TrackingDesk

analytics·attribution·advertising

GA4 budgets by ad platform now, and UTMs decide which

Published 2026-11-21 ·Google primary source

What it means

GA4's budget planner can now tell you to move money from Meta to Google Ads, but it only sees Meta, TikTok or LinkedIn spend that arrives with a paid utm_medium and a source on Google's list, and it learns from click-through conversions plus YouTube engaged views. Fix your mediums before you build a plan, and treat any shift it recommends away from view-led platforms as a hypothesis to test, not a finding.

Google Analytics’ budget planner now splits spend and return by ad platform: Meta Ads, TikTok Ads, LinkedIn Ads and so on, side by side with Google Ads. Google announced it on LinkedIn this week and added it to the cross-channel budgeting help page. Google’s traffic is labelled from its own inventory. Every other platform is labelled from your UTM tags only, by rules Google publishes. We ran 21 tags through those rules. 11 landed in Unlabeled or Other Ads, where the planner cannot treat them as a platform, and a 12th depends on letter case that Google does not document.

What Google changed

Cross-channel budgeting is the planner under Advertising → Budgeting in GA4. It has two plan types: a projection tracks how this quarter is pacing, and a scenario models where next quarter’s budget should go. Until now its rows were channel groups such as Paid Search or Paid Social. Now they can be ad platforms, and an alpha adds platform by campaign type.

The rest of the planner stays the same, and the conditions are strict:

  • A year of data, from Google and at least one other platform. Each conversion needs a year of history, and so does cost from at least two channels, one of them non-Google. Non-Google cost arrives by data import or a connector. Web conversions only; app conversions and key events are not supported.
  • The model is GA4’s own attribution. The DDA-based model, in beta for GA4 and GA4 360, feeds the property’s data-driven attribution into a Bayesian regression trained on 12 months. A Meridian-based model is in alpha for select 360 customers. The plans change no budgets: they are recommendations.
  • It learns from clicks, plus one kind of view. The help page says the models consider “all click-through conversions, YouTube engaged view conversions and all cost data.”

What we checked: 21 tags through Google’s rules

Google’s Source Platform page gives the rule for non-Google traffic in two steps. First the utm_medium must look paid: it contains cp, starts with paid, or is exactly ppc, retargeting, display or appremarketing. Otherwise the visit is Unlabeled. Then the utm_source is looked up in a table: facebook, fb, ig or instagram give Meta Ads, tiktok or tt give TikTok Ads, and so on. A source that is not in the table goes to Other Ads.

We wrote those two steps into a script today and ran 21 source and medium pairs through it. We chose the pairs to probe each rule, and added TikTok’s own auto-fill. They test the rules. They are not a survey of how often each tag is used.

utm_sourceutm_mediumSource Platform
facebookpaid_socialMeta Ads
fbpaidsocialMeta Ads
igcpcMeta Ads
facebooksocialUnlabeled
facebooksocial-paidUnlabeled
m.facebook.comreferralUnlabeled
msgpaidOther Ads
anpaidOther Ads
tiktokpaidTikTok Ads
TikTokPaidTikTok Ads or Unlabeled (case not documented)
linkedinpaid-socialLinkedIn Ads
linkedinsponsoredUnlabeled
bingcpcMicrosoft Advertising
redditcpmReddit Ads
pinterestadsUnlabeled
snapchatpaid_socialSnapchat Ads
twitterpaidX Ads
x.compaidOther Ads
openaicpcOther Ads
chatgptcpcOther Ads
newsletter-sponsorcpcOther Ads

What stands out:

  • The medium decides first, and social-paid fails it. paid_social and paid-social pass because they start with paid. social-paid does not start with paid and contains no cp, so a Meta campaign tagged that way is Unlabeled, as are sponsored and ads. GA4’s Paid Social channel uses the same medium regex, so these visits already sit outside Paid Social today.
  • Abbreviations below the brand level fall through. fb and ig are in Google’s table. msg for Messenger and an for Audience Network are not, so they go to Other Ads. A Meta campaign tagged by placement this way is split: its Facebook and Instagram clicks go to Meta Ads, the rest to Other Ads.
  • x.com is not X. The table matches x exactly, or anything containing twitter.
  • TikTok’s own auto-fill may not match Google’s rule. TikTok’s Auto-attach fills source TikTok and medium Paid, with capitals. Google says its channel definitions are not case sensitive. The Source Platform page does not say either way. We could not test it in a live property today, so treat that row as open.
  • AI ads get no platform. Google’s table names gemini, openai and perplexity, and maps all three to Other Ads. Spend in an AI assistant can be planned only as part of a bucket that also holds every unknown source. A tag written chatgpt lands in the same place. GA4’s AI Assistant channel is no help here: it counts only untagged clicks, and the planner counts only paid ones.

The read: the planner compares platforms on Google’s evidence

It will now name a loser. A channel-group plan said “Paid Social is underperforming.” A platform plan says “move money from Meta to Google Ads.” That recommendation is only as good as what GA4 sees of each platform, and for everything outside Google that comes down to the tags.

