What Google's Conversion Lift Update Asks of Your Tracking

As of November 2026, Google Ads has rewritten its "Set up Conversion Lift based on users" help page, and the change is a tracking audit rather than a new feature: the study mechanics are the same, while the requirements a conversion setup must meet before Google will use its data are written down in far more detail. An advertiser that cannot show hashed first-party data, separate app conversion actions and a recent history of firing conversion actions will spend holdout and budget on a study whose result Google discards. A sem agency working on Google Ads accounts in Singapore now has to answer that readiness question before it proposes an incrementality test.

The short answer: audit tracking before you book a study

Conversion Lift measures the incremental conversions, site visits and other actions that ads drove, by holding back part of a randomly split audience and comparing it with the exposed group. Nothing in the November update changes how that comparison is run. What changed is the list of conditions under which Google will accept a conversion action's data into the calculation.

Search Engine Roundtable reported in mid-November that the revised document covers hashed customer data requirements for enhanced conversions, the insufficiency of click-ID-only offline imports, unsupported setups such as Google Analytics web conversion imports, more specific consent mode guidance for EEA, UK and DMA campaigns, app conversion requirements across Android, iOS, Firebase and third-party attribution partners, and data quality rules including the 7 to 14 days of continuous conversion data.

Read as one document, those points describe a data pipeline. The experiment design sits on top of it, and no amount of clever study configuration repairs a pipeline that sends Google data it will not match.

Key takeaways

• The known thresholds did not move: a campaign needs at least 1,000 observed conversions and a campaign budget of at least $5,000 USD before it can be studied.
• Data quality is now spelled out. Conversion actions must have been firing with valid enhanced conversions and consent mode for at least 7 to 14 days before launch, and API uploads should arrive within 24 hours of the event.
• Offline imports without hashed personal data no longer qualify. Google's own page states that GCLID-only imports cannot be used, because the control group is never served an ad to click on.
• App and web conversions must be separate conversion actions. When Google detects both sources on one action, Conversion Lift drops all of that action's data.
• The decision for most advertisers is sequencing: fix the tracking, then test, or accept that the attribution figure is the only number available until the setup is repaired.

Where Conversion Lift sits next to your normal reports

Every paid-media report answers a question about credit. Conversion Lift answers a question about cause. The first asks how many conversions the platform attributed to the ads; the second asks how many happened because of the ads and would not have happened without them.

An advertiser used to reading return on ad spend should be clear about which one they are looking at. A report that divides signed-deal value by ad spend includes brand-name searches, existing demand and repeat purchases in the numerator. The platform's own attribution model assigns those conversions using clicks and fingerprints. Neither number tells you what the holdout group delivered, which is the only figure a lift study produces.

That distinction matters when a budget decision is being debated. If the account cannot run a study, the attribution figure remains the working number, and it should be labelled as such in the report rather than retitled "incremental".

What the account has to clear before it can be studied

The account and campaign thresholds

Google's page opens with the account threshold, and it has not changed. The account needs at least 1,000 observed conversions, and conversions that carry supplementary data do not count towards that minimum. The study requires a minimum campaign budget of $5,000 USD. At least one conversion action in the account must be compatible with Conversion Lift.

Campaign eligibility is narrower than the account check. Display, Search, Video, Demand Gen, App and Performance Max campaigns can take part. App campaigns targeted at iOS and Travel Ads cannot. A campaign can sit in only one study at a time, whether that is Brand Lift, Search Lift or Conversion Lift, so a campaign already enrolled elsewhere has to be removed first.

That last rule trips up accounts that run continuous testing: the same Search campaign cannot carry a Brand Lift study and a Conversion Lift study at the same time.

Study power, and why 90% keeps coming up

Google calls its feasibility estimate study power. The engine expresses it as a range with stepped increments, and it expresses the likelihood that the study will measure the true causal impact rather than noise.

The guidance is direct. A study power below 90% gets budget guidance from the tool aimed at reaching 90% certainty of lift. Results between 50% and 90% can be used as directional, and a setup above 90% is where the most reliable results sit. The dials that move the number are the selected conversion actions, the daily campaign budget, the study duration, the traffic split and the account's own campaign history.

For a small advertiser the practical reading is blunt: if the tool asks for more budget than the campaign can carry, the study is not ready regardless of how good the tracking is, and a google ads singapore account run on a modest monthly budget will often land here before it lands anywhere else.

Study duration and the cost of running short

The page sets a floor and a recommendation. Studies can run as short as 7 days, but running for more than 14 days is recommended. The reason is conversion lag, the average time between an impression and the conversion it drives.

Where lag is long, a short study loses data at the tail. Google states that studies with a long conversion lag that run for less than 14 days can experience up to a 17% drop in absolute lift. Quick purchase cycles such as food ordering can be studied in a short window; expensive, long-considered products cannot.

A sem agency client selling consultation-led services should read its own lag before booking dates. If the average lead closes weeks after the enquiry, a 7-day study window does not capture the conversions it was meant to measure.

Holdout size and the cost of being cautious

Holdouts can go up to 50%. A larger holdout increases the sample and shortens the time needed to detect a difference, but it also raises the opportunity cost, because the control group is never shown the ad. A smaller holdout needs a longer study to collect enough sample.

Google's own guidance on frequency is worth quoting in a budget meeting: based on historical data, most advertisers run about 1 to 2 studies per year. Every study carries that opportunity cost, so the sensible pattern is testing around a major budget decision or validating a hypothesis that a media mix model produced.

There is also a selection rule that decides what the result means. Including every campaign of the same type, audience and targeting strategy measures the value of that whole channel against no ads at all. Measuring only a subset tells you the value of those specific campaigns while others keep running, and it reduces the chance of detecting lift, because holdout users can still meet your unmeasured ads.

The tracking checklist Google now spells out

The bulk of the November revision sits here. Google lists each issue, the condition it flags, and the action it recommends.

Three of these deserve attention because they fail quietly.

Phone-only or single-identifier setups rarely reach the matching thresholds, so Google recommends sending email alongside phone and physical address on every enhanced conversions setup. Offline imports that contain no personal data never enter the holdout comparison, because a member of the control group has no click to match against. And the order or transaction ID has to be identical in the website tag and in the API upload; a mismatch between the two leaves server-to-server conversions unmatched even though both systems report success.

Which setups Google will reject outright

The revision adds an explicit list of configurations that Conversion Lift does not support, even when the diagnostics page reads "Good". This is the part most likely to be missed, because normal click-based attribution accepts all of them.

• Google Analytics web conversion imports, including Google Analytics user-provided data, are not supported in a user-based study. Native Google Ads website conversion actions with enhanced conversions are the supported route.
• Offline imports carrying only a click ID cannot be used. Upgrading them to enhanced conversions for leads and passing hashed personal data is the fix.
• Importing BRAIDs together with personal data makes the personal data unusable for the study, so BRAIDs should be kept out of the actions a study will use.
• Custom conversion variables on enhanced conversions for leads uploads are not processed.
• Multi-source conversions add extra identifiers such as IP address or device data that the study ignores; website tags and hashed offline imports remain the supported path.

There is a consent twist for advertisers selling into Europe. Google's page states that basic consent mode is insufficient for Conversion Lift, so a campaign aimed at EEA, UK or DMA regions needs advanced consent mode before it is studied.

A readiness sequence for a Singapore account

The order below follows the document's own logic and the author's experience of accounts that arrive with a CRM, a tag manager and several conversion actions of uncertain provenance.

1. Inventory the conversion actions. List every action that fires, note whether it is a website tag, an API feed or an app event, and mark the ones a study would select.
2. Separate the app. If any action receives both app and web conversions, split it and re-run diagnostics before anything else.
3. Confirm the identifier chain. Each offline or CRM stage change needs a stable identifier and hashed personal data attached; an import that carries only a click ID is outside the study.
4. Check consent and tag serving. For EEA, UK or DMA targeting, confirm advanced consent mode. For every account sending web conversion traffic, confirm the share routed through a first-party domain or CDN against the 75% to 80% threshold.
5. Let the data run. Keep the selected actions firing with valid enhanced conversions and consent mode for at least 7 to 14 days before launch, and confirm API uploads land within 24 hours.
6. Read study power, then book dates. If the estimate falls under the target, take the budget guidance or extend the study before committing a holdout.

The consent and pre-study steps are the ones a sem agency usually has to own, because they span the website, the CRM and the tag deployment, three places no single media buyer controls.

What this changes for google ads singapore advertisers

The accounts that will feel this update hardest are consultation-led businesses: clinics, professional services, interior firms, B2B suppliers. Their conversions close in a CRM days or weeks after the enquiry, their tracking often leans on a click identifier passed through a form, and their budget rarely reaches the level a study needs.

