Ad Attribution for Shopify: Models, Windows, and How to Read the Numbers Honestly (2026)

If you have ever added up what Google, Meta, and TikTok each said they made you and found the total exceeded everything your store sold that month, nothing was broken. Each platform applied its own generous rule to the same orders, and nobody deducted anything. This guide explains how those rules work, what each platform counts out of the box, who benefits from that, and how to check your own numbers in about ten minutes.

The models, defaults, and windows here are current as of August 2026. Three of these platforms changed their attribution behaviour in the last twelve months, so check the setting in your own account before relying on any figure.

The quick checklist

• Attribution is a rule, not a measurement. It decides who gets credited for a sale that already happened. Change the rule and the same sale changes owner.
• A click is an observed event. A view is not. View-through attribution credits an ad because somebody scrolled past it.
• The longer the window, the more coincidence gets counted as performance.
• Google offers two models now: last click and data driven. First click, linear, time decay, and position based were withdrawn, and conversion actions using them were moved to data driven automatically.
• Google’s defaults are thirty days for clicks, three days for engaged views, and one day for views. Clicks stretch to ninety.
• Meta’s default is seven days of clicks, one day of engagement, and one day of views. Since January 2026 one day is the longest view window it offers.
• TikTok offers one to twenty eight days of clicks and up to seven days of views, and the setting locks once the ad group publishes.
• ChatGPT Ads excludes view-through from its headline number entirely.
• Shopify picks one winner per order, crediting the last channel the customer clicked through from and ignoring direct visits. Every model on its channel report works this way, which is why those rows sum to your attributed sales.
• Shopify also documents an “any click” model that gives the full order value to every channel clicked, so one sale can be credited three times over. It is deliberate, and it is not on the Growth attribution table, so you have to know where to look.
• Last click is not as blunt as it is made out to be. It ignores introductions that failed to close, not introducing itself. A conversion ad that gets somebody from never having heard of the product to buying, in one click, is credited in full.
• The gut check: add up what every platform claimed for a month, compare it to your actual Shopify revenue, and you have measured your own inflation.
• Your true number is a range, not a figure. The platform total is the ceiling and Shopify’s own channel report is the floor, because Shopify undercounts paid for reasons unrelated to performance. The honest figure sits between them, nearer the floor.
• The honest default: seven-day click, last click, view-through off, counted once.

Attribution is a rule, and the rule is written by the people it pays

A sale happens. That is a fact, recorded in your Shopify admin with a payment attached.

Attribution is everything that comes after: deciding which of the things that happened beforehand should be credited with causing it. It is not a measurement of what occurred, it is a rule applied to what occurred, and the rule is a choice.

The rule deciding how much credit Google Ads gets is written by Google. The rule deciding how much credit Meta gets is written by Meta. Each platform is graded by you on the number its own rule produces, and each is paid more when that number looks better. No other part of your business works like this. You would not let a supplier write its own invoice, grade its own delivery performance, and then decide which of your sales it was responsible for.

That does not make the platforms dishonest, and their defaults are defensible on their own terms. It does mean the defaults all lean the same way, which is toward crediting the platform, and that nobody in the chain has an incentive to lean the other way.

Two settings do the work. The model decides how credit is split when there is more than one candidate. The window decides how far back a candidate is still eligible.

The six attribution models, and which ones still exist

Only two of those six are still a live choice anywhere, and most ad platforms never offered a menu at all. Meta, TikTok and ChatGPT Ads do not let you pick a credit-splitting rule: they each credit a conversion to one of their own ads by logic they do not publish, and what they give you control over is windows and interaction types instead. Shopify offers the widest selection of any of them, with last click, last non-direct click, first click, linear and any click available in its marketing reports. Google is where the withdrawals happened, and they matter more than the models.

Per Google’s attribution documentation, first click, linear, time decay, and position based are no longer supported, conversion actions using them were upgraded to data-driven attribution whether the advertiser asked or not, and data driven is now the default for most conversion actions. Google Analytics removed the same four in November 2023.

