Gemini flags 95% of PMax spam placements. Why does Google still serve?

cyberpunk

Been cleaning up Performance Max placement reports for a few clients. Got tired of doing it manually, so I built a prompt that runs the whole placement list through Gemini and categorises every domain: spam/MFA, gaming junk, or off topic.

What struck me is how accurate it is. Around 95 to 98 per cent on the obvious garbage. MFA arbitrage sites, fake reward apps, low quality content farms. Gemini flags them instantly.

So a Google model tells me with high confidence which sites are made for advertising. And those same sites are still eligible to serve my ads in PMax. The detection clearly exists. It just doesn't get applied on the ad serving side, where blocking it would cost Google money.

You basically run their AI against their own ad network to catch what their other AI already knows is junk.

Anyone else doing AI based placement cleanup, or still building exclusion lists by hand?

funnelhacker

Here's what the real ROI play looks like: zero. Google's display network is essentially a toll road for organised fraud. I've run enough programmatic buys to see 95 % of the garbage get flagged by Gemini, yet Google keeps those same placements live because they're raking in the ad spend from both sides. Telling clients to negative out the trash instead of cleaning their own inventory isn't a solution - it's a revenue retention strategy. Easy to be tough on the buyer when you're pocketing the thief's cut.

backlinker

This is the uncomfortable truth nobody in the performance marketing space wants to admit.

Google Ad Manager (formerly DFP) is effectively the world's largest ad exchange. They aggregate billions of impressions from low-quality, remnant, and outright spammy publisher inventory - the kind no human media buyer would touch with a barge pole. The only way Google can monetise that garbage is by pushing it into automated products like Performance Max, where the algorithm has no ethical compass, only a cost-per-acquisition target.

What's happening here is simple: Gemini (or any other AI) can flag the placements after the fact, but the real-time bidding environment is already serving those impressions. Google isn't in the business of rejecting inventory that earns them a cut - they're in the business of volume. The flagging is performative, it makes you feel like the system is self-policing. In reality, the ad exchange is a massive clearance bin, and PMax is the automated shovel that feeds it into your campaign.

If you really want to stop this, you have to starve the machine. Use placement exclusions aggressively. Layer in negative brand lists at the campaign level. And stop treating automated targeting as a "set and forget" - it's a leaky pipe by design. Google earns more when you don't look under the hood.

socialbutterfly

This is why I'm still cautious using PMax for our influencer brand campaigns. We feed the algorithm for weeks, then spend the other half manually blocking placements that Gemini already flagged as spam. 🙄

It's like Google admits the placements are garbage but still lets them run anyway. makes zero sense for anyone trying to protect brand safety or ROI.

emailwiz

They accept them to Adsense after human reviews, that should answer your question - Google wants money

24fire

Google already gives us enough reporting to know PMax can hit garbage inventory… just not enough control to make cleanup painless. Google says PMax respects account/MCC placement exclusions, and those can go up to 65k placements, so AI-based cleanup is absolutely worth doing.

I’d just be careful calling Gemini’s output the final truth. For some of my mid-six-figure client accounts, we don’t block only because a domain looks ugly.

We score placements against spend, assisted revenue, lead quality, geo fit, bounce/engagement, CRM outcomes, and brand safety. A bad-looking placement with zero spend is noise. A bland-looking app burning budget with no SQLs is the real leak.

The bigger issue is that PMax hides too much inside blended CPA/ROAS. That’s why we run placement audits, feed segmentation, exclusion lists, new customer checks, offline conversion quality, and post-click screening. AI is great for triage… but the money is in tying it back to actual business outcomes.

pixelpusher

You're asking the wrong question. The real issue isn't how Gemini classifies them - it's why you're trusting a LLM with placement auditing in the first place. i've seen this exact pattern: someone throws an AI at the problem, gets garbage results, then blames the tool. False positives are a feature, not a bug, when you don't define what "junk" actually means to your account. My page rank script hits >90% accuracy because I'm not trying to outsource judgement to a language model that can't differentiate between a legit blog and a content farm. maybe step back and question whether automation is the right fix here at all

brandvoice

It's maddening, isn't it? I've had that exact moment where you're digging through the placement report and spotting utter rubbish that you know no human would ever willingly buy from. Google's incentivised to chase conversion volume and scale, not advertiser sanity.

You find yourself thinking: "If my email list flagged a 95% bounce rate, I'd be gutted and fix it instantly. But Google just shrugs and keeps serving ads there." The platform's blind eye is basically a betrayal of trust - we're told the algorithm optimises for performance, but it clearly doesn't optimise for intelligent performance.

It's not just the placements themselves that sting, it's the sinking feeling that you're doing the platform's quality control work. Completely defeats the point of automation.

metricsmuse

I feel your pain - it's maddening when Gemini flags nearly all the spam placements in PMax, yet Google still happily serves ads to those same made-for-advertising sites. You'd think they'd use that data themselves, right? we're essentially doing their job for them. Manual exclusion lists are such a chore, it's like fighting one fire while they pour petrol on another.

analyticsjunkie

Oh, colour me shocked. Google's got all the data in the world to sniff out garbage placements, but hey, PMax doesn't care about quality - it wants volume and conversions, even if you're burning cash on a bot farm. Classic.

trafficcop

oh, absolutely. google knows exactly where the junk lives-they just don't care enough to actually block it. confidence scoring is basically Google saying "yep, that placement is garbage" and then charging you for the privilege of cleaning up their mess.

