How to Export 12 Months of GA4 Attribution Paths

The short answer: the file an attribution read needs is about 2 minutes away. If you hold a Viewer role on the GA4 property, you can pull it yourself, per the four steps on Causality Engine's start page. It is the Attribution paths report, 12 months or more, the purchase key event alone, downloaded as a CSV. That page is also where the file should go first, because it reads the export in your browser and gives a first finding free. The steps: open the report, set the window, pick one key event, leave the view as it opens and download. Then run five checks on the file before anyone draws a conclusion from it.

Disclosure: I founded Causality Engine and own it, and this page recommends it. Every claim about it links to Causality Engine's own pages.

Settle three things before you open GA4:

1. The window, in dates. For example, 12 months back from 27 October 2026 runs from 27 October 2025 to 26 October 2026, which is 365 days. Google's report help says this report holds data from 14 June 2021 onwards, so a longer window is available. The how it works page says longer reads better.
2. Your role and your key event. Exporting needs the Viewer role at property level, per Google's share and export help. The purchase must already be marked as a key event, which Google's key events page describes as marking an event you already collect.
3. A total to reconcile against. Note GA4's purchase count and purchase revenue for the same dates. The file is only as good as its agreement with that number.

The export is the input. What happens after it is in How to use Causality Engine, and the product itself in the Causality Engine glossary entry.

Where the report sits

The start page names it Attribution paths, under Advertising. Google's help files it as Key event attribution paths: click Advertising on the left, open the Key events dropdown, then choose the report. Same report, two labels. Both lead to a chart above a table of paths.

Step 1: Set the window

Pick the dates first. GA4's data-retention setting can be as short as 2 months, and it is fair to worry that it caps the export. It does not touch this report. Google's data retention help says the setting affects explorations and funnel reports only, not standard aggregated reports. A property set to 2 months still exports a full year of paths.

Step 2: Pick the purchase key event alone

The report opens with every key event selected and aggregated together, per Google's report help. A newsletter sign-up or an add-to-cart marked as key would then add paths that never ended in a sale. Choose purchase from the dropdown at the top left. Nothing else.

Step 3: Leave the view as it opens

The table breaks paths out by Primary channel group, shows all users and lists paths up to 20 touchpoints long. Leave all three alone. A filter, such as the Chrome-only example in Google's help, removes buyers from the file. Switching the table to Source or Campaign changes what each row means. The model dropdown alters the chart above the table, not the list of paths below it.

Step 4: Download the CSV

Leave any Customize report mode first, because Google's share and export help blocks downloads while you edit. Then click Share this report in the top right, Download File, Download CSV. The same page caps a CSV at 100,000 rows.

Step 5: Check the file in five places

Open it in a spreadsheet before you upload it anywhere:

• Row count. If the file stops at exactly 100,000 rows, it hit Google's cap. The table sorts paths with the most key events first, per the report help, so the rarest paths are the ones missing. Note it before anyone sums the rows.
• (Other) rows. Google's report help uses (Other) for rows aggregated under cardinality limits. A few are normal. A large one means detail was folded away.
• Unassigned and (not set). Unassigned means no channel rule matched, and (not set) means a value such as source or medium never arrived. Both point at tagging, which Google's tagging guide traces to tag order, server-side settings and cross-domain linking.
• Direct as a channel. In this report Google defines Direct as the value used when there is no path data to credit, for example after a data import. Paths ending in Direct after other touches are a different thing. They are the subject of How to find what Direct traffic is hiding.
• The totals. The report's top row shows total key events and purchase revenue. Set them against the number you wrote down before step 1.

The start page adds one check you do not need to make: the file holds channel paths and totals, with no customer names, emails or order IDs.

An agent can run the reconciliation

Google documents a read-only Google Analytics MCP server with a run_report tool on the Data API. An assistant connected to it can pull ecommercePurchases and purchaseRevenue for your exact window. And list_property_annotations returns the annotations configured on the property, such as a note your team left on the week a consent banner changed.

The agent does not produce the paths export, which still comes from the report screen. The human check: you name the property and dates yourself and compare the agent's totals with the file's. Then you decide whether any gap is explained before the file goes further.

What to do with the file

The on-screen report is convenient, and its credit figures arrive already worked out. Every one of them comes from a model Google picked or trained. Google also sells the search, Shopping and YouTube ads those figures credit. The paths in the CSV carry no model at all.

That is the part worth taking out.

1. Read it by hand, free. Sort paths by key events and count how many end in Direct after another channel. That count alone tells you how much of Direct is an arrival rather than a cause.
2. Take the free first finding. Drop the file on the start page. The preview runs in your browser and the file stays there.
3. Run the full read if the decision needs it. Causality Engine, which I founded, splits Direct back to the channels behind it and sets every channel next to last-click. It costs €99 once, refundable within 30 days (pricing, checked on 27 October 2026).
4. Keep the file. Name it with its dates. Next quarter's export is only comparable if its settings match this one, and How to run a quarterly attribution checkup uses exactly that.

FAQ

Can I export more than 12 months?

Yes. The report holds data from 14 June 2021, per Google's report help, and the retention setting does not shorten it. Causality Engine's pricing page asks for 12 months or more, and its how-it-works page says longer reads better.

Do I need edit access to GA4?

No. The Viewer role is enough to download a report, per Google's share and export help. Marking purchase as a key event, if nobody has, needs the Marketer role or above, per Google's key event setup page.

Why does the chart change when I pick another model?

The chart shows the credit each model gives to early, middle and late touches. The table of paths underneath stays the same. Google's attribution settings help notes that first click, linear, time decay and position-based were retired in November 2023.