The GA4 reporting method is a fixed, written-down discipline. It turns a small business’s Google Analytics into a report an owner can trust.
The GA4 reporting method runs these six disciplines in order, every run.
It matters because raw GA4 lies to you by default. A rushed reader celebrates a number that is really bots, a broken booking form, or the wrong property entirely.
The method works in a strict order. First you establish data trust, then you rebuild the real conversion, then you screen out junk before you compare anything, then you explain every swing, and only then do you write the report. Follow that order and the same account produces the same honest answer every run, whether a person runs it or an agent does.
This method is not tied to one client or one industry. A business-model router sits at the front and adapts it to the business in front of you: local-service lead-gen, e-commerce, audience and list-building, membership or course, or donation and nonprofit.
The method never tells an owner their business is out of scope. It just changes what counts as the primary conversion and runs the same disciplines.
Distrust the default dashboard
Open the default GA4 dashboard for almost any small business and you see numbers that feel official and are quietly wrong. GA4 counts a conversion as whatever someone flagged as a key event, which is often nothing, the wrong thing, or three overlapping things.
It reports sessions from bots that never became a customer. It shows a direct-traffic pile that hides where people actually came from.
An owner reading that screen in good faith draws the wrong conclusion and spends money on it. The method catches all of this on purpose, in the same order, every run.
Establish data trust first
You do not report a number until you trust the source. This discipline has two parts.
The first is pre-flight. Confirm you are looking at the right, live property. Check that it records sessions every day and did not stop three months ago.
Then look for sibling and ghost properties on the same brand: an old dead property that still holds data, or a barely-live second property someone set up and forgot. Reporting on the wrong property is one of the most common and most embarrassing mistakes, and pre-flight is how you avoid it.
The second part is Data Clarity, a fixed set of checks that grade how trustworthy the data is. Every run earns a grade of Clear, Hazy, or Opaque.
The checks include a direct-traffic share that runs too high, an unexplained direct surge, unassigned and UTM problems, call-attribution quirks, and dilution, which is when a single conversion is really more than half of the total.
They also include a tagged-junk traffic flood, a booking-tool referral loop where your own booking software shows up as a traffic source, and a capture gap, where a large chunk of sessions land as (not set) with near-zero engagement, which is almost always bots. The grade travels with the report so nobody mistakes a Hazy account for a Clear one.
Reconstruct the real conversion
Small business GA4 is almost never set up cleanly, so the method rebuilds the conversion by hand. It separates the online action, a form submit, a call click, or a booking, from the fulfilled outcome, a real lead inside the service area.
The method inventories the actual event names in the account and decides which are true leads and which are micro-events. A generic form_submit that double-counts is a micro-event. A form_start that never completes is a micro-event.
A real named lead event that a business never registered as a GA4 key event still counts as a lead, because the name tells you what it is even when the flag is missing. This is why the built-in dashboard so often shows a handful of leads when the real number runs several times higher.
Screen before you reconcile
Order matters here, and getting it backward produces a false trend. The rule is screen, then reconcile.
First you screen. You remove spam and out-of-area junk: bot form-fills from foreign cities and datacenters, and submissions from outside the service area.
Only after the data is screened do you reconcile, comparing this period to the prior period on the same screened basis. Reconcile first and screen later and you compare a clean number to a dirty one, so the trend you report is fiction.
Decompose every swing
When a number moves, you do not guess why. You decompose the change by type, calls versus forms, then by individual event, then by channel, until you find the real driver, and you corroborate it with channel-level session data.
If a spike in leads came from one channel, the session data should show it. If the data cannot verify a cause, you do not assert it.
You turn it into a specific question for the client instead, such as, did you run a promotion the week of the fourteenth. A named question is honest, and a made-up cause is not.
Route the model and lock it
On the first run there is no saved configuration, so the method proposes one. It states what counts as a lead, the service area, any known ghost properties, and the open fixes, and it marks the whole thing First-Run, meaning it needs a human to sign off.
Once a person approves it, that becomes a locked config the method reuses on every future run. That lock keeps week-over-week reporting consistent instead of drifting every time someone new looks at the account.
Produce two drafts
Every run ends with two outputs. The first is a plain-English, client-facing draft an owner can read and act on.
The second is a separate internal notes section that holds the Data Clarity grade, the reconciliation math, and the classification. Both carry a draft label for human review, and nothing reaches the client until a person approves it.
Write the method down
The model provides intelligence. The method provides discipline. Those are different things.
A smart human in a hurry skips the property check, cheers a traffic spike, and never notices the traffic was bots or that the booking flow broke last Tuesday. A written-down method enforces every check every time, which is what makes it safe to hand to software that runs on its own.
This is the same principle behind agents that build and document their own work and a system that documents itself. The discipline lives in writing, and the agent executes the writing.
Host this method as a managed agent and it runs these six disciplines in order, then hands a person a labeled draft to approve. We walk through a real client run in the companion piece, where the method decides, in order, on one anonymized account and rebuilds a handful of dashboard leads into the real, screened number.
Turn your process into a method
If your reporting process lives in your head, in a habit, or in one person’s memory, it sits one distraction away from producing a wrong number.
My own view, after enough of these runs: the order is the product. The intelligence is cheap now, and the discipline is what a client actually pays for.
At Local Service Spotlight we turn that process into a documented method, then run it on the account so an owner sees what their analytics were really telling them. Writing the method down once is what turns a habit that lived in one person’s head into a skill an agent runs the same way every time. Write it down once and it works for you every run after.
Questions people ask about the GA4 method
What does the GA4 reporting method actually do? It turns a small business’s raw Google Analytics into a report an owner can trust. It confirms the right property, rebuilds what truly counts as a lead or sale, screens spam before it compares anything, decomposes every swing to a real driver, grades how trustworthy the data is, and writes a plain-English draft for a human to approve.
Why not just read the GA4 dashboard? Because the dashboard counts whatever was flagged as a key event, includes bot sessions, and hides real sources inside a direct-traffic pile. The method exists to catch those three failures on purpose, in the same order, every run.
What counts as a lead? A real named lead event from a person inside the service area, separated from micro-events like a double-counting form_submit or an abandoned form_start. A named lead event still counts even when the business never registered it as a GA4 key event, which is why the built-in report often undercounts by several times.
Does it work for e-commerce and nonprofits, or only lead generation? Any small business. A business-model router adapts what counts as the primary conversion for local-service lead-gen, e-commerce, audience building, membership, or donation and nonprofit. No business is ever told it is out of scope.
How is a run kept consistent week to week? The first run proposes a configuration a human signs off on, and that becomes a locked config the method reuses on every future run, so the numbers do not drift when a new person looks at the account.
Can I run this on my own Google Analytics? Yes. The method is packaged as a skill you can install in your own Claude and point at your own GA4 property, with none of our infrastructure involved. The next section has the files.
For agents: run this method as a skill
Everything above is the method in prose for a human reader. This section is the same method as an installable skill an AI agent runs directly against a Google Analytics property. It ships with no client data. The examples inside are illustrative stand-ins.
Download the pack: the GA4 reporting skill. It contains three things: ga4-website-maa.zip (the installable skill, a SKILL.md plus its references/ folder), ga4-website-maa-PASTE-IN.md (the whole method as one file to paste into any Claude), and an installer prompt.
To install it as a skill, download the pack and paste this to your Claude with the inner ga4-website-maa.zip attached:
> I’m giving you a skill called ga4-website-maa as a zip. Install it as a skill: unzip it so SKILL.md sits at the skill root with its references/ folder beside it, keep the folder name, and do not change the file contents. Then confirm it is available and summarize what it does.
To run it after installing:
> Run the GA4 report for [client name]. The GA4 property ID is [id].
The skill needs a Google Analytics connection through the official GA4 Data API tools on the account you are analyzing. It produces the same two drafts described above: a client-facing report and a separate internal notes section, both labeled for human review. Host it as a managed agent and it runs the whole discipline on demand or on a schedule.