Google Ads attribution ignores offline conversions uploaded after 7 days

Offline conversions that arrive in Google Ads more than seven days after the event never reach the data-driven attribution model, while standard reporting columns record them in full, which is why one account can show two different conversion totals for the same set of events. The rule sits in the Google Ads Help Center page on attribution reports, was spotted by Hana Kobzova, and was carried into wider circulation by Google Ads consultant Adriaan Dekker in a LinkedIn post published on 22 August 2026, reported by PPC Land on 23 August 2026.

What the documentation actually says

Google frames the difference as expected rather than broken. According to the Help Center text, differences in conversion totals between attribution reports, including the Model Comparison report, and standard Google Ads reporting columns are normal, and they are caused by data processing constraints and network configurations inside the attribution system.

The mechanism is a rolling window. The daily attribution modeling engine, named in the text as Search funnels, processes conversions that occurred within the last seven days. An offline conversion uploaded more than seven days after the initial event is, in Google's words, "bypassed by the attribution modeling calculations". Standard reports carry no equivalent restriction and retroactively record uploaded offline conversions whenever they arrive.

Dekker's summary of the find, which credited Kobzova with sharing it, was that conversions uploaded after the window still appear in standard reports but are "ignored by the attribution engine".

The arithmetic of a single lead

The worked example in the reporting is the clearest way to see the gap. A lead clicks a search ad on the first of the month. The deal closes on the twelfth. The conversion is uploaded on the fifteenth, which is fourteen days after the click.

On the Campaigns page, that conversion appears. In the Model Comparison report, it does not exist. Neither number is wrong and neither report is broken. They are counting under different rules, and the account holds both numbers at the same time.

Google's stated remedy in the same passage is a target rather than a fix. Keeping offline conversion upload latency below seven days, the documentation says, allows the modeling engine to fold those offline journeys into its daily calculation suites, which maintains model accuracy and supports automated bidding efficiency.

The wording describes the gate twice, and the two tests are not the same

This is the most valuable part of the story and it is also the part that stays unresolved. Two statements sitting in the same bullet describe the gate differently.

The first keys the rule to the age of the conversion: the attribution engine only processes conversions that occurred within the last seven days. The second keys it to upload latency: an offline conversion uploaded more than seven days after the initial event is bypassed.

Those two tests agree in the easy cases and split in ordinary ones. A conversion that occurred six days ago and is uploaded today satisfies both tests. A conversion that occurred fourteen days ago but was uploaded two days after the event satisfies the latency test while falling outside a strict seven-day event window. Whether the engine looks at the event date, the upload delay, or both, the documentation does not say, and nothing in the reporting resolves it.

What each reading would imply

Under the latency reading, what matters is the speed of your pipeline. A business-to-business advertiser whose deals close after six weeks can still feed the model, provided the upload follows the close quickly. Under the event-age reading, the same advertiser cannot, because the conversion is already older than seven days by the time it exists at all, and no amount of pipeline speed changes that.

The divergence is sharper for historical backfills. A one-off upload of six months of closed deals satisfies neither reading on the event-age test, and fails the latency test for almost every row in the file. Under either interpretation, a backfill is a reporting exercise rather than a modelling input. That is the one conclusion the two readings share, and it is the safe one to plan against until Google clarifies which test the engine applies.

What the Model Comparison engine leaves out

The seven-day upload gate is not the only exclusion. Under a heading covering network and campaign type exclusions, the documentation states that the Model Comparison engine is designed to explicitly exclude conversions originating from the Search Partner Network, Gmail and App campaigns. Standard account reporting columns include conversions from those networks by default.

That matters most for anyone using attribution reports to decide where budget goes. Conversions from Search Partner inventory sit outside the modelling engine while appearing in standard columns, which puts any partner network return assessment drawn from an attribution report on incomplete ground.

Time of event against time of query

A third source of divergence has nothing to do with offline data. Attribution reports count by time of conversion. The Campaigns page counts by the time of the ad query that preceded the click that led to the conversion.

Those are different days for any conversion that does not happen immediately after the click, which is most of them in considered-purchase categories. The documentation names the fix: adding by conversion time columns to the Campaigns page aligns the two counting conventions. That is a report configuration change, not a data change, and it removes one of the four sources of disagreement without touching anything else.

Four structural differences, side by side

The comparison table published alongside the Help Center passage sets out where the two views diverge. The rows below reproduce that comparison as reported.

Read together, an advertiser reconciling a Model Comparison figure against a Campaigns page figure is comparing outputs that differ by timing convention, network inclusion, campaign type inclusion and upload latency at the same time, before any question of which attribution model to pick even arises.

How the seven-day gate reaches Smart Bidding

Data-driven attribution feeds Smart Bidding as well as the reports on screen, and the documentation makes that link itself when it ties upload latency to model accuracy and automated bidding efficiency in the same sentence.

Offline conversion import exists precisely because the conversion is separated in time from the click. The click identifier is captured at click time, stored alongside the business record, and uploaded once the outcome is known, which can be days or weeks later. In business-to-business pipelines, a fortnight between click and closed deal is unremarkable. Under the documented gate, those conversions reach standard reporting and bypass the model, which means the bidding system is optimising against a subset of the outcomes the account can actually see.