What is untagged or mis-tagged is invisible as a platform. Spend imported for Meta but sitting against Unlabeled sessions breaks the link between cost and return. The model either fails its quality checks or reads Meta as cost with little revenue. The TikTok connector shows how tight the join is: its source and medium must exactly match the tags on your URLs, and cannot be edited once set.

Views count on one side only. The model is trained on click-through conversions and YouTube engaged views. A Meta or TikTok ad that someone watched and later acted on through search adds nothing to Meta in this model: the click that GA4 sees is the search click. A Google video view does count. That is a rule in the model, not a bias in your data, and it points the same way every time. It is the old last-click problem with response curves on top.

What to do

  • Run your own mediums through the regex before you build a plan. Rename social-paid, sponsored, ads and the like to paid_social or cpc, and write sources the way Google’s table spells them. Old data keeps its old labels, and the model trains on 12 months.
  • Check Source Platform in an exploration for the last 12 months. A large Unlabeled or Other Ads share means the planner is short of what it needs to compare platforms.
  • Test before you move money on its advice. A recommendation to cut a view-led platform comes from a model that cannot see its views. Hold out a region or a week and measure the change, as a holdout test does.

Each platform also counts its own view-through conversions, with windows that differ by an order of magnitude. How much of Meta’s or LinkedIn’s reported total comes from views, and over how many days, is the number to set against what GA4’s planner leaves out.

Sources

  • Google Analytics help, 'About cross-channel budgeting': Google is launching new source grouping dimensions, source platform dimensions and (alpha-only) source platform x campaign type dimensions for budgeting plans, so performance can be analysed for both Google and other platforms. The tools are for planning only and won't affect budgets or spend in connected accounts.Googleverified 2026-11-21
  • Google Analytics posted on its LinkedIn page that cross-channel budgeting now supports source platform dimensions, so individual ad platforms can be compared side by side instead of broad channel groups. LinkedIn showed the post as one day old when read on 21 November 2026.LinkedInverified 2026-11-21
  • Google Analytics help, 'About cross-channel budgeting': the budgeting models consider all click-through conversions, YouTube engaged view conversions and all cost data; Google asks advertisers to include impressions in data imports when available, to prepare for future updates.Googleverified 2026-11-21
  • Google Analytics help, 'DDA-based budgeting': cross-channel budgeting needs at least one year of data for a conversion, and at least one year of campaign data from at least 2 channels, including Google and non-Google. Web conversions are supported, app conversions aren't yet, and key events aren't supported. Non-Google campaign data is uploaded through data import.Googleverified 2026-11-21
  • Google Analytics help, 'DDA-based budgeting': the model uses the property's historical data-driven attribution as the primary input to a Bayesian regression, is trained on 12 months of data, uses the Source platform dimension to report and forecast spend, revenue and conversions, and includes only paid channels.Googleverified 2026-11-21
  • Google Analytics help, 'About cross-channel budgeting': the DDA-based model is in beta for Google Analytics and Google Analytics 360 customers; the Meridian-based model is in alpha and only available to select Google Analytics 360 customers.Googleverified 2026-11-21
  • Google Analytics help, 'About the Source Platform dimension': Google traffic is mapped by inventory source (for example YouTube to Google Ads, Display & Video 360, Search Ads 360 or Campaign Manager 360). For all other sources, Google Analytics first checks whether the UTM medium is paid: it matches ^(.*cp.*|ppc|retargeting|paid.*)$ or is exactly 'display' or 'appremarketing'. If no paid medium is found, Source Platform is 'Unlabeled'.Googleverified 2026-11-21
  • Google Analytics help, 'About the Source Platform dimension': with a paid medium, the UTM source is mapped by a table, including %facebook%, meta, fbook, fb, messenger, %instagram%, ig and %whatsapp% to Meta Ads; %tiktok%, tt, %tik-tok% and %bytedance% to TikTok Ads; x and %twitter% to X Ads; %bing% and %microsoft% to Microsoft Advertising; %linkedin% to LinkedIn Ads; and %gemini%, %openai%, %perplexity%, (direct), (not set) and all others to Other Ads. For Unlabeled traffic, the page says to update UTM parameters to the recommended paid values to resolve it for future data.Googleverified 2026-11-21
  • Google Analytics help, '[GA4] Default channel group': Paid Social is manual traffic whose source matches a regex list of social sites and whose medium matches ^(.*cp.*|ppc|retargeting|paid.*)$. Channel definitions aren't case sensitive and can't be edited. The 'About the Source Platform dimension' page does not say whether its source and medium matching is case sensitive.Googleverified 2026-11-21
  • TikTok Ads Manager help, 'How to add URL parameters to your website URL': the URL Auto-attach feature auto-fills the campaign source (TikTok), medium ('Paid'), ID and name.TikTokverified 2026-11-21
  • Google Analytics help, 'Connect TikTok to Google Analytics': the connection imports TikTok campaign data into the Ads cost, Ads clicks and Ads impressions metrics; utm_source and utm_medium are required on destination URLs, and the Campaign source and Campaign medium entered in the connector must exactly match those values. A connector's ad account, source and medium can't be edited after creation.Googleverified 2026-11-21