GemPages With Causality Engine: Weigh the Funnel and the Test

The short answer: the advertorial built in GemPages is getting the ad clicks, GemX says its new version has a 95% probability to win, and the store's orders still fall short of what Meta and TikTok claim between them. GemPages, a Shopify builder for landing pages, store pages and post-purchase funnels, now sends page testing to a separate app, GemX, which credits every order to the first variant a shopper was assigned, even when the purchase comes in a later session (orders and revenue in GemX). That settles which page converts better. Which channel sent the buyers is a different file: the GA4 export that Causality Engine reads to estimate what each channel caused, starting with a free first finding at app.causalityengine.ai/start.

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

Three numbers GemPages and GemX put in front of you:

1. Upsell revenue, which sets a fee. On Optimize the first $1,000 a month of post-purchase upsell revenue is commission-free, then $10 is charged for every further $1,000, per the Sales Funnel article. For example, $8,000 in a month costs (8,000 - 1,000) ÷ 1,000 × $10 = $70.
2. Experiment orders, a subset by design. GemX counts only sessions that enter an experiment, so its totals sit below Shopify's and GA4's, per How Experiment Analytics Work.
3. Probability to win, on one metric. GemX treats around 95% as a likely winner, calculated on Conversion Rate or Revenue, whichever you chose (statistical significance article).

The facts side by side

What GemPages counts, and where GemX takes over

GemPages' own help says testing moved out in September 2024: the old V6 testing needed a Google Analytics integration because it had no tracker of its own, so GemX was launched as a separate A/B testing app (A/B article). In GemX a conversion is an online store session that completes an order. Assignment is sticky, and orders follow the first variant seen: a shopper put in Variant B who returns days later through another channel and buys is still a Variant B order. Revenue is counted after discounts and may leave out shipping and tax. An order paid straight through Apple Pay, with no standard checkout session, can be missed altogether (orders and revenue).

GemX is plain about scope. Its analytics article says experiment figures should not be used to measure overall store performance or replace Google Analytics. That is the right reading, and it leaves the channel question open.

The Sales Funnel stays inside GemPages. Up to two post-purchase offers follow payment, each can be split-tested with a traffic slider, and every plan can use it, capped at $100 of upsell revenue a month on Free and Build (Sales Funnel). An accepted offer does not create a new order: Shopify edits the existing one, which is why GemPages' order tagging guide uses Shopify Flow to tag edited orders. The store's order count does not move when an upsell lands.

Headless pages and the GA4 session

A regular GemPages page publishes on the store's own domain, so it sits under the same GA4 setup as the rest of the theme and the visit keeps its source and medium. Google then sorts it by rule: a TikTok ad click lands in Paid Social only with a social source and a paid medium such as cpc (default channel group).

The Headless Page on Optimize is different. It lives on a domain hosted by GemPages' instant page service, not by Shopify, or on a subdomain from another provider (Headless Page), and it takes GA4 only through Google Tag Manager, pasted into its publish settings (tracking apps on Headless Page). Google's cross-domain measurement page explains the consequence: without cross-domain measurement, a visitor moving between two root domains becomes two users and two sessions, and self-referrals can credit traffic to the wrong source. Shoppers who land on a headless advertorial and check out on the store can then reach GA4 as referrals from your own page. Google's unwanted referrals page is the fix to apply before any export.

Neither kind of page puts its variant into GA4. The GemX articles read on 11 November 2026 describe no such setting, so a read of the export sees channels, never versions.

Worked example: an upsell funnel and a GemX test

For example, with invented figures and no store behind them, a month with a headless advertorial for paid traffic, a GemX test on the product page and an upsell funnel on Optimize:

• Shopify books 2,310 orders.
• Meta claims 1,402, TikTok 690 and Google Ads 512: 1,402 + 690 + 512 = 2,604, which is 2,604 - 2,310 = 294 more than exist.
• GemX reports 1,076 experiment orders, 1,076 ÷ 2,310 = 46.6% of the store's total, with Variant B at 95.4% Probability to win on Conversion Rate.
• 187 of the 2,310 orders took an upsell, worth $8,000, so the fee from point 1 is $70. The order count stays 2,310.
• GA4 shows 266 orders under Referral from the headless page's own domain, because cross-domain measurement was never set.

Those 266 are the first thing to fix, since no read can return them to Meta or TikTok while GA4 files them as a referral from your own site. With the domains configured, the next export goes to Causality Engine. Per its how-it-works page, the read splits Direct back to the channels behind those buyers and puts each channel's caused share beside a last-click rebuild, with a data-health score and a next step. It will not say whether Variant B or the upsell helped Meta buyers more. If TikTok's caused share lands far below its 690, test a TikTok cut in one region first.

A free test, then the export

Two free checks need no new app. For offers, the funnel's own A/B test splits buyers between two versions with a traffic slider (Sales Funnel), so the better offer is found on any plan. For the upsell as a whole, the check is by hand: four weeks with the funnel paused, four with it live, and revenue per order across all orders in each window, not the upsell figure. The difference is what the funnel adds, with the usual caveat that a sale or an ad change inside either window spoils it. For page tests, GemX's own checklist asks for a stable trend over several days before acting on a 95% reading (significance article).

For channels, export the Attribution paths CSV and drop it on the start page; the first finding runs in your browser and is free. Causality Engine, which I founded, sells the full read for €99 once, excluding VAT, with 30 days to ask for a refund (pricing, checked on 11 November 2026).

Limits on both sides

• The read needs 12 months or more of GA4 history (how it works) and key events from GA4's Advertising section (one-off read).
• A channel under 1% of credit on both methods is too small to act on, by its own rule (how it works).
• Page design, offers, journey analysis and variants stay with GemPages and GemX.

On Optimize, the upsell revenue figure GemPages reports is also the base it charges commission on, and the funnel article adds that the fee is not refunded when a customer's order is (Sales Funnel). A number that sets the vendor's fee deserves a check the vendor did not run, which is what the paused-funnel comparison above provides.

See also

• GemPages: plans, funnels, the MCP beta and complaints
• How to use GemPages: one advertorial per ad angle
• GemPages on the marketplace: GemPages' affiliate program terms
• Causality Engine: the read, what it costs, where it declines
• How to use Causality Engine: GA4 export to budget decision
• Causality Engine on the marketplace: its affiliate program terms
• GemPages vs Shogun: page caps against testing apps
• How to export GA4 data for attribution: the file the read needs

FAQ

Do GemPages or GemX feed Causality Engine?

Neither does. The read takes a GA4 export and measures the channels recorded in it (integrations), so GemPages matters only through the visits GA4 logged.

Why does GemX show fewer orders than Shopify?

GemX counts orders from shoppers who entered an experiment and can be matched to a tracked session; failed, draft and cancelled orders and some wallet payments fall out (orders and revenue).

Does a Headless Page need cross-domain setup in GA4?

When it sits on a different root domain from the store, yes. Google's cross-domain page says otherwise one visitor counts as two users in two sessions.