What the two settings decide
Google documents conversion tracking as a record of what people did after interacting with an ad. Every conversion it records passes two tests before it appears in a report. It must fall inside a window: the period after the ad interaction during which the account will still credit it. And if several ad interactions preceded it, the attribution model decides how the credit is shared among them.
The two settings answer different questions. The window asks whether a conversion counts for ads at all; a conversion that arrives after the window has closed is not counted, however much it owed to the ad. The model asks where the credit goes, between campaigns, ad groups, keywords and ads. Change the window and totals rise or fall. Change the model and totals stay close while credit moves between campaigns.
Both feed bidding. The Smart Bidding strategies guide notes that the model and the window decide how many conversions and how much value each campaign is credited with, so a target set under one combination needs revisiting when either changes. That is why these settings belong with the people who set targets, not only with whoever installed the tags.
| Row label | Attribution model | Conversion window |
|---|---|---|
| Question it answers | Which interactions share the credit | Whether a conversion is credited to ads at all |
| Set where | Per conversion action | Per conversion action, separately for clicks, engaged views and views |
| Effect of a change | Credit moves between campaigns; totals change little | Totals rise with a longer window and fall with a shorter one |
| Effect on bidding | Changes which campaigns and keywords look valuable | Changes how many conversions the strategy learns from, and how late |
Attribution models in Google Ads
The measurement hub describes the two models Google Ads currently offers.
Data-driven attribution uses the account's own conversion paths to distribute credit among the ad interactions that preceded a conversion. A purchase that followed a generic search click and, days later, a brand search click can give part of its credit to the generic click. The model learns from the account's data, so the shares it assigns are not published as fixed rules.
Last click gives all the credit to the final ad click before the conversion. It is simple, stable and easy to explain, and it systematically favours whatever people click last, which is often a brand search.
The conversion tracking setup tutorial notes that data-driven is the default where the account has enough data, with last click otherwise. Both models distribute credit only among Google Ads interactions; channels outside the account are invisible to them, which is the main reason the platform and analytics disagree, as the comparison of Google Ads and GA4 figures explains. The attribution models framework sets out the wider family of models and what each assumes.
Choosing a model for each action
The model is set per action, and the choice is mostly about what the action is used for.
Primary actions that drive bidding generally suit data-driven attribution where it is available, because it credits the earlier clicks that start journeys, and a strategy that sees that credit can bid for them. Under last click, an account whose customers search generically first and by brand name later will see its generic campaigns starved of credit, and bidding will follow the credit.
Actions read mainly for a simple comparison, such as a secondary action kept to check against another system, may be better on last click, which behaves predictably and changes only when behaviour changes.
All the primary actions used by one kind of campaign should share a model. A campaign whose goals mix a data-driven action and a last-click one reports a sum of two different kinds of credit, and its cost per conversion is not comparable from month to month if the mix shiftsA change in a blended metric caused by a change in the mix of segments, such as devices or sources, rather than by any segment performing differently..
General practice is to settle the model once, deliberately, and leave it. Every change moves credit, resets comparisons and changes what bidding learns, so a model that is switched whenever a report disappoints never produces a trend anyone can read.
Conversion windows: clicks, engaged views and views
Google Ads sets windows per conversion action and separately for three kinds of interaction, as the measurement hub describes, each with a maximum the help page documents.
The click-through window is how long after an ad click a conversion can be credited to it. It is the window that matters most for Search and Shopping, and the one that should follow the sales cycle.
The engaged-view window applies to video. The YouTube hub describes an engaged-view conversion as one that follows a viewer watching a documented portion of a skippable ad without clicking.
The view-through window applies when a person saw a display or video ad, did not interact with it, and converted later. A View-Through ConversionA conversion credited to an ad that was shown but not clicked, within a set view window. is the weakest evidence of the three, because seeing an ad and buying anyway is common; the view-through versus click-through comparison recommends reporting such conversions on their own line with their window stated, not mixed into the main figure.
The lookback window entry compares how other platforms and analytics tools set their windows, which matters when Google Ads figures are read beside them: a window that runs forward from the click in one system and backward from the conversion in another will never count the same set.
Setting windows from your own conversion lag
A window should match how long real customers take, which is the account's Conversion LagThe time between an ad interaction and the conversion it leads to, which leaves the most recent reporting days incomplete., not a default inherited from setup. The conversion lag entry explains how to measure it: from the platform's time-to-conversion reporting or from the CRM, find how long it takes for most conversions from a day's clicks to arrive.
The reasoning then runs in three steps. First, read the distribution: how many conversions arrive within a day, within a week, within a month and later. Second, set the click-through window to include the bulk of that distribution, so the account credits the conversions its ads actually started; a window shorter than the typical lag undercounts every campaign and teaches bidding from a biased subset. Third, look at the tail. A window long enough to catch every late conversion also credits actions that owe little to the ad, which the hub names as the cost of a long window, so the tail is better handled by the business's own records than by stretching the window to reach it. The academy exercise on setting windows from a piano retailer's own conversion lag practises this reasoning for each conversion action before reconciling a closed month with the retailer's sales records.
