GA4 + Claude via MCP: 10 Questions That Beat Your Dashboard

Why GA4 dashboards fail in practice

Three structural problems:

1. Pre-built dashboards answer the wrong questions. GA4's default reports cover what Google thinks you should care about. Your actual question is rarely on the default screen.
2. Custom dashboards rot. A custom report built in March is wrong by August — events get renamed, conversions get redefined, your business focus shifts. Maintenance overhead is real.
3. Exploration takes too long. Five clicks to filter, two clicks to add a comparison, three clicks to change the time range. Each individual interaction is fast; ten of them in a row is slow.

Conversational analytics solves all three. You don't need a dashboard for a question you can ask in a sentence.

Question 1: What's actually working this week?

For last 7 days vs prior 7 days, list the top 5 pages by session
increase AND the top 5 by conversion increase. Show the absolute
and % change for each.

The dashboard equivalent: pull GA4's Pages report, set comparison to prior 7 days, sort by one metric, screenshot, then repeat for conversions. About 10-15 clicks. Conversational equivalent: 8 seconds.

What you typically see: 1-2 expected winners (a recent blog post) and 1-2 surprise winners (an old page that started getting AI citations or social referral). The surprises are where the strategic value lives.

Question 2: Where are users dropping off?

Build a funnel from landing page → /pricing → /auth.html → signup
completion. Show conversion rate at each step for last 30 days.
Identify the biggest drop-off step and recommend what to investigate.

GA4's native funnel builder works but requires building the funnel UI for each new question. Conversational lets you reshape funnels on the fly: change the steps, add a filter, compare segments — all in follow-up prompts.

Question 3: Which pages had sudden conversion drops?

For each page that drove 10+ conversions in the prior 30 days,
compare conversion rate this 30 days vs prior 30. List pages
where conversion rate dropped > 30% — investigate why.

This question is impossible in GA4's UI without building a custom report — and even then, exporting to a sheet and computing the deltas manually is the only way to get a clean answer. Conversational does it in seconds and surfaces an investigate-list you can actually act on.

Question 4: Which channels deserve more budget?

For last 30 days, compute conversion rate by acquisition channel.
Then compute LTV per channel (using Stripe data via MCP). Rank
channels by LTV ÷ CAC. Which has the best unit economics?

This pulls from two data sources (GA4 for conversion, Stripe for LTV) and does cross-source math. In a dashboard, that's two reports, a manual export, and a spreadsheet. In Claude, one prompt.

Question 5: What does my best customer look like?

For customers acquired in the last 90 days (from Stripe), cross-
reference their acquisition source from GA4. List the top 10 customers
by lifetime spend. What do they have in common?

Used to build an Ideal Customer Profile. The "what do they have in common" question is where AI reasoning adds value beyond the dashboard — Claude can synthesize patterns across acquisition source, geography, device, and behavior.

Question 6: When is my traffic actually arriving?

For last 30 days, show sessions by hour of day in my GA4 property's
timezone. Identify the 3-hour peak window and the lowest 3-hour window.
Recommend when to schedule social posts, email sends, and any
manual outreach.

Surprisingly underused question. Most teams send emails at 9am because that's when they're in the office, not because it's when their audience reads. Five minutes with this prompt corrects months of bad timing.

Question 7: Which source has the highest LTV?

For all customers acquired in the past 12 months (Stripe), group
by first-touch acquisition source (GA4). Show LTV per source.
Which 3 sources are most valuable per customer acquired?

This question changes most strategies. The source that drives volume is rarely the source that drives value. Knowing which is which lets you allocate budget and effort accordingly.

Question 8: What's my AI traffic worth?

For last 90 days, compute conversion rate for sessions from AI sources
(chatgpt.com, claude.ai, perplexity.ai, gemini.google.com, copilot.com)
vs Google organic vs direct. Then compute revenue per session per source.
How does AI traffic compare in value to organic?

In 2026 this is the question every marketing team should be asking and few are. AI traffic has very different conversion economics than organic — sometimes better (AI users are pre-qualified), sometimes worse (they got their answer from the AI without clicking). Knowing your specific number is gold for strategy.

Question 9: Which pages do users actually engage with?

For last 30 days, list pages with: 1000+ sessions, scroll depth
above 75% on average, time on page above 90 seconds. These are
genuinely engaged pages. Cross-reference with conversion — do the
engaged pages also convert?

Engagement and conversion don't always correlate. A page can be heavily read and never convert (interesting but not commercial). A page can be barely read and convert well (high-intent buyers don't need much). This question separates the two.

Question 10: What's broken right now?

For each tracked event in GA4 over the last 7 days, show the absolute
count and the % change from the prior 7 days. Flag any event where
volume dropped >50% week-over-week — likely indicates a tracking
break, a deploy broke something, or a real user behavior change.

This is the question that catches problems within hours instead of weeks. Most tracking breaks go undetected for 7-14 days because nobody is looking at the specific event that broke. Running this weekly costs 30 seconds and catches issues before they cost real money.

What conversational GA4 actually replaces

• 50% of your custom Looker Studio dashboards. Most custom dashboards exist to answer questions you ask occasionally. Conversational answers occasional questions on demand.
• The "I'll look at GA4 later" reflex. When the friction drops to typing a sentence, you actually look.
• Most ad-hoc analyst requests. The 10-minute Slack request "can you pull X for me?" becomes "I'll just ask Claude" — except the asker gets the answer themselves.

What it doesn't replace: scheduled stakeholder reports, client-facing dashboards, board-level summaries. Those still need polished, predictable formats.

Frequently Asked Questions