Same three orders, five conversion rates

A conversion rate that moves looks like a change in your shoppers. Often it is a change in the bottom of the fraction. Every tool agrees on what goes on top. Each one picks its own session rule, or does not divide by sessions at all, and few of them say so on the report.

We ran one visit log through the documented rules of six tools. Three orders came out as a conversion rate of 27.3%, 30.0%, 33.3%, 50.0% or 60.0%. Nothing in the shoppers’ behaviour changed between the rows. Only the rule did.

What each tool divides by

All of these come from each vendor’s own documentation, read on 26 October 2026.

Tool	A session ends after	Also starts a new session	Conversion rate divides by
GA4	30 min of inactivity (adjustable). No limit on length	Nothing else documented	Sessions (session key event rate) or users (user key event rate)
Adobe Analytics	30 min of inactivity (can only be shortened on standard suites)	12 h of activity, 2,500 hits, 100 hits in the first 100 s	Visits, by default
Matomo	30 min of inactivity	A new campaign or ad click ID, midnight in the site’s time zone	Visits
Plausible	30 min of inactivity	Nothing else documented	Unique visitors, and a visitor is one person on one day
Shopify (since 21–23 Sept 2026)	30 min of inactivity	Nothing else documented. Before the update: midnight UTC	Sessions, and only sessions with cookie consent
Google Ads	—	—	Ad interactions (clicks), not visits

Three of these are not session counts at all. GA4’s user key event rate divides by people. Plausible divides by people per day, so one person who visits on Monday and Tuesday is two visitors over that range. Google Ads divides by its own ad clicks, and its rate can pass 100% when a click gets more than one conversion.

The worked example

The log is constructed, so you can check every line. It is a US store on New York time, which in October is UTC−4. Midnight UTC is 20:00 in the store’s time zone.

Shopper	What they did on 20 October (and 21)
A	Clicks a Google ad at 19:40 and browses until 20:20. Clicks a Meta ad at 20:22. Leaves, comes back direct at 21:10 and buys at 21:15. Returns from the newsletter at 10:00 the next morning
B	One visit at 14:00, six pages, no purchase
C	Visits at 10:00, comes back direct at 15:00 and buys
D	Clicks a Google ad at 13:00, leaves the tab open 45 minutes, checks out at 13:50
E	Clicks a Pinterest ad at 19:50 and browses without a break until 20:10. No purchase

Five shoppers, 28 pageviews, 3 orders. A script split the log into sessions under each rule:

Rule	Sessions	Converting	Conversion rate
30 min of inactivity only (GA4, Adobe, Plausible visits, Shopify now)	9	3	33.3%
Matomo defaults (plus new campaign, plus local midnight)	10	3	30.0%
Shopify before September 2026 (plus midnight UTC)	11	3	27.3%
Plausible: unique converting visitors ÷ unique visitors per day	6 visitors	3	50.0%
GA4 user key event rate: converting users ÷ users	5 users	3	60.0%

Where the extra sessions come from:

• Matomo splits A’s visit at the Meta ad. A new campaign or a new click ID starts a new visit, even two minutes after the last page. A shopper who clicks two ads in one sitting is two visits.
• Shopify’s old rule split A and E at 20:00. For a US store, midnight UTC falls in the evening, when people shop. E browsed for 20 minutes without a break and was counted as two sessions. Shopify moved off this rule between 21 and 23 September 2026, and says the conversion rate can change even when orders have not. A Shopify conversion rate that rose in late September may be the rule change, not the store.
• D is two sessions in every tool. The tab sat open 45 minutes, so the checkout is a new session. Adobe’s documentation gives this case as the most common gap between a visit and a browser session.
• A is one person in GA4’s user rate and two in Plausible’s. A came back the next day. Plausible counts people per day, so A is a visitor on 20 October and another on 21 October.

The log is small, so the rates are large. The spread is the point. With real traffic, the size of the gap depends on how often your shoppers come back, click a second ad, or browse across a boundary.

What each count leaves out

The denominator is also a filter, and each tool filters something different.

Shopify only counts sessions with cookie consent. Its analytics field reference says so for the Sessions metric. A Shopify session count in a region with a consent banner is a count of the visitors who accepted it. If you compare it with a tool that counts differently, part of the gap is consent, not traffic.

Shopify also filters identified bots by default since the September update. A tool that keeps them has more sessions and a lower rate for the same orders.

Google Ads only sees its own clicks. In the log, shopper A clicked a Google ad and then a Meta ad before buying. Each platform can count that order against its own click. Google Ads divides by its own interactions, so its rate describes Google traffic, not the store.

What AI changes

An order placed inside an AI assistant can arrive with no visit at all. OpenAI’s Agentic Checkout Spec renders the checkout inside ChatGPT. ChatGPT collects the buyer’s details and calls the merchant’s endpoints, and the merchant takes payment on its own systems. The buyer never loads a page of the store.

That order has no session, so it cannot enter a numerator of “sessions that completed checkout”, which is how Shopify defines the rate. If you calculate orders ÷ sessions yourself, it enters the top and not the bottom, and the rate rises with no change on the site. As more sales move into assistants, the two ways of calculating it drift apart. Ask which one a dashboard uses.

What to do differently

1. Write the denominator next to the rate in every report: sessions, users, daily visitors or ad clicks. A rate without one cannot be compared with anything.
2. Compare a rate only with the same tool’s rate. GA4’s session rate and Shopify’s conversion rate use similar rules, but Shopify counts only consenting sessions and drops bots. They will not match.
3. Check the calendar before you read a trend. A change in session rules, a timeout setting or bot filtering moves the rate with no change in behaviour. Shopify’s 21–23 September update is one.
4. Read the rate next to the volume. Three orders is three orders in every row of the table above.

The bottom of the fraction is one moving part. The top moves too: an ad platform’s conversion count for a day keeps changing after the day is over, because conversions are written back to the click date.