Lead businesses with long sales cycles face the hardest version. A signed contract may follow the click by weeks or months, beyond the reach of any sensible window. The hub's advice is the longest click window available for the enquiry action, with offline import carrying the final outcome, so that bidding learns from enquiries promptly and from contracts as they close. The B2B measurement guide covers the reporting side of long cycles, cohorts and leading indicators, beyond what one ad platform can see.
Reporting by click date and by conversion time
Google Ads places each conversion, in its standard columns, on the date of the click or interaction that led to it, not on the date it happened. The measurement hub notes the consequence: recent days keep filling in after the fact, and reports that compare conversions by click date and by conversion time exist because the window makes recent periods look weaker than they will become.
| Row label | By click date (standard columns) | By conversion time |
|---|---|---|
| A conversion is placed on | The day of the click or interaction | The day the conversion happened |
| Recent periods | Incomplete until the window closes | Complete for the day, but credited to older spend |
| Cost and conversions | Matched: the same clicks' cost and results | Mismatched: today's cost beside conversions from older clicks |
| Best for | Judging campaigns, keywords and bids | Comparing with orders and leads in business systems |
Use click date for anything that sets cost against results, because only then do the spend and the conversions belong to the same clicks. Use conversion time when matching the account against the business's own records, which are dated by when the order or lead arrived; the tutorial on reconciling with sales records does exactly that. Never compare a recent click-date figure with a closed period, which always flatters the past. When a fall in conversions has to be told apart from days still filling in, the diagnosis for Google Ads conversions that dropped suddenly sets this reading beside the other causes, from changed primary actions to broken tracking.
Changing settings without misreading the result
A change to a model or window changes the numbers without any customer behaving differently, so the change itself has to be visible and its effect separated from everything else.
Change one setting at a time. A new window and a new model on the same day produce one shift with two causes. Leave bid targets and budgets alone in the same week, for the same reason.
Record it. Note the action, the old and new settings, the date and the reason in change history. The view-through comparison makes the same point for reports: a window change breaks the trend line at that date.
Expect bidding to respond. The hub notes that a model change alters what Smart Bidding learns from, and Google's Smart Bidding documentation lists changes to conversion actions among the changes that can start a learning period. Treat a settings change as a bidding event: avoid sales and holidays, and judge the campaigns after they settle.
Read the result only after the new window has closed. Compare a period whose clicks fall entirely under the new settings with an equivalent period under the old ones, each read only once its own window has finished. How far a change applies to data already recorded is set out on the help page; until you have checked, treat the change date as a break in the series rather than assuming earlier figures were restated.
Revisit targets. A longer window credits more conversions, so a target CPA met under the old window will look easier under the new one without anything improving. Reset targets on the new basis, from figures read after the window closed. The checklist for changing a conversion goal gathers these steps into a list to tick before a model or window is changed. The academy checkpoint on conversion windows, reconciliation and offline uploads tests reading a month split by a settings change, alongside reconciling a closed period and checking an offline upload.
Settings for a slow-buying landscaping firm (illustrative)
Gravelquill Landscaping lengthens its click window
Gravelquill Landscaping, an invented garden-design firm, counts a design consultation request as its primary conversion, on data-driven attribution with a 30-day click-through window. Its CRM shows more paid-search enquiries than Google Ads credits, so the owner exports the delay between first ad click and enquiry for every CRM lead that carries a click identifier over the previous six months.
| Days from click to enquiry | Share of enquiries |
|---|---|
| Same day to 7 days | 46% |
| 8 to 30 days | 33% |
| 31 to 60 days | 16% |
| More than 60 days | 5% |
A 30-day window was missing about a fifth of the enquiries the ads had started, mostly from people planning a garden over a season. The owner changes only the click window, to 60 days, on the first day of September, notes it in change history, and leaves the model, budget and target untouched. The comparison is timed so both sides are complete: March clicks under the old window, read in early May, against September clicks under the new window, read in early December, both months having similar spend and season.
| Row label | March, 30-day window | September, 60-day window |
|---|---|---|
| Spend | £4,200 | £4,300 |
| Enquiries credited by Google Ads | 61 | 77 |
| Paid-search enquiries in the CRM | 80 | 81 |
| Cost per credited enquiry | £68.85 | £55.84 |
Credited enquiries now sit much closer to the CRM, and the apparent fall in cost per enquiry is a measurement change, not an improvement, so the owner resets the target CPA from £70 to £58 rather than letting the strategy spend into the gap. Contracts, which follow enquiries by months, are left to an offline import rather than a longer window.
Common mistakes
- Leaving windows at whatever setup chose. The window should follow the account's own lag.
- A window shorter than the sales cycle. Every campaign looks worse, and bidding learns from the fastest buyers only.
- Mixing models across one campaign's primary actions. Two kinds of credit summed into one figure.
- Switching models whenever a report disappoints. Credit moves, comparisons break and nothing is learned.
- Changing window, model and target together. One shift, three causes.
- Reading the effect before the new window has closed. The new figures are still filling in.
- Comparing click-date figures with business records. Use conversion time for that comparison.
- Keeping the old target after a longer window. The target is met more easily without anything improving.
- Mixing view-through conversions into the main figure. Report them on their own line.
