In the AI Era, Whose Shelf Is Your Ad Money Landing On?

A while back, I had dinner with a friend who runs brand media buying. He had just come off a quarter-long mega promotion, and as he talked about one thing in particular, he got more and more frustrated.

A good chunk of the budget he put out through programmatic had landed on websites he had never heard of. Those sites seemed to have everything — dense, crammed content, ad slots stacked shoulder to shoulder. Click in and take a closer look: the articles were unreadable, with that mass-produced, machine-spewed smell all over them.

Later he had a verification company run the numbers. The verdict: a significant share of the traffic came from low-quality sites. The industry calls these sites MFA (Made-for-Advertising) — born for ads, not for readers.

Dig one layer deeper and there was something even more intimidating underneath: content sites mass-produced with generative AI, pages stuffed with ad slots, set out specifically to hook programmatic budgets. The industry gave these a name too — slop sites. Slop, as in pig swill.

Yikes.

That episode stayed on my mind for a long time, and today I want to unpack the whole topic with you: how brand safety and brand suitability should actually be done in the AI era.

First, Let's Get Two Terms Straight

What is brand safety?

Simply this: don't let your ads appear next to terrible content. Violence, hate speech, misinformation — any brand that runs into one of these has an incident on its hands.

And what is brand suitability?

It goes one step further than safety. Safety means steering clear of the bad; suitability means picking the right.

Here's an example. An ad sits next to content with no violence and no misinformation — completely "safe." But the article drips with passive-aggressive snark, cracks jokes about disasters, and grates against the brand's tone. Safe? Safe. Suitable? Not.

That is what suitability is for.

Think about picking a seatmate for your kid. The one with poor grades who loves to fight is a clear no — that's safety. Then, among the quiet, well-mannered kids, you pick the one whose temperament clicks with your own child — that's suitability.

So where do most brands get stuck? In treating these two things as one, and trying to run the whole show with a single keyword blacklist.

The blacklist sees the word "shot" and flags danger — could be violent content. But in sports coverage, a shot is a shot on goal. Does soccer reporting talk about violence? No. A one-size-fits-all list doesn't care, though, and the ad gets blocked anyway.

That's over-blocking. The money is spent, the volume is gone, and huge amounts of quality inventory get blocked at the door by your own side.

The New Business AI Built

The safety problem wasn't even solved before generative AI cranked the difficulty up another notch.

Slop sites produce no value — only "things that look like content." Their goal is singular: turn your ad money into their revenue.

There are stealthier plays too: cloning legitimate media sites, faking user identities, mass-faking clicks. Back in 2023, Juniper Research predicted that by 2025, global losses to ad fraud would reach $100 billion a year.

$100 billion. That is the scale of the scammers' business.

And in 2024, IAS (Integral Ad Science) offered an even bigger call: in the near future, as much as ninety percent of online content could be AI-generated.

Ninety percent.

What does that mean? It means that from now on, most of the "content" you see online may have been written by machines. And some of it exists purely to fish for ad budgets.

Fight back with old lists and old rules? You can't read it all, and you can't keep up.

Massive junk generated by AI can only be intercepted by AI.

40 Years of Video a Day — Who Could Watch It All?

So let's see how the professional verification companies take this on.

Start with IAS. The first time I saw this number, I froze for a good while: IAS processes more than 280 billion ad interactions every day, and each day it analyzes video content equivalent to 40 years of footage.

Forty years. An analyst working nonstop from graduation to retirement couldn't watch one day's worth.

So it goes to the machines: machine learning, natural language processing, and computer vision, the full trio. A pre-bid pass before delivery, a post-bid watch after, millions of URLs classified one by one, video analyzed frame by frame. There is even a dedicated Threat Lab keeping daily watch on what new kinds of fraud look like.

Then there's DoubleVerify (DV). Its Universal Content Intelligence engine blends visual recognition, OCR, and natural language processing, keeping the web, mobile, connected TV, and social media all within view.

Its anti-fraud model has been refined over 15 years. Fed fifteen years of fraud samples, the model can put a risk score on fake reviews and forged media identities on sight. It also has a product called Scibids AI that turns campaign data into bidding strategy, auto-optimizing thousands of campaigns at once.

Impressive. No human hands can keep up with that speed.

The interesting thing is that these models learn fraud the way people do: see enough of it and you start to recognize it. Feed in the patterns of known schemes, the model learns what a scammer looks like, then takes that experience and sweeps traffic across the whole web. Every new scheme discovered is another bit of experience gained.

And the loop can't stop. Scammers use AI to speed up, so defenders have to speed up with AI too.

This is an arms race of AI versus AI.

Looking ahead, the direction is already clear: detection models will get finer, catching deepfakes through metadata, language patterns, and source credibility; the yardstick for ads will shift from "was it visible" to "did anyone actually watch," with attention metrics like Attentive CPM; and repetitive campaign optimization will be handed to AI more and more.

Your Money, Someone Else's Shelf

Enough about the arms makers — now the ones spending the money. What should brands and agencies do?

Let's start with a few numbers — all of them scary.

In DV's 2025 Asia-Pacific consumer survey, nearly half of respondents say that seeing a brand's ad next to offensive content makes them less willing to buy.

Kantar's 2024 survey: 65% of Asia-Pacific consumers somewhat distrust brands tied to low-quality, misleading content.

Now the upside. An IAB Southeast Asia report from 2023 noted that in Asia-Pacific markets, AI-powered contextual targeting can lift click-through rates by as much as 30%.

Think about it: where your ad shows up is no longer just a media question, it's a brand-equity question. Consumers don't distinguish between "the brand" and "the content next to the brand." In their eyes, you and whatever sits next to you are in it together.

So what should you actually do? Four things, all worth starting right away.

First, find a solid verification partner instead of cobbling together your own tools. Companies like IAS and DV have all the tech stacked up, and they can sweep content across languages and markets. In Asia-Pacific especially, where languages and cultures shatter into dozens of pieces and every market draws the line between "appropriate" and "offensive" in a different place, this is exactly where AI sentiment analysis can help.

Second, define your own brand suitability guidelines instead of using the industry defaults. What is your brand's tone? What is your risk tolerance? Which content gets flagged and which gets a pass? You have to decide for yourself. Otherwise you either lose volume to over-blocking or lose money to junk traffic.

Third, monitor in real time instead of waiting for the month-end review. The delivery environment changes minute by minute, and your tools need to alert and act the moment something looks wrong.

Fourth, take consumers' wariness of AI seriously. In that Kantar survey, 60% of Asia-Pacific consumers said they were worried about misleading AI-generated content. So do the AI disclosures you should do, and don't skimp on high-quality creative. And rising with it all: everyone's standard for what counts as "real."

The Old Guard Proves Itself

Last, the ones selling the shelf space: