GA4 Conversion Window: Which Value to Set (BigQuery)

Since August 11, 2026, GA4 lets you type any integer as your conversion window: 1 to 90 days for click-through, 1 to 30 days for engaged-view. The 1/7/14/30/60/90 ladder is gone, and so is the 3-day ceiling that was clamped on video views. Good news. Except Google removed the constraint without offering a single recommendation. The value you enter is no longer a preset you put up with, it is a decision. And a decision with no method behind it is just another number picked out of thin air. This article gives you the method: measure your real click-to-conversion delay in BigQuery, then set the window to it. Plus the three traps the news briefs never mention.

What exactly changed on August 11, 2026

Two settings became configurable, and neither is cosmetic. The click-through conversion (CTC) window, which used to accept only six values, now takes any integer. The engaged-view conversion (EVC) window, locked at 3 days for every property with no exception, now opens from 1 to 30 days.

The setting lives in two places, which matters when the GA4 property and the Google Ads account are run by different teams. On the GA4 side: Advertising > Conversion management > more options icon (⋮) > Settings. On the Google Ads side: inside the linked account’s conversion management interface. Same value, two doors. The official rationale fits on one line in the release notes, with no blog post and no press release: align the window with your business cycle. No suggested value. The responsibility for the number now rests entirely on whoever enters it.

Trap 1: the conversion window is not the key-event lookback

This is the confusion that costs the most, because it goes unnoticed. You have two separate settings, two numbers, two effects. The conversion window we are discussing here is the Google Ads-facing setting: how long after a click (or a view) a conversion is still credited to the campaign. The key-event lookback, by contrast, is GA4’s internal attribution setting: how far back GA4 looks to distribute credit across touchpoints in its own reports. By default, that lookback is 30 days for acquisition events (first_visit, first_open) and 90 days for other key events.

Mixing the two up is the number one cause of discrepancies between GA4 and Google Ads. If you want to understand the internal attribution mechanics and what Google changed in April 2026, read GA4 Attribution Changes in 2026 first. This article touches only the Ads-facing setting, not the lookback.

The method: measure your real delay in BigQuery

Nobody should pick 30 days out of habit. Your GA4 export holds the exact answer: the distribution of the delay between the first touch and the conversion, for each of your key events. The query below returns the median, p75, and p90 in days using APPROX_QUANTILES. If you have not wired up the export yet, start with working with the GA4 BigQuery export: it lays the groundwork this query assumes.

-- Distribution of first-touch -> conversion delay, per key event
-- Adjust the project, dataset, and date range.
DECLARE key_events ARRAY<STRING> DEFAULT ['purchase', 'generate_lead'];

WITH first_touch AS (
  -- Each user's first touch in the period
  SELECT
    user_pseudo_id,
    MIN(event_timestamp) AS first_ts
  FROM `project.analytics_XXXXXX.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20260827'
  GROUP BY user_pseudo_id
),
conversions AS (
  -- Every occurrence of a key event
  SELECT
    user_pseudo_id,
    event_name,
    event_timestamp AS conv_ts
  FROM `project.analytics_XXXXXX.events_*`
  WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20260827'
    AND event_name IN UNNEST(key_events)
)
SELECT
  c.event_name,
  COUNT(*) AS conversions,
  -- Delay in days (event_timestamp is in microseconds)
  ROUND(APPROX_QUANTILES((c.conv_ts - f.first_ts) / 1e6 / 86400, 100)[OFFSET(50)], 1) AS median_d,
  ROUND(APPROX_QUANTILES((c.conv_ts - f.first_ts) / 1e6 / 86400, 100)[OFFSET(75)], 1) AS p75_d,
  ROUND(APPROX_QUANTILES((c.conv_ts - f.first_ts) / 1e6 / 86400, 100)[OFFSET(90)], 1) AS p90_d
FROM conversions c
JOIN first_touch f USING (user_pseudo_id)
WHERE c.conv_ts >= f.first_ts
GROUP BY c.event_name
ORDER BY conversions DESC;

The cost. The query scans only three columns (user_pseudo_id, event_name, event_timestamp) and filters on _TABLE_SUFFIX, so the volume stays low. Always preview the estimated scan in the BigQuery editor before running, it shows at no charge. This query is a good candidate to join your library alongside the 10 essential BigQuery queries for GA4.

