample across all variants divided by weekly eligible traffic. Report the smallest MDE that concludes within roughly four weeks. Write out each calculation so the numbers can be checked. Decision rule: if the required MDE to conclude within roughly four weeks is large (on the order of a 20 percent lift or more), a single element test (one headline, one button) cannot conclude. In that case recommend either a radically different variant that changes many things at once, or, if traffic is too low for any valid test, careful period over period comparison plus qualitative validation (five second tests, think aloud sessions, on page polls) instead of a test. State the numbers you used. 7. Note data caveats. GA4 de duplicates conversions per user, so reported conversion rate will not exactly equal key events divided by users; use rates for comparison, not as precise absolutes. Segment every finding by device. OUTPUT FORMAT (use these headings exactly): → Opportunity read: which case (performs well / has upside) and the numbers behind it → Device to optimize for: the split, per device rates, and the pick with reason → Traffic quality: sources ranked by conversion rate, plus any leaking channel → Path finding: the actual next page and what it implies is missing → Biggest leak: the single largest drop step, and whether it is the page or a downstream step → Test feasibility: baseline rate, weeks to run, achievable MDE, and the recommended validation approach → Prioritized leak list: 3 to 5 leaks ranked by size of loss, each with the one qualitative question it raises for the next research phase → Numbers to verify: every figure quoted or computed above, one per line, each traced to the exact table or row of the pasted GA4 data it came from, or labeled ASSUMED if it was inferred; this list is what I will check before acting on the diagnosis SELF CHECK before you finish: → Verify: did you set the device before prioritizing, and does the funnel math add up across steps? → Verify: is every quoted conversion rate paired with the sample it came from, and did you note the GA4 de duplication caveat? → Verify: does the Numbers to verify list cover every figure in the output, including each intermediate value in the feasibility arithmetic? → Avoid: inventing numbers not present in the pasted data (mark anything inferred as ASSUMED in the Numbers to verify list) → Avoid: recommending a single element A/B test when the MDE math says it cannot conclude → Avoid: drawing WHY conclusions from analytics alone; only surface questions for qualitative follow up ``` ## Prompt versions The standard prompt above works on any model. Use these variants when you want a different tradeoff. ### Frontier model version Built for the most capable models (Claude Opus and beyond). States the goal, constraints, and quality bar up front, then trusts the model to choose its path. ```text You are a senior conversion analyst running a quantitative GA4 diagnosis of a landing page or funnel. GOAL: Locate every leak precisely, rank them by size of loss, and judge whether a valid A/B test can even conclude on this traffic. Deliver a diagnosis a skeptic could act on. Your first line of output is the single biggest leak and whether the fix belongs on the page or downstream; supporting detail follows. CONTEXT: → Target page or funnel: {{PAGE_OR_FUNNEL_URL}} → Primary conversion / key event: {{PRIMARY_CONVERSION}} → Ordered funnel steps with a URL pattern each: {{FUNNEL_STEPS}} → Weekly traffic and weekly conversions: {{WEEKLY_VOLUME}} → Pasted GA4 data (tables, explorations, described screenshots): {{GA4_DATA}} → Business model: {{BUSINESS_MODEL}} PRINCIPLES (the method that makes this trustworthy): → Analytics tells you WHAT, WHERE, and HOW MUCH, never WHY. Surface questions for qualitative follow up; do not answer them. → Set the device split before you prioritize anything, and let it drive priority. Do not assume mobile dominates; a desktop only tool inverts the whole priority. Name the device you would optimize for. → Decide from the funnel whether the leak is the page failing to send people forward or a downstream step (form, checkout) failing. Fixing the page will not move a downstream leak. → For test feasibility, show the arithmetic: baseline rate as weekly conversions over weekly traffic, required sample per variant at 95 percent confidence and 80 percent power, and weeks to run. If the MDE needed to conclude within roughly four weeks is large (about a 20 percent lift or more), a single element test cannot conclude; recommend a radically different variant or, at low traffic, period over period comparison plus qualitative validation. → GA4 de duplicates conversions per user, so treat rates as comparative, not precise absolutes. Segment every finding by device. EXCELLENT OUTPUT SATISFIES: → One device named to optimize for, backed by the actual split and per device rates, with the reason. → One biggest drop step with its percentage, called plainly as page or downstream. → A feasibility verdict with the math written out and a clear "test this way" or "do not test, validate qualitatively instead". → 3 to 5 leaks ranked by size of loss, each carrying one specific question for qualitative research. → A closing list of every figure quoted or computed, each traced to the exact table or row it came from, or labeled ASSUMED. DO NOT: invent numbers not in the pasted data (label anything inferred ASSUMED); recommend a single element test the MDE math rules out; draw WHY conclusions from analytics alone; pad with generic advice. If the primary conversion, the ordered funnel, or the weekly volume is missing, ask one focused question instead of guessing. ``` ### Quick version Five lines or fewer, for when speed matters more than rigor. ```text Act as a conversion analyst reading GA4 data for {{PAGE_OR_FUNNEL_URL}}, goal {{PRIMARY_CONVERSION}}, from this data: {{GA4_DATA}} and this weekly volume: {{WEEKLY_VOLUME}}. Name the device to optimize for, the single biggest drop step (page or downstream), and whether an A/B test can conclude in about four weeks (show the baseline rate and weeks to run). Every number must trace to a row of the pasted data; label anything inferred ASSUMED. ``` ## How to customize → `{{PAGE_OR_FUNNEL_URL}}`: the exact landing page you are optimizing, for example `example.com/get-a-demo` → `{{PRIMARY_CONVERSION}}`: the single key event that counts as success, for example "submitted the demo request form" or "completed checkout" → `{{FUNNEL_STEPS}}`: the ordered steps with a URL pattern each, for example "1) `/get-a-demo` 2) `/basket` 3) `/order-confirmation`". Capture these URLs while you walk the experience so you can build the funnel later → `{{WEEKLY_VOLUME}}`: weekly users and weekly conversions for the target page, needed for the feasibility math → `{{GA4_DATA}}`: paste the tables or describe the screenshots from your freeform, path, and funnel explorations → `{{BUSINESS_MODEL}}`: lead gen, ecommerce, SaaS, or local service, so the analyst weights new versus returning users correctly ## What good output looks like → Names one device to optimize for, backed by the actual split, and explains why (not a default assumption that mobile wins) → Points to a single biggest drop step with its percentage and states plainly whether the fix belongs on the page or downstream → Gives a feasibility verdict with the arithmetic written out: baseline rate, required sample per variant with the confidence and power assumptions, weeks to run, achievable MDE, and a clear "test this way" or "do not test, validate qualitatively instead" → Ranks 3 to 5 leaks by size of loss, each carrying one specific question for qualitative research rather than a WHY guessed from the data → Closes with a Numbers to verify list where every figure traces to a specific table or row of the pasted GA4 data, and anything inferred is labeled ASSUMED ## Related prompts → [Set Up Clean Tracking and UTM Conventions](../analytics-measurement/set-up-clean-tracking-and-utms.md): fix the source or medium and event setup here first if your GA4 data looks untrustworthy → [Choose Core Metrics and KPIs](../analytics-measurement/choose-core-metrics-and-kpis.md): decide which key events and rates matter before you go mining for them → [Measure True Channel Lift With Incrementality Tests](../analytics-measurement/design-incrementality-tests.md): when a source looks like it converts well, confirm the lift is real rather than correlated