For them the result is not a broken experiment but a clearer order of work. The same requirements that make a study possible, hashed first-party data, timestamps on offline events, separate actions per environment, also improve the platform's conversion-based bidding, which adjusts bids towards the conversions it can see. Fixing the data changes how the account bids before any study runs. The question a marketing lead should bring to the budget meeting is therefore not "can we run a Conversion Lift study" but "how far is our conversion data from being usable if we wanted one".

Where Topkee sits in that picture

Topkee is the attribution layer, not a Conversion Lift provider. Its work covers the data prerequisite a study needs, and Conversion Lift answers a causal question that attribution cannot, so the two sit beside each other rather than on top of one another.

Topkee Media's account work runs on a single measure: return on ad spend calculated as signed-deal value divided by ad spend, reported each period from Google Ads, Yahoo Ads, Facebook Ads and GA4 in one report. Each lead carries a TTCLID identifier holding the UTM parameters, and stage changes are sent back to Google Ads Conversions through TTO TAG, so an offline deal can be matched to the click that produced it. That is the identifier chain a lead-based study reads.

TTO TAG manages ad accounts, tags and conversion events in one place, with one-click conversion event setup and sales-stage events relayed to Google Ads Conversions, Google Analytics, Meta Pixel and Bing UET. Site-wide tagging through TAG or GTM adds the Google tag across a site and exposes tag parameters for deeper conversion tracking, which is the same plumbing that decides whether a web conversion action can carry enhanced conversions.

What these services cannot do is state a causal result. Attribution return on ad spend, including the agency's reported cases, counts brand searches and repeat purchases in the numerator, and the agency's own reporting materials treat it as a credit measure, not an incrementality one. A holdout study measures the difference the ads made. A cleaner attribution chain prepares the data for that study; it does not replace it. Where the measurement window itself is the problem, the same constraint applies: platforms accept back-end CRM conversions only within fixed windows, so conversions outside them never reach platform training and have to be read from the advertiser's own end-to-end reporting.

Advertisers who want to see how channel reports are defined before building one can read the agency's Google Ads channel report definitions and its brief on AI-generated content and search behaviour.

How often to test, and what to do with the result

Google's frequency guidance is a good planning anchor: 1 to 2 studies a year for most advertisers, aligned to budget cycles and used before a large reallocation or to validate a media mix model's conclusion. A result arrives inside a confidence interval, so a single study is a data point, not a verdict.

The practical follow-through, for anyone using a sem agency or running the account in house, is a decision rule written down before the study starts. If incremental cost per action comes back above the account's target, the campaign or the budget line moves. If it comes back below, the budget moves toward the studied channel. Writing the rule first stops a result from being reinterpreted after the fact, which is the most common way a study gets wasted.

FAQ

What changed in Google's Conversion Lift setup documentation?

Google expanded the technical requirements and troubleshooting guidance for advertisers running incrementality studies. The changes cover enhanced conversions data requirements, offline imports that carry only a click identifier, unsupported setups such as Google Analytics web conversion imports, consent mode for EEA, UK and DMA campaigns, app conversion requirements and data quality rules. The study mechanics and the main thresholds stayed the same.

Can I run a Conversion Lift study with only GCLID offline imports?

No. Google's documentation states that click-ID-only imports cannot be used in a user-based study, because the control group is never served an ad to click on. Offline conversions need enhanced conversions for leads with hashed personal data, imported through Data Manager rather than the legacy API.

How long before a study should my conversion actions be firing?

At least 7 to 14 days of continuous conversion data is required before launch, with the selected actions firing with valid enhanced conversions and consent mode. That window lets Google verify the setup and calculate feasibility accurately.

Is a 7-day study enough?

Only for short purchase cycles. Google recommends running a study for more than 14 days, and states that studies with a long conversion lag running less than 14 days can lose up to 17% of absolute lift. Match the window to your average conversion lag, not to the length that fits the calendar.

Does a high ROAS figure prove that ads caused the conversions?

It does not. An attribution return on ad spend counts every conversion the platform credits, including those that existing demand or brand familiarity would have produced anyway. Only a holdout comparison separates the conversions the ads caused from the ones that would have happened regardless.

Sources

1. Barry Schwartz, "Google Ads Updates Conversion Lift Set Up Help Document", Search Engine Roundtable, 2026-11-13. https://www.seroundtable.com/google-ads-conversion-lift-set-up-doc-update-42530.html
2. "Set up Conversion Lift based on users", Google Ads Help, undated (retrieved 2026-11-13). https://support.google.com/google-ads/answer/12005564?hl=en