What your attribution is still worth under Consent Mode V2.
Someone searches, clicks your ad, declines the cookie banner, and fills in the form ten minutes later. You may still see that conversion in your Google Ads account. Not because a cookie was set, because it wasn’t, but because Google estimated the missing piece. That’s exactly where most Consent Mode V2 conversations go wrong: people treat every conversion in the dashboard as equally solid, when some of it is observed and some of it is modelled. Those aren’t the same thing, and the difference decides what your attribution is still worth.
What denied consent does to your measurement. With Consent Mode V2 in place and marketing consent declined, Google’s gtag still loads, but it sends a cookieless ping instead of a normal conversion hit. You can spot it in the network traffic by gcs=G100, the denied state. No _ga cookie, no identifier to tie that person back to anything later. So Google Ads does get a signal that something happened, but without the cookie pinning it to a specific click. If the visitor does accept, the ping carries a granted state and the cookie gets set after all.
Why you don’t cut that signal off. The temptation is to be stricter than required: block the tag entirely when consent is declined, just to be safe. In the tag manager that’s the needed setting, and it’s almost always wrong. You’d be severing the rail Google designed itself. The cookieless ping isn’t a leak, it’s the intended path. It hands Smart Bidding a signal without placing a cookie, so without touching the GDPR. Set the tag to needed and every visitor without marketing consent contributes exactly zero signal, and the bidder steers on an ever-smaller slice of reality. Stricter than the law asks, and worse for your campaigns.
Where the modelled part comes from. From those cookieless pings, combined with the pattern of observed conversions, Google estimates how many conversions the declined group likely produced. Those are the modelled conversions in your report. For Smart Bidding that’s useful: the bidder has enough volume to optimise on, instead of flying blind on only the visitors who happened to accept. For your own read on the account it’s a different thing. A modelled conversion is a statistical estimate, not a recorded event. You can’t drill it down to one person, one click, one search term.
The honest answer: directionally yes, precisely no. That’s what it comes down to. At the campaign and account level, your attribution under Consent Mode V2 is solid enough to steer on: trends, the balance between campaigns, whether the bid algorithm is pushing the right way. At the level of one conversion, one keyword, one euro, it isn’t, because part of what you’re counting is estimated. That’s no reason to switch the modelling off. It’s a reason to know which part of your number is observed and which part isn’t, and to match your language to it. You don’t present a modelled conversion as a hard lead.
That’s the same discipline we bring to a CPL that isn’t measuring what you think: knowing what a number is built on before you base a promise on it.
We look at this early in every audit. Are the tags on notSet or on needed, are the cookieless pings coming through, and do you know which part of your conversions is observed and which part is modelled? Checking that takes half an hour and it’s free.