This lands in the middle of a long sequence of changes to how offline data enters Google's advertising stack: enhanced conversions for leads in March 2022, a mandatory conversion environment parameter for offline uploads on 6 February 2025, the Data Manager API launched on 9 December 2025 as a unified ingestion point, and the 15 June 2026 block on new offline conversion import adopters using the older Google Ads API path. Every one of those addressed how offline data gets in. None addressed what the attribution model does with it once it is inside.

Report controls that bound what you are looking at

Two further limits are worth knowing before drawing conclusions from the attribution tab. Date ranges reach back two years, and path report and attribution credit data older than two years is deleted and cannot be restored. The lookback window control adjusts to 30, 60 or 90 days, with most reports opening at 30, and the Model Comparison report adds a Default setting that matches the lookback window to the conversion window configured for each conversion action.

The documentation also separates two terms that get conflated constantly. A lookback window sets how far back from a conversion an ad interaction remains eligible for credit. A conversion window is the period after an ad interaction during which a conversion is recorded at all, and any conversion without a preceding ad interaction inside its conversion window is not counted in Google Ads in the first place.

What to check in your own account

Four checks follow from the documented rules, and none of them require a guess about which reading of the gate is correct.

1. Measure your actual upload latency. Pull the gap between event timestamp and upload timestamp for a recent month of your offline conversion feed, and look at the distribution rather than the average. If a meaningful share of rows lands past seven days, the model is working from less than your reports show.
2. Stop comparing the two reports as if they should match. If a performance review puts a Model Comparison number next to a Campaigns page number, the difference is expected by design across four dimensions. Add by conversion time columns to the Campaigns page to remove the timing dimension, and treat the rest as structural.
3. Work out whether Search Partners, Gmail and App traffic is material to you. Those conversions are excluded from the Model Comparison engine and included in standard columns. If they are a large share of volume, an attribution report is a poor basis for moving budget between networks.
4. Treat historical backfills as reporting, not training data. Under either reading of the gate, a bulk upload of old conversions is unlikely to reach the model, so do not expect a backfill to improve Google Ads bidding performance.

What this means for Thai marketers

Neither the documentation nor the PPC Land report mentions Thailand, so what follows is reasoning about local conditions rather than anything the source stated.

The first point is about sales cycle length. Categories that lean heavily on offline conversion import in Thailand, including property, motor vehicles, private education, insurance and business-to-business services, routinely run weeks between the ad click and the closed deal. A property enquiry that turns into a booking after two site visits is nowhere near a seven-day cycle. Those are exactly the accounts where the documented gate bites hardest, and they are also the accounts most likely to be running Smart Bidding on imported conversions.

The second point is about how the conversion actually gets recorded. A large share of lead handling in Thailand happens on LINE and by phone rather than inside a web form, so the close is logged by a salesperson in a CRM or a spreadsheet and uploaded later. Upload latency in that setup is not a technical property of the tag, it is a function of how quickly a human updates a record. If the sales team updates the CRM weekly, the pipeline is already at or past the seven-day line before anyone touches Google Ads.

The third point is about reporting discipline. Where an in-house team and an external partner both pull numbers, two people looking at the same account can produce two totals honestly. Agreeing in advance which report is the reference for performance decisions, and configuring the Campaigns page with by conversion time columns, removes an argument that has no winner.

The fourth is a measurement hygiene point. Before blaming attribution for a gap, it is worth confirming the underlying tracking is sound, because a broken or half-migrated measurement setup produces discrepancies that look identical to this one. A clean Google Analytics 4 setup does not fix the seven-day gate, but it removes the competing explanations.

What the article does not resolve

• Which test the engine applies. The event-age reading and the upload-latency reading both appear in Google's own wording, and the documentation does not say which governs.
• Whether the Help Center passage is new. The report notes the page carries no publication date for the added section, so it is not known how long the rule has been documented.
• Whether Google plans to change the rule. Nothing in the reporting indicates a fix, a rollout or a timeline.
• Any figure for how much conversion volume is affected. No percentage, no account level data and no market level estimate appears anywhere in the reporting.

FAQ

Are my offline conversions being lost?

No. Conversions uploaded more than seven days after the event are still recorded in full by standard Google Ads reporting columns, which record uploaded offline conversions retroactively regardless of when they arrive. What they do not reach is the data-driven attribution model, so they are missing from attribution reports including Model Comparison rather than missing from the account.

Why does my Model Comparison total differ from my Campaigns page total?

Because the two are counting under different rules, and Google describes the difference as normal rather than faulty. The documented differences cover four dimensions at once: attribution reports count by time of conversion while the Campaigns page counts by time of the ad query, attribution reports exclude Search Partner Network, Gmail and App conversions, campaign and video format coverage differ, and offline conversions uploaded after seven days are bypassed by the modelling calculations.

Is the seven-day limit about when the conversion happened or when I uploaded it?

The documentation says both and does not say which one governs. One statement keys the rule to conversions that occurred within the last seven days, the other keys it to an offline conversion uploaded more than seven days after the initial event. Those tests give different answers for a conversion that happened fourteen days ago but was uploaded promptly, and the source does not resolve the inconsistency.

Will a historical backfill of old conversions improve my Smart Bidding?

Under either reading of the documented rule it is unlikely to, because the conversions in a backfill are old, the uploads are late, or both. The reporting does not test this directly, so treat it as the conservative conclusion the two readings agree on rather than a confirmed outcome.

What should I do first?

Measure the gap between event time and upload time across a recent month of your offline conversion feed. That single number tells you whether the gate applies to your account at all, and it is the only one of these questions you can answer without waiting on Google.