[paxys]

Meh. I read the original report when it was released, and it was a whole lot of nothing.

Their claim was basically – "we found some Google video ad fraud!". Well guess what, Google knows and agrees that there is fraud. So do all the advertisers who pay them. The question is really about how much fraud there is. Google says that it actively keeps it under a certain %, and advertisers are generally okay with that number.

Now the report makes the usual claims – ads run on shady sites, ads viewed by bots, ads muted or obscured, ads unskippable and against policy etc. When you read through and try to find actual numbers for the severity of these issues, they just skip over it with weasel words. Just read their opening statement:

> However, this research report finds that for years, significant quantities of TrueView skippable in-stream ads, purchased by many different brands and media agencies, appear to have been served on hundreds of thousands of websites and apps in which the consumer experience did not meet Google’s stated quality standards. For example, many TrueView in-stream ads were served muted and auto-playing as out-stream video or as obscured video players on independent sites. Often, there was little to no organic video media content between ads, the video units simply played ads only.

"significant quantities", "many different brands", "appear to have been served", "many ads were served", "Often"...you can keep reading as long as you like, and will not find a single objective number or percentage. Unless someone can conclusively find that, no one is going to take them seriously.

[eddythompson80]

> "significant quantities", "many different brands", "appear to have been served", "many ads were served", "Often"...you can keep reading as long as you like, and will not find a single objective number or percentage

I’m a bit confused. The Adalytics report definitely defines numbers and percentages of the samples they looked at. Obviously only Google will have the authoritative final numbers, but they are also the ones most incentivized to keep them as opaque as possible. I admit I’m not very familiar with the ad business, and know that a fraud is rampant, but there is a difference between a rampant 5-10% fraud to 50–60%. The first few paragraph list a lot of percentages from that report

Original report: https://adalytics.io/blog/invalid-google-video-partner-truev...

[codedokode]

It is obvious to me that third-party sites are motivated to employ every trick to increase ad views. Google cannot show ads to bots, because it will be a fraud, but a third-party site can because they are in a different country and you won't hold them responsible anyway.

[macNchz]

I’ve never run video ads on Google, but I have run a lot of search ads, and if the 3rd party sites where they show the videos are anything like the “search partners” network, you’re correct, the sites are total trash.

Several years ago I looked through our referrers and immediately disabled search partners for all of our ads when I saw the actual sites.

Many were just clones of the Google search results UI but with ads indistinguishable from search results. Their play is to buy ads on Google themselves, while showing partner network ads on their own confusing “search results” page. They arbitrage traffic by buying ads on cheaper keywords and then getting visitors to click a more expensive ad they serve by being in the partner network.

From what I saw, I can only imagine the vast majority of human clicks through these pages are totally unintentional or confused.

[makeitdouble]

At the same time, Google is the only entity that could provide the numbers you want to see.

Who else could get actual numbers of scams they run vs legit ads ? There's no other entity that sees every single of the ads they serve, so we'd need a trustworthy insider leak to meet your standard.

[whatshisface]

There is a concept from polling called a "random sample" which would apply here. If you take 10,000 clicks and sort them into legitimate and illegitimate, as long as you were able to rule out any way that your sample would be significantly more or less fraudulent than the whole pool, you would understand (roughly) the ratio in the greater pool.

It's like taking a sample of a swimming pool, to check the chlorine levels.

[janalsncm]

I’m unfamiliar with how Google Video Partners works, but how could anyone go about randomly sampling the sites their ads are running on? According to the article Google won’t tell you where your ads are running. And just finding shady video websites is also not an unbiased method.

Also just to be clear, clicks are not likely to be randomly distributed between legitimate and spam sites. If advertisers are paying for plays, they need to be able to see that information as well.

If I was spending money on video ads now I would be somewhere between cancelling my ad spend and calling a lawyer.

[makeitdouble]

This would be a good measure for your own site, and you could get back to Google to say your results are garbage.

