How ConvRadar works — Guide

Four ready-made workflows

Type one slash-command, get a structured answer.

These are the fastest path to value. Each one orchestrates the right tools in the right order so you don't have to learn the API. Use them as defaults; ask follow-up questions in chat.

/full_audit≈ 5–8 min

Complete CRO audit on the connected property

Pulls overview metrics, scans for anomalies, and diagnoses the biggest funnel drop.
Matches the finding against the hypothesis library and verifies the top candidate on your actual page.
Returns three prioritised findings + the single next thing to ship.

"Run a full audit on my property."

/weekly_checkup≤ 2 min

Week-over-week health check

Headline WoW deltas: sessions, conversion rate, revenue, AOV.
Anomaly scan over the last 7 days; flags statistically-significant breaks.
Status of in-flight hypotheses + verdicts on changes you shipped recently.

"How is the site doing this week?"

/diagnose_drop≈ 2 min

Focused diagnosis when one metric drops

Splits the drop across device / source / country / landing page.
Surfaces the top 3 contributors with confidence intervals.
Suggests candidate hypotheses without auto-running verification — fast and cheap.

"My checkout conversion dropped last week — diagnose."

/suggest_next_test≈ 1 min

Pick the next A/B test by leverage

Reads your confirmed-but-unshipped hypotheses + recent change journal.
Frames the highest-leverage one through the A/B test design playbook.
Hands you a primary metric, MDE, sample size, and a ramp plan.

"What should I A/B test next?"

Plain-English questions

Or just ask in your own words.

ConvRadar maps your question to the right tools — you don't need to know the names. Pick a row that matches what you want to learn and paste it into Claude.

AHow am I doing?

Status, trends, benchmarks. Use these as a baseline before drilling deeper.

Show me a 28-day overview with revenue, conversion rate, and AOV vs the previous period.Headline KPIs with prior-period deltas.
How does my conversion rate compare to industry benchmarks for my vertical?CR / AOV / bounce vs benchmark band, calibrated to your vertical.
Where is most of my traffic coming from this month?Top sources/mediums by sessions and revenue.
How does mobile conversion compare to desktop?Side-by-side segment comparison with deltas.

BWhy did X happen?

Diagnostics — the part that makes ConvRadar more than a dashboard.

Why did conversions drop last week?Anomaly detection + segment breakdown of the drop.
Where is my checkout funnel leaking the most revenue?Step-by-step funnel with leak ranking and dollar value.
Did my paid traffic quality change recently?Source-share shifts + per-source spike/drop detection.
Which products are dragging the catalogue conversion rate down?Per-SKU classification (underperforming / watchlist / healthy) with broken funnel step.

CWhat should I do?

Recommendations + verification + measurement loop.

What should I A/B test next on my product page?Pulls confirmed hypotheses; frames the test (metric, MDE, sample).
Show me hypotheses for checkout friction in subscription e-commerce.Browse the verified hypothesis library by category + vertical.
I shipped a price-display change on Apr 18 — did it work?Logs the change, runs pre/post z-test 14 days later.
Verify this hypothesis on my actual product page.ConvRadar fetches the page, extracts observables, returns a calibrated verdict.

Under the hood

Diagnose, match, verify, then act.

This is the loop that separates ConvRadar from chatting with a raw analytics dump. The product enforces it; you skip steps only when you want to.

1

Diagnose

ConvRadar pulls the right slice of GA4 — funnel, segments, traffic mix — and surfaces where the loss concentrates. Statistical, not vibes-based.

2

Match

The finding gets matched against a library of 126 verified hypotheses across 21 categories and 11 verticals. You see what's likely, ranked.

3

Verify

ConvRadar fetches your actual page (or asks for a screenshot), extracts observables, and returns a calibrated verdict — matches: yes / partial / no.

4

Act + measure

When you ship the fix, log the change. Two weeks later, ConvRadar runs a pre/post z-test on the metric you targeted and tells you whether it moved.

Capability reference

25 tools across three tiers.