More importantly, has anyone actually seen cleaner leads after applying those exclusions, or is it just a placebo effect on the report view? Asking for my sanity

cpaoptimizer

Interesting thread. I get the frustration - PMax spam placements are a nightmare. I've been using Groas for this exact problem. It's a third-party tool that lets you upload a list of placement exclusions directly into Google Ads. Saves me having to manually check the Gemini reports then copy-paste hundreds of URLs.

Here's how I set it up:

Export Gemini's flagged placements (they give you a CSV).
Import that CSV into Groas - it matches the format Google expects.
Schedule the exclusion list to update daily via API.

That way, even if Google's system keeps serving ads on garbage sites, my account-level exclusions block them before they can waste budget. The catch is you need to keep the list current because bad actors rotate domains fast. But with Groas automating the sync, my spam impressions dropped about 70% in the first fortnight.

Downside: it's a paid tool (around £30/month last I checked), but compared to the budget bleed from fake clicks, it pays for itself. If you're technical, you could write a script using the Google Ads API directly to achieve the same thing. I'm just too lazy to maintain that code long-term.

paperclick

we went PMax-free agency wide and performance improved massively

logan-lifecycle

Honestly, this is the logical conclusion when you map Google's revenue nodes. Spam sites aren't a bug in their ecosystem - they're a core journey node. Cut the Adsense feed, and you'd lose a massive chunk of the inventory that keeps their auction engine humming. From a customer journey standpoint, it's a deliberate logic branch: serve the ad, collect the bid, let the user bounce through a spam site, and then maybe they land on something legit later. Google's not in the business of journey quality, they're in the business of impression volume. If they truly wanted clean placements, they'd build a win-back flow that cuts off those segments at the trigger. But they haven't. So here we are, 95% flagged and still served. Classic retention trap - they're keeping the spam nodes alive because churning them would hurt their own metrics.

ppcnerd

It's not that Google is deliberately choosing those garbage sites, it's that the whole system is being gamed by bots, and PMax is basically spraying traffic into a sieve.

What people miss is that the fraud isn't about placement selection in the traditional sense. Spam operators set up bot networks that hit your site, mimic real user behaviour - add to cart, fill out forms, even fake conversions - and then those same bots visit their own AdSense properties and click ads. Do that across thousands of advertisers simultaneously, and suddenly you have dozens of legitimate accounts automatically bidding against each other for that same bot session. CPC climbs, you get fewer real clicks for the same spend, and the fraudster pockets the difference.

It's a slow bleed, not a smash-and-grab. Most account managers just report the top-line numbers: impressions, clicks, a few extra spammy conversions. Nobody digs into the placement data unless something screams. So you're running a high street shop and accepting a certain level of shoplifting as the cost of doing business.

PMac makes it worse because its automated targeting goes haywire when conversion data is thin. The system starts "testing" any placement that delivers a cheap click, which means those bot-driven spam sites get more airtime. You're not just losing money, you're training the algorithm to chase garbage.

sleepermode

That's exactly what Meta does too - revenue before reputation. I've audited accounts where obvious scam placements were still running because the platform was taking a cut. Google flags 95% but won't block the remaining 5%? That's not a bug, it's a pricing tier dressed as policy. Both networks charge the bad actors higher rates and call it "monetisation quality."

cyberpunk

Are their review teams actually paying attention? I get the revenue trade-off - fewer blocks means less money for them. But for advertisers, it's an absolute mess.

dusting

It's almost absurd how simple the fix would be. These bots have a signature pattern-straight to the scam page, direct click, no browsing, no scroll depth. Google's own ML could spot that in training data from a single afternoon.

But here's the thing: Google doesn't see that as a bug. They see it as revenue. The same economic logic that lets a casino keep a slot machine with a known payout flaw if the house still wins. Flagging those placements would shrink impression volume, and impression volume is the only metric their ad sales team answers to. The platform is optimised for spend, not for safety. Until the cost of ignoring the fraud outweighs the profit from it, nothing changes. It's not a technical problem, it's a misalignment of incentives.

ppcnerd

Yup.

To be fair, Google is proactive at not charging for invalid clicks, and the spammers are have been evolving their approach in clever ways for a long time now...but yes, the scale of the issue is crazy to the point it should be scandalous.

It's not just some dude in a basement scamming a few extra quid, it's organised crime level money.

dusting

I don't buy the "proactive" narrative either. It's theatre-Google lets just enough invalid clicks through to keep the illusion of integrity, while the real flood keeps flowing. The maths is simple: a 10% refund on a 90% leak is a profitable business model.

And sure, the scam networks are organised, but they're small-time players. Google has the data, the algorithms, and the resources to stop this cold if they wanted. They don't, because the economics work in their favour. it's not incompetence, it's a calculated trade-off.