But you would have nothing to retort if they tell you you're an outlier, or your specific niche has bad actors that skew the results and they'll look into banning one or two of them (and come back to us again when they get replaced, we'll again clear a few token users)

[ineedasername]

>There is a concept from polling called a "random sample"... It's like taking a sample of a swimming pool, to check the chlorine levels.

And maybe the edge of the pool closest to the person taking the sample is the deep end where mostly adults swim, but if they walked a bit further and took the sample from the shallow side with the babies and toddlers they'd find out the water is actually 50% piss with more than trace amounts of feces mixed in as well.

[alvarezbjm-hn]

Strawman. Parent was talking about [(Cl^(-1))]

If you care about excrement content, that should be another study.

Bottom line: Representative sample https://en.m.wikipedia.org/wiki/Sampling_(statistics)

[autoexec]

I think the point was that you need to ensure a representative sample and that requires an idea of the scale of the fraud taking place to be able to measure its impacts. I have no idea if that's possible without insider data or not, but I suspect that was the point of the argument, and not which substance was being tested for in the imaginary body of water being used as an example.

[ineedasername]

Yes, I am-- very-- well aware of what a representative sample means, and made my comment specifically with that knowledge in mind when I responded.

[hsjqllzlfkf]

Of COURSE google says "yeah tottally below %".

I'm reminded of that quote "a man doesn't understand when his pay depends on him not understanding". Do you by any chance work in the advertising industry? (To be absolutely clear, if you work for google the answer to that question is yes)

Edit: not just a rhetorical question, I'm genuinely waiting for an answer to this.

[TAForObvReasons]

> It is difficult to get a man to understand something when his salary depends upon his not understanding it.

- Upton Sinclair

[nkozyra]

Yeah they really want to sell this as a bombshell but don't provide enough specifics or quantified data to do anything but say "well yeah ... and?"

I'm trying to find them motivations behind this thing.

[autoexec]

> I'm trying to find them motivations behind this thing.

They aren't exactly shy about their motivations. The site says their mission is:

> "holding the surveillance adtech industry accountable for abuses against advertisers and consumers, and spearheading the development of a transparent, efficient and privacy-focused digital advertising marketplace. "

Google is the #1 game in town (arguably the only game) so they need to convince advertisers that Google's ads are not serving their interests so that they'll be receptive to an alternative. It's entirely possible that they are correct and that Google is actually screwing over advertisers, I have no idea, but I suspect that it'd be more persuasive to demonstrate that their new ad platform actually serves advertisers better in real world scenarios than to just complain about whatever crap google is pulling.

[hedora]

All available credible evidence says they’re sending money to sanctioned entities and intentionally defrauding their customers. The fraud rate is somewhere between 50-90% of revenue, and adds up to billions.

The article cites specific examples of fraudulent spend, with screenshots, etc., on comically easy to detect domain names, including ones that other parts of Google block.

Also, they’ve progressively taken action to expand the percentage of advertisers they can defraud, and also to ensure the fraud is difficult to detect.

On top of all that, it is only possible due to monopolistic bundling of YouTube and the ad networks, in a scheme that could be a textbook example from “anti-trust litigation 101”.

[paxys]

Why not post this "credible evidence" then?

> The fraud rate is somewhere between 50-90% of revenue, and adds up to billions.

This is the only part that is relevant, and if you have a source for this then do share it.

[blincoln]

It's in the original report, which several other people have already linked to:

https://adalytics.io/blog/invalid-google-video-partner-truev...

There's a detailed methodology section. I haven't verified their results, but they report finding things like muted, autoplaying ads even on big-name sites like the New York Times.

A few key quotes:

"For one major infrastructure brand approximately 73% of their TrueView skippable in-stream budget was spent on Google Video Partner sites or apps whose ad delivery does not appear to be consistent with the TrueView or in-stream requirements."

"A major consumer goods brand had 75% of their TrueView budget delivered against Google Video Partner (GVP) sites or apps. 56% of their total TrueView budget was spent against websites which served TrueView ads in muted or non-in-stream video players [...] 19% of their TrueView skippable in-stream ad budget delivered against ineligible mobile apps which are not video streaming apps[...]"