So on Google the choice is now binary: the platform’s own model, or last click. Every rule-based alternative you could reason about from first principles has gone, and the model that replaced them cannot be audited, reproduced, or compared against anything except last click. When your reported conversions rise after a model change, there is no way to establish whether more sales happened or the model became more generous.

Your model choice also drives bidding, not just reporting. Target ROAS and Target CPA optimise toward whatever the selected model counts.

What each platform counts out of the box

Google’s figures come from its conversion windows documentation, which also recommends windows of at least seven days on the grounds that they provide richer conversion data. That is a fair point about signal. It is also, unavoidably, a recommendation to count more.

Thirty days is a long time in ecommerce. It credits an ad clicked on the first of the month with a purchase made on the thirtieth. For a store selling a forty-dollar item that people buy in one sitting, most of what that window captures is coincidence.

TikTok’s settings come from its Ads Manager documentation, which also warns that they cannot be changed after the ad group publishes. Choose before you launch. TikTok is also the one platform that will let you switch view-through crediting off entirely, and its attribution overview confirms that each conversion is credited to only one touch point, so there is no double counting inside TikTok’s own figures.

Clicks, views, and the distinction that decides everything

A click-through conversion means somebody clicked, arrived at your store, and later bought. There is a recorded event, a timestamp, and a chain you can follow.

A view-through conversion means an ad was loaded onto somebody’s screen, they did not click it, and at some later point they bought from you. That is the whole basis of the claim. There is no evidence they looked at it or noticed it.

An engaged view sits between the two. TikTok sets the bar at six seconds of watch time; Meta’s engage-through covers likes, saves, comments and shares. Something happened, but the shopper never came to your store from the ad.

The case for counting views is not stupid. Advertising works partly through exposure, and a click-only model will undervalue upper-funnel work.

The case against, for a merchant deciding where next month’s budget goes, is stronger. View-through credit is not a measurement of influence, it is an assumption of influence, made by the party that benefits from it. It is structurally biased toward whichever platform served the most impressions, because more impressions mechanically produce more coincidental overlaps with purchases you were getting anyway.

That bias is worst in retargeting, which is where most Shopify stores lean on it hardest. A retargeting audience is by definition people who already visited your store. Many of them were coming back regardless. Jay Stampfl of Blackbird PPC, writing in Search Engine Land in April 2026, makes the point directly: view-through credit inflates reported performance most in retargeting, precisely where genuine incremental impact is lowest, and platforms fold it into default reporting without flagging the problem, so you have to ask for it to be separated out. On his agency’s accounts, stripping view-through credit from retargeting has more than halved the reported conversions.

Who benefits from a loose default

Three incentives point the same way, and understanding them is more useful than memorising any setting.

The platform wants to look good to you. Its reported conversions are the argument for your next budget increase. A longer window and looser crediting produce a better number at no cost to the platform, and the merchant never sees the counterfactual.

The platform also wants to look better than the other platforms. This is the part most merchants miss. When you compare Google’s dashboard against Meta’s to decide where to move budget, you are comparing two self-graded scorecards produced under different rules. The platform with the loosest window and the most generous crediting wins that comparison, regardless of which one actually sold more. Loose defaults are not only a tool for extracting budget from you, they are a weapon against the other bidder for it.

Which makes what happened at Meta in 2026 genuinely interesting, because Meta moved the other way. Per Meta’s developer announcement, from 12 January 2026 the seven-day and twenty eight-day view-through windows stopped returning data, leaving one day as the longest view window it offers. Then in March 2026 Meta narrowed click-through attribution to genuine link clicks only, moving likes, saves, comments and shares into a separate one-day engage-through bucket. eMarketer reported that the stated aim was to close the reporting gap between Ads Manager and third-party analytics. In other words, Meta decided that being believable was worth more than being flattering, and voluntarily reported less. Plenty of merchants saw their Meta numbers fall in early 2026 and assumed their advertising had broken. It had not. The measurement got stricter.