An honest caveat: this version measures the delay between the user’s very first touch and their conversion, which is a clean, robust proxy for the click-to-conversion delay. To pin it to the exact paid click you would join on the session traffic source, but for deciding a window, this distribution is more than enough.

The decision rule: set it to the p90, round up

Once the distribution is in front of you, the rule is simple: take the p90 of your delay, round up, and enter it. The p90 captures 90% of your attributable conversions without inflating the window with the long tail of edge cases. A concrete example: if your purchase has a median of 4 days but a p90 of 26 days, a 14-day window silently costs you one conversion in ten. Move up to 30. Conversely, if the p90 lands at 9 days, staying at the default 30 credits clicks that probably had nothing to do with it: come back down.

Above all, one value per key event. The per-conversion setting has been available since January 2026: a lead form and an e-commerce transaction do not share a cycle, so they do not deserve the same window. Measure each event, decide each event.

The engaged-view case: why your YouTube count will rise

EVCs count a viewer who watched a qualifying portion of your video without clicking, then converted afterwards. The thresholds: at least 10 seconds for skippable in-stream (or the full duration if the format is shorter), 5 seconds for in-feed and Shorts. Until August 11, that credit expired after 3 days, regardless of your property configuration. In other words: someone watches your brand film, thinks it over for a week, buys, and the video earned no credit at all. Every video budget was judged against a metric capped at 72 hours.

Push the window to 30 days and the EVC count will mechanically climb. Watch the misreading: this rise is not incremental. It is a reattribution. The credit your videos are about to reclaim was already counted elsewhere, taken by last-click search and direct. You are not creating conversions, you are redistributing their authorship. Say it plainly internally, before someone presents the rise as proof of YouTube effectiveness.

Trap 2: retroactivity is undocumented

The release notes say nothing about retroactivity. Yet the usual behavior of GA4 lookback windows is to be non-retroactive: the change applies from the edit date forward, not backward. If that logic holds, then the day you widen your window, you create a dated discontinuity in your conversion series: before, counted under the old rule; after, counted under the new one. Two definitions inside the same curve.

The practical consequence: annotate your dashboard on the exact date of the change, and never compare year over year across that date. A YoY that straddles the switch compares apples to oranges. This is the kind of unexplained gap that triggers a bug hunt where there is only a setting. If you are auditing your configuration, add this point to the list in GA4 Audit: 11 Configuration Mistakes.

Trap 3: do not change the window the same week as anything else

Timing is treacherous right now. On August 17, 2026, Google Ads tightened how bidding targets are enforced, a change that, according to some advertisers, forces you to double your stated target CPA to hold the same delivery volume. And the number of reported conversions feeds Smart Bidding directly. If you change your conversion window that same week, you touch the denominator of every efficiency metric at the precise moment the bidding system changes behavior. The result: an unattributable effect. You will never know what moved what.

The rule: a 4-week freeze window around the change. One lever at a time. It is the same discipline that applies to any media steering, as in GA4 Cross-Channel Budgeting: you isolate the variables or you measure nothing.

To place the move in context: Meta went the opposite way from Google. It removed the 7-day and 28-day view-through windows from the Ads Insights API (effective January 12, 2026), then redefined click-through to link clicks only in March 2026. One side widens the range of values, the other narrows it. Comparing YouTube and paid social therefore happens across two measurement systems whose flexibility is diverging. Worth keeping in mind whenever a report puts the two channels side by side.

The checklist before you touch the setting

Five actions, in order. One, measure your real delay in BigQuery with the query above. Two, decide one value per key event, set to the rounded-up p90. Three, document the exact date of the change somewhere durable. Four, annotate the dashboard on that date to cut off any YoY that straddles it. Five, recheck your Smart Bidding once the dust settles, keeping the August 17 switch in mind.

The real win from August 11 is not having 90 values instead of 6. It is finally being able to match the setting to the data. You have the data in BigQuery. Use it, and stop typing 30 days because it happened to be the middle box.