"A Fortune 500 brand had approximately 57% of their total TrueView in-stream ad budget served on GVP sites and apps. 35% of their total TrueView budget was spent against websites which served TrueView ads in muted or non-in-stream video players [...] 8% of their TrueView skippable in-stream ad budget delivered against in-eligible mobile apps which are not video streaming apps."

I'm no marketing expert, but 43-73% of one's ad budget basically being thrown away seems like not great results.

[ineedasername]

Every thing you said sums up to: "Just trust Google".

[floomk]

> Google says that it actively keeps it under a certain %, and advertisers are generally okay with that number

How about actively keeping that at 0%?

[mwint]

Relevant: Bits About Money, The optimal amount of fraud is non-zero: https://www.bitsaboutmoney.com/archive/optimal-amount-of-fra...

[hedora]

The title doesn’t generalize to “the optimal illegal ad spend at sanctioned websites and countries is non-zero”.

They’re running EU ads on household-name Russian government propaganda sites in the middle of a war. Manual curation of a block list should’ve caught that!

[paxys]

Is there a block list of every Russian propaganda site in existence? Who maintains such a list? Or should Google have to manually review every website every day to check if it is part of that set?

[fireflash38]

They're some of the biggest ones, and easily discernable by the most cursory glances. Are you trying to say that their efforts will catch more hidden ones but not the obvious ones? If they can't even get the easy ones, why do you think they get any of them?

[Sai_]

Relevant that this is an opinion piece, not a study.

[lolinder]

It's not an opinion piece or a study, it's a pretty straightforward application of the law of diminishing returns. It doesn't matter if you're talking about fraud or uptime or test scores, the cost of a 100% success rate is almost never worth the extra effort it takes to get there.

[Sai_]

The law of diminishing returns is not natural law - it is just an observation of something that happens in practice but not always. E.g., people often choose to spend time and treasure treating an incurable cancer or send the Coast Guard to search for five people in a submersible.

Patio11's soundbite as often shared puts the cart before the horse.

He says that the "optimal amount of fraud is non-zero" as if allowing some fraud is necessary for society to function. But what he actually means is that since not all fraud is prosecuted and prevented, allowing some fraud is just the price we pay for actually getting things done.

Take the analogy of food processing. It is accepted that some small percentage of ground coffee will be insect or rodent remains and their feces. It's not like making ground coffee requires animal remains and feces.

I read Patio11's soundbite to mean that ground coffee requires animal remains and feces.

[lolinder]

You keep calling it a soundbite—did you actually read the article, or just the title? It doesn't sound like you actually disagree with him.

> The law of diminishing returns is not natural law - it is just an observation of something that happens in practice but not always. E.g., people often choose to spend time and treasure treating an incurable cancer or send the Coast Guard to search for five people in a submersible.

Most of the time we choose to value human life at approximately infinity (especially our own lives!). When one side of the equation is near-infinite, it makes sense to keep going long after the curve has begun to flatten, but that doesn't prove that returns don't diminish, just that the value of the goal is extremely high.

Diminishing returns isn't a natural law in its own right, but it's the direct result of natural laws. It has its parallels in half lives, in the inverse square law, and in basic probability theory. While the probability of something is high, it's easy to find and eliminate because random chance is on your side. When it becomes lower, random chance starts to work against you.

[sofal]

Do you think the tradeoffs discussed in that article are unproven?

[Sai_]

The tradeoffs are not unproven but the soundbite suggests that the tradeoffs are a necessary precondition. That's the crux of my issue with the soundbite.

[sofal]

So the tradeoffs are proven, but not necessary?

[x-complexity]

> How about actively keeping that at 0%?

Practically impossible. A sufficiently well-developed bot is indistinguishable from the least technically-savvy legitimate users that clicked on an ad. Unless you have a novel way of filtering out the 2 from each other, the costs for performing such filtrations compound linearly at best & exponentially at worst.

[floomk]

It should still be their goal, instead of some arbitrary percentage. Or I suppose they could offer everyone a blanket discount for the fraud percentage.