Meanwhile the newest entrant went further still. Per OpenAI’s measurement documentation, ChatGPT Ads reports view-through conversions as a separate campaign-level metric and excludes them from the Conversions total outright, so cost per acquisition, bidding, and billing all remain click-through based. Google, over the same period, retired every rule-based alternative and made its own unpublished model the default.

And your agency may want the same thing you are being sold. An agency paid a percentage of your ad spend is paid more when you spend more, and a flattering ROAS is the argument for spending more. An agency paid a share of attributed revenue is paid more directly still, because a looser setting raises the number its fee is calculated on. Neither requires anyone to lie. The agency simply reports the platform’s number, and the platform’s number was set by defaults nobody in the meeting has to defend. Loosening a conversion window is invisible in a monthly report. It looks like a good month.

Three questions worth asking whoever runs your advertising, and the answers tell you a great deal:

• What percentage of the conversions in this report are view-through rather than click-through?
• What conversion window is each of my conversion actions set to, and why that number?
• If two platforms both claim the same order, which of these reports has it removed?

We are not exempt from this. Our fee is a share of the advertising revenue we are credited with, so measuring the way the platforms do would earn us more from you for exactly the same advertising. We do the opposite, in two ways that matter more than any setting. We never take a platform’s reported revenue as our number: a sale is counted from your own order record, and only where the platform confirms the order followed a click. And when more than one platform claims the same sale, we count it once instead of adding them together, so what you get is a single figure for your whole advertising programme rather than the sum of several platforms marking their own homework. The full rules are published rather than buried in a contract.

The ten-minute proof: when the platforms claim more than you sold

This is the most useful thing in this guide and it costs nothing.

Pick a completed month. Write down your actual store revenue from your Shopify admin. Write down what each platform claimed it made you for the same month. Add the platform figures together. Compare.

You cannot have sold more than you sold. Your store also made sales from email, organic search, direct visits, repeat customers and word of mouth, so the honest ceiling for what advertising could possibly have produced sits below your total revenue, and certainly not above it. Advertising can be the lion’s share of what a store sells. It cannot be all of it.

Here is the arithmetic on an illustrative store. These figures are made up to show the shape of the problem, not benchmarks for your category.

The platforms have claimed 105% of everything the store sold. Every order is spoken for, and then some, before email, organic search, direct traffic or returning customers get a dollar. The merchant looking at three dashboards showing 650%, 610% and 475% has three plausible numbers and one impossible total.

Run the same month strictly, on seven-day click-through only, view-through off, each order credited once to the last click that brought the customer:

242% is a far less flattering number. It is also the only one you can act on. The first set tells you to increase spend everywhere, because everything looks excellent. The second tells you what your advertising is actually worth.

Shopify itself documents why the platforms disagree with each other, using an example most stores will recognise. Where a customer clicks both your email and your Google Shopping ad, Shopify’s help documentation notes that each can record its own separate conversion, while Shopify credits the sale only to whichever was clicked most recently in the past 30 days. Two platforms claim it. Shopify gives it to one.

There is also a way to see the size of that over-claim inside Shopify, though it is not on the screen you will find first, and this is worth understanding before you go looking for it.

On the Growth page, the Channel performance report offers last click, first click and last non-direct click. All three credit an order to exactly one channel, which is why those channel rows add up to your attributed sales rather than exceeding them. That is the report most merchants will open, and it will never show you an over-claim, because it is not built to.

In the Analytics reports, Shopify documents five models rather than three, and one of them behaves in the opposite way. Any click gives 100% of the credit to every channel the customer clicked, so a $200 order touched by an email, a Google ad and a Meta ad produces $600 of credit from $200 of revenue. Shopify’s own guidance is that because the model allocates more credit than orders you have received, it is best used to analyse a single channel, or to reconcile the attribution each channel reports. Along with linear, it appears on the performance by referring channel, performance by marketing activity, and performance by UTM campaign reports.

Shopify’s documentation is not entirely consistent about which models appear where, and the Growth page and the Analytics reports do not open on the same model as each other. If you cannot find any click on one screen, try the other, and check which model is selected before you compare two reports and conclude you have found a discrepancy.

The truth sits between the two numbers

The exercise above gives you a ceiling. It does not give you an answer, and the mistake most merchants make next is to swing to the opposite extreme and treat Shopify’s own channel breakdown as the truth. It is not. It is the floor.

Shopify’s marketing reports understate paid, systematically, and for five reasons that have nothing to do with how well your advertising is working.

It counts clicks and nothing else. Shopify credits a channel when it can see a click that brought the customer to the store. Any genuine influence that happened without a click is invisible to it by design. That is the right call for a fee calculation, and it still means the number is a floor rather than a measurement.

The source has to survive the whole journey. Shopify reads where a visitor came from at the start of the session and carries it to the order. Anything that breaks that chain sends the order to Direct: a redirect or link-shortening tool that drops the query string, a payment provider that bounces the shopper off your domain and returns them without the original parameters, a post-purchase upsell app that reloads the confirmation page, or simply a link somebody forgot to tag. A large Direct share is not a loyal customer base typing your URL from memory. It is mostly paid and email traffic whose source could not be read.

The browser forgets, and it forgets ad clicks fastest. This is the mechanism almost nobody accounts for. Apple’s Intelligent Tracking Prevention deletes cookies created in JavaScript, along with all other script-writable storage, after seven days without interaction with the site. Worse for advertising specifically, ITP caps that storage to a single day when the visitor arrived from a domain classified as having cross-site tracking capability and the landing URL carried a query string, which is an exact description of a paid social click into a Shopify store carrying UTM parameters. Apple’s own worked example in that post is a user clicking a link on a social network and landing on a shop. So on Safari, and on every browser on iOS, a shopper who clicks your ad on Monday and buys on Thursday can arrive as a stranger, with no record of where they came from.

Cross-device only works when the shopper identifies themselves. Somebody clicks your ad on their phone at lunch and buys on their laptop that evening. Shopify does stitch some of these journeys, and has included cross-device reporting in storefront data since March 2023. Its own worked example is a shopper who clicks an Instagram ad on a mobile, hands over an email address, and buys from a laptop days later, and it is the email address that makes the example work. Where the shopper never identifies themselves on the first device, your store sees two unrelated visitors, while Meta and Google can still join the sessions through a logged-in account. That is the one place where the platforms genuinely know something you do not, and it means a portion of what they claim above Shopify’s number is real.

One winner per order. Shopify picks a single channel per sale. Where a paid click brought the customer in and an email closed them a week later, paid receives nothing.

Why the honest number sits closer to the floor

Both numbers are wrong. They are not equally wrong, and the reason is arithmetic rather than opinion.

The ceiling has no upper bound. Every platform inflates in the same direction, and the errors add: three platforms claiming the same order produce three times the credit, and view-through crediting and long windows push each of those claims higher again. Add a fourth channel and the ceiling rises further, with no mechanism anywhere in the system to bring it back down.

The floor cannot exceed reality. Every model on Shopify’s channel report credits an order to a single channel, so its attributed total can never be more than your actual sales. Its errors are bounded, and they run in one direction only: a lost source moves an order to Direct once, and that is the worst it can do.

So treat the pair as a range rather than as two competing claims. The platform total is the largest number anybody could justify. Shopify’s channel report is the smallest. The honest figure lies between them and, on the weight of the evidence, nearer the floor, because only the ceiling has an unlimited capacity to be wrong.

Then watch the gap rather than either number. Record both once a month. If the gap widens without your media mix changing, either your tracking is degrading or a platform has loosened something, and both are worth knowing about before the quarter ends.

Where shopAds sits in that range. Deliberately near the floor. An order only counts as advertising revenue when the platform confirms it followed a click, and orders are matched using the click identifier the platform attaches to the ad click rather than by reading a referrer, which survives some of the ways a source normally gets lost. It does not survive all of them. Cross-device journeys and browser storage limits cut our number down just as they cut Shopify’s, and we have not tried to model our way back up. Our fee is charged on the conservative figure, which is the point.

The honest default: seven days, clicks only, last click, counted once

A click is a fact. Everything else is an inference. Somebody clicked or they did not. A view infers that an impression was noticed. An engaged view infers intent from a few seconds of attention. A data-driven allocation is an inference produced by a model you cannot see, built by the company being paid.

Seven days is long enough for a real decision and short enough to exclude coincidence. The further a touchpoint sits from the purchase, and the more touchpoints in between, the weaker its claim to have caused it. Extending to thirty or ninety days does not capture more causation, it captures more overlap.

Last click credits the interaction closest to the decision, and it is the only allocation rule two people can verify and agree on.

The usual objection is that last click credits the closer and ignores the introducer. That is only half right, and the half it gets wrong matters. Last click does not ignore introducing. It ignores introductions that failed to close. Where one interaction does both jobs, and a great many do, last click credits it in full and nothing is being undercounted at all.

That case is more common in ecommerce than the multi-touch literature suggests. Somebody sees a Shopping ad for a product they did not know existed, clicks, and buys in the same session. The ad introduced and closed. It is one touchpoint carrying the whole journey, and it is precisely what a direct-response ad is built to do. So the measurement and the objective point the same way: an ad that converts on the click is both better advertising and more honestly measurable. Chasing a model that spreads credit backwards over interactions that did not finish the job is, in that light, chasing credit for the wrong thing.

You do not have to take a view on how common it is, because your own account will tell you. The path length report in Google Ads attribution shows how many of your conversions came from a single interaction rather than several, and GA4’s attribution paths report can be filtered by number of touchpoints. Run it before you decide how much last click is costing you.

Counting once across platforms is the part most people miss. If two platforms both claim the same order, one of them is wrong, and adding them together makes both wrong.

This is not only the more honest setting, it is the better one to optimise on. What an automated bidding system learns from is not a count, it is a set of labels: this click led to a sale, that one did not. A click seven days before a purchase is a strong label. A click twenty nine days before, with a fortnight of email and three other ads in between, is a weak label and often a false one, and false labels teach the algorithm to bid for the wrong people. A long window also slows the feedback loop, so the system reacts to last month instead of to now. Fewer, better labels, delivered sooner, is not less signal.

The clearest published test of that comes from Maggie Humphrey of Cypress North, writing in Search Engine Land in March 2026. Her direct-to-consumer client was optimising on Google’s thirty-day default while its actual average time from click to purchase was 2.2 days, with most conversions inside a single day. They rebuilt the purchase conversion on a seven-day window, ran it as a secondary action for a fortnight, then promoted it. Reported cost fell 6.3 percent and reported ROAS rose 62.3 percent, while in the store’s own Shopify data total sales rose 20 percent and net profit rose 30 percent. Most tellingly, marketing mix modelling had Google’s incremental return climbing to 182%, a rise of 10 percent, while Meta’s fell to 59%, a drop of 25 percent, which is to say Meta was returning less revenue than it consumed in spend. Shortening Google’s window stopped it claiming delayed conversions other channels had influenced, and the real contribution of each became visible. Humphrey is careful that campaigns were being restructured at the same time, so the window change cannot take all the credit, and her stated conclusion is the defensible one: performance was not harmed and the quality of the signal improved.

Two honest costs, and one that is smaller than it is usually made out to be. It undercredits genuine hand-offs. Where a channel brought somebody in and something else finished the job days later, that channel receives nothing, and if you run substantial discovery activity you should read the number knowing that. It is not a reason to loosen the counting, which credits coincidence rather than contribution, and it applies only to the share of your sales that actually take more than one interaction.

The transition disrupts, because changing a primary conversion action resets learning phases, which is why the change above was staged over a fortnight rather than flipped. And it has to match how your customers actually buy. Seven days suits most Shopify stores selling physical products because most of those purchases are decided within a day or two, but a genuine three-week consideration cycle would make seven days wrong rather than conservative. Check your own conversion lag first, and take the shorter option whenever it is a close call.

One thing it does not tell you. Attribution answers which sales can be credited to advertising. It does not answer which would not have happened anyway. That is incrementality, and no attribution model of any kind answers it. The strongest evidence on that question comes from inside Facebook: Gordon, Zettelmeyer, Bhargava and Chapsky, in Marketing Science (2019), compared fifteen randomised advertising experiments at Facebook against the observational methods the industry actually uses, and found that in half the studies the estimated lift in purchases was out by a factor of three, with the bias running upward. That is the case against inference-heavy measurement, not an argument for it. The causal question is answered by holdout and geographic tests, or simply by turning a channel off for a fortnight and watching what your store revenue does.

How shopAds measures

shopAds runs on exactly this position, and the detail is published on the how we measure and charge page.

An order only counts as advertising revenue if the platform confirms it followed a click, and the sale itself is read from your own order record rather than from anything a platform reports. We ask each platform for the shortest click window and the least view-through crediting it allows, and switch view crediting off entirely where the platform permits. We use last click rather than any platform’s own model. When more than one platform claims the same sale, we compare the click identifiers recorded on that visit and the times they were recorded, and credit the platform whose click came last, so the sale is counted once rather than added up twice. Each day is held open for seven days to let confirmations arrive, and once a day settles its numbers never change.

Our fee is charged on that number, which is smaller than the one the platforms would give us for the same work. None of this works if the underlying tracking is broken, which is a more common problem: the Google Ads conversion tracking guide covers getting that layer right, and clean product data underneath it is what makes any of the reporting worth reading.

Frequently asked questions

Reading your numbers from here

Attribution will never be settled, because the question it tries to answer does not have a clean one. What you can control is whether the number in front of you is built from things that happened or things that were assumed.

Do the arithmetic once a quarter. Add up what every platform claimed and hold it against what your store actually sold. Watch the gap, and watch what happens to it when you shorten a window or switch view-through off. A merchant who knows the size of their own inflation is in a far stronger position than one who does not, because they are the only one of the two who can tell the difference between advertising that is working and advertising that is taking credit.

Sources

• About attribution models (Google Ads Help, 2026)
• About conversion windows (Google Ads Help, 2026)
• Get started with attribution (Google Analytics Help, 2026)
• Ads Insights API metric availability updates (Meta for Developers, 16 October 2025)
• Meta adds engage-through attribution data for social ads (eMarketer, 2026)
• How to create attribution windows for your ad group (TikTok Ads Manager, November 2025)
• Attribution overview (TikTok Ads Manager, 2026)
• Measurement pixel (OpenAI Developers, 2026)
• Marketing reports (Shopify Help Center, 2026)
• Measuring marketing performance (Shopify Help Center, 2026)
• Full third-party cookie blocking and more (John Wilander, WebKit, 24 March 2020)
• Intelligent Tracking Prevention 2.2 (John Wilander, WebKit, 24 April 2019)
• 30-day vs. 7-day attribution in Google Ads: what the shorter window revealed (Maggie Humphrey, Cypress North, in Search Engine Land, 18 March 2026)
• Google, Meta, and the long history of misaligned incentives in paid media (Jay Stampfl, Blackbird PPC, in Search Engine Land, 7 April 2026)
• A comparison of approaches to advertising measurement: evidence from big field experiments at Facebook (Gordon, Zettelmeyer, Bhargava and Chapsky, Marketing Science, 2019)