You almost never need to know these by name — Claude picks the right one from your question. But if you're evaluating ConvRadar for a team or running it for a client, here's the full surface.

Show all 25 tools9 read · 8 diagnostic · 8 hypothesis & verification

Tier 1 — Read your data (9)

Account info: Connected GA4 property name, currency, timezone, and the data window we have on file.
Overview metrics: Headline KPIs (sessions, conversion rate, revenue, AOV) with prior-period deltas.
Funnel: Step-by-step e-commerce funnel with conversion rate per step and the biggest leak ranked.
Traffic breakdown: Top traffic sources by (source, medium) with sessions, revenue, conversion rate.
Landing pages: Top landing pages with engagement, bounce, duration, purchases, and revenue.
Device breakdown: Mobile / desktop / tablet session share with conversion rate and revenue split.
Geo breakdown: Top countries by sessions or revenue, with per-country conversion rate.
Product performance: Top products by revenue/views/purchases, with cart-to-view and purchase-to-view rates.
Compare segments: Side-by-side segment comparison (device, source, country, new vs returning) with deltas.

Tier 2 — Diagnostics (8)

Find conversion anomalies: Statistically-significant spikes and drops in daily metrics (28-day rolling baseline, weekly seasonality).
Diagnose funnel drop: Segment-level leak analysis split across device, source, country, and landing page.
Compare to benchmark: Headline metrics vs industry benchmark band — calibrated to your vertical.
Detect traffic-quality change: Flags shifts in source share and per-source spikes or drops over a window.
Assess traffic quality: Quantifies analytics noise (referral spam, internal, datacenter, low-quality paid) and estimates distortion.
Product analysis: Pre-computed per-SKU diagnosis (underperforming / watchlist / healthy) with the broken funnel step.
Describe data: Metadata: which fact tables, dimensions, metrics, and events are queryable, plus the date window.
Query metrics: Flexible ad-hoc aggregation over GA4 fact tables with dimensions, metrics, and filters.

Tier 3 — Hypotheses & verification (8)

List hypotheses: Browse the verified hypothesis library — paginated index by category and vertical.
Get hypothesis: Full hypothesis content by ID — description, expected impact, suggested test, evidence basis.
Capture page: Server-side fetch of a URL; ConvRadar reads the live HTML to verify a hypothesis.
Record verification: Submit observables from the captured page; runs the matcher and returns a verdict.
Mark hypothesis status: Move hypotheses through surfaced → testing → confirmed / rejected / dismissed / archived.
Save AI-suggested: Persist a Claude-generated hypothesis when the library has no good match.
Log change: Record a deployed change (copy / design / pricing / flow / experiment) for impact tracking.
Verify change impact: Two-proportion z-test on pre/post metrics — did your change actually move the number.

Get sharper answers

Five tips that pay back fast.

Be specific about the time window."Last 28 days", "this quarter", "Apr 1 to today" — ConvRadar defaults to 30d if you don't say.
Mention the segment you care about.Mobile, paid search, returning users, a specific country. Diagnoses get sharper when scoped.
Use a slash-command when in doubt./full_audit is the right default for "something is off, find what." /weekly_checkup for routine Mondays.
Ask Claude to verify on your real page.Once a hypothesis is proposed, say "verify on my actual product page". ConvRadar will fetch the page and return a verdict — not a guess.
Log every change, then check back.After you ship a fix, ask "log this change against hypothesis X". Two weeks later, run /weekly_checkup and ConvRadar will report the pre/post verdict.

Calibrate expectations

What ConvRadar doesn't do.

Honesty saves trust. These limits are intentional — most of them are about staying out of your users' private data and out of your platform settings.

No live or real-time queries.GA4 itself has a ~24-hour data delay. You're seeing yesterday and back.
No user-level data.Aggregated rows only. No emails, no IDs, no individual session traces.
No competitor data, no Ads, no Search Console.GA4 only. We don't connect to BigQuery exports, paid platforms, or third-party tools.
No autonomous actions.ConvRadar recommends. You decide and ship. Nothing changes in your GA4 or your site without you doing it yourself.

Step-by-step setup

Plug ConvRadar into your AI in five minutes.

Click-by-click walkthrough with screenshots — pre-flight (account + magic link), then a tab for Claude and a tab for ChatGPT. The MCP URL you paste in both clients is https://mcp.convradar.com/mcp.

I use Claude

Custom connector via claude.ai/settings

→

I use ChatGPT

Workspace admin → Apps (Business)

→

Before you start

Get a ConvRadar account ready

Three steps to create the account and start a trial. Once you're logged in, the rest of the setup happens inside Claude or ChatGPT — pick your AI in the next section.

1
Open convradar.com and click Start free trial
The blue button lives in the top-right of every public page; the same CTA is repeated in the hero.
2
Enter your email — we send a one-time link
No password to remember. Supabase magic links handle auth; new email → a fresh ConvRadar tenant is created automatically.
3
Open the link from your inbox
The page confirms with “Magic link sent to ...”. Sender is noreply@mail.convradar.com. Click the link — it signs you in and routes you to the right onboarding step based on your trial state. Existing accounts go straight to the Connector page; brand-new ones run a one-shot GA4 + Stripe pass.
Tip: search Gmail for 'ConvRadar magic link' if it doesn't arrive within a minute, or check spam.

Step 4 onward

Pick the AI you want to plug ConvRadar into

The remaining steps differ between Claude and ChatGPT. Use the tabs below.

/onboarding/connect-claude with Open Claude / Open ChatGPT buttons — The MCP URL you’ll paste into both clients is https://mcp.convradar.com/mcp.

4a
Open Claude → Settings → Connectors
Direct link: claude.ai/settings/connectors. Scroll to the bottom of the connector list and click Add custom connector.
5a
A modal opens — fill Name and the MCP server URL
Name something memorable like ConvRadar (Claude makes Name required). Paste https://mcp.convradar.com/mcp into the Remote MCP server URL field. Leave Advanced settings alone.
6a
Hit Add — ConvRadar appears in your connector list
No separate OAuth screen pops up: Claude registers the connector and lands you back on /settings/connectors with a fresh ConvRadar row.
7a
(Optional) Click Configure to fine-tune tool permissions
Configure opens a per-tool list — you can set every cr_* tool to Always allow, Ask, or Never. Defaults are sensible; you can come back to this any time.
Tip: keep cr_describe_data on Ask while you're learning what each tool does.
8a
Test it: open a new chat and ask for your funnel
Try “Use the ConvRadar connector to show my funnel for the last 30 days.” Claude calls cr_get_funnel and replies with a real funnel table from your GA4 property.

If something goes wrong

Troubleshooting

My magic link didn't arrive▾

Check the spam folder. If still missing after 2 minutes, request a new link from /login. Same address — we don’t create duplicate accounts.

GA4 sync seems stuck▾

The first sync pulls 90 days and can take up to 5 minutes for properties with high traffic. Refresh the page; progress is read live from the database.

Claude says “Couldn’t reach the MCP server”▾

Almost always a stale OAuth token. In Claude → Settings → Connectors, click Configure on ConvRadar → Uninstall, then re-add the connector. Takes 30 seconds.

ChatGPT can't find ConvRadar in my chat▾

The connector lives at the workspace level, not personal apps. Make sure the workspace admin published it (step 6b). Personal-account ChatGPT (Plus / Free) currently uses a different MCP UX — see the note on the ChatGPT tab.

I want to revoke access▾

From myaccount.google.com/connections to drop the GA4 grant; from claude.ai/settings/connectors or the ChatGPT workspace admin to remove the connector. Both take effect immediately.

Anthropic and OpenAI iterate on these settings UIs frequently. If a screenshot here doesn’t match what you see, tell us at /contacts. Captured on May 8, 2026.

Already have an account? Skip to /onboarding/connect-claude →

Your conversion analyst, inside Claude — here's how to use it.