Here is the sequence most brands will experience: your reporting breaks before your AI revenue shows up.
Fall '26 shipped "Sell in Google Search, including AI Mode, and the Gemini app with checkout powered by Shopify." According to multiple industry reports, Google began enrolling eligible Shopify merchants into native AI Mode and Gemini checkout on 22 September 2026 — as an opt-out, with merchants notified by email that their store had been matched to Merchant Center. The off switch lives in Sales channels → Agentic.
The orders are yours. The measurement around them is not.
Key takeaways
- Orders, customer records, fulfilment, support and refunds all stay in your Shopify admin, in standard form with customer data attached. Ownership does not change.
- Anything bound to a thank-you page on your domain does not fire: GA4
purchase, Meta Pixel, Google Ads conversion tags, and pixel-built Customer Match audiences. - Multi-touch attribution tools develop gaps. Server-side sourcing from the order store is the only reliable fix.
- Fall '26 also changed session counting with smarter bot filtering — numerator and denominator moved in the same quarter.
- Most brands should not switch it off. The sensible first move is to repair measurement, then decide.
What does not change
Worth stating first, because the instinct is to read this as the platform taking your customers.
It does not. Orders land in your Shopify admin as standard orders with customer information. Fulfilment, returns, support and refunds run exactly as before. You remain the merchant of record. Nothing about ownership of the customer relationship moves.
What does change
Every event that depends on a browser loading a thank-you page under your domain. On this path, that page is never loaded.
Measurement system | Status on this path | Replacement |
|---|---|---|
GA4 page events |
| Measurement Protocol / server-side, sourced from the order store |
Meta Pixel | Does not fire | Conversions API, with the order as the event source |
Google Ads conversion tag | Does not fire | Enhanced conversions or server-side upload |
Multi-touch attribution tools | Gaps appear | Shopify order webhooks; accept coarser channel resolution |
Customer Match audiences | These buyers are missing | Rebuild audiences from order data |
Shopify admin orders | Normal | Treat it as the single source of truth for this path |
Notice the pattern: every fix is the same fix. Stop deriving revenue truth from the browser and start deriving it from the order store. Teams that already did this for iOS privacy changes have most of the work behind them.
The second-order trap
The same release shipped: "More accurate session measurement — sessions in analytics more accurately reflect human visitors with smart bot filtering and unified user journeys."
Sessions are the denominator of conversion rate. Agent-path purchases are missing from the numerator. Both moved in the same quarter.
If you compare conversion rate before and after this release, the number you produce is not an improvement or a regression. It is an artefact. Draw an explicit version boundary in your reporting, annotate it, and rebuild the baseline after it. Fall '26 added manual annotations in analytics for exactly this kind of event — use them.
Should you turn it off?
Our view: most brands should not.
Switching it off does not recover the exposure. It pushes the sale back onto click → landing page → checkout, which is the lower-converting path, and you do not get the impression back by refusing the transaction. You have closed a better route to protect a report.
Three cases justify a serious evaluation:
- You depend heavily on first-party on-site data for personalisation or loyalty, and cannot move to server-side sourcing in the short term.
- Average order value is high enough that a human needs to be in the sales process.
- Compliance does not permit completing a transaction inside a third-party interface.
Even in those three, the correct order is: fix server-side attribution first, then decide. A decision taken while your numbers are wrong is not a decision.
A 30-day repair plan
Week | Action | Done when |
|---|---|---|
1 | Check your actual switch state in Sales channels → Agentic. Do not assume the default | The current state is documented, not guessed |
1 | Subscribe to order webhooks and land them in your warehouse | Orders arrive independently of any pixel |
2 | Move GA4 purchase to Measurement Protocol, Meta to CAPI, Google Ads to enhanced conversions | Order-sourced conversions reconcile with admin totals |
3 | Rebuild Customer Match and lookalike seeds from order data | Audience sizes recover |
4 | Draw the version boundary, annotate it, rebuild the baseline, add an agent-channel line to reporting | Agent-sourced orders can be counted on their own |
That last row is the one teams skip and later regret. If agent-sourced orders are not a separate line in your reporting, you will spend 2027 arguing about whether any of this worked.
FAQ
How do I tell whether an order came through an agent path? Order-level attributes and the cart/checkout attribution carried by the protocol are the starting points; Fall '26 also added attribution data to cart and checkout permalinks. Expect this to be coarser than classic UTM tracking, and design the taxonomy before volume arrives rather than after.
Does this affect ChatGPT and Copilot purchases too? The attribution mechanics are the same wherever checkout completes outside your domain. The enrolment story differs by channel — see what the ChatGPT Shop at button actually does.
Is the 22 September date confirmed by Shopify? It comes from industry reporting rather than a first-party announcement, which is why we flag it as second-hand. Your own admin switch state and the terms you accepted are the authoritative source. Everything else in this article is first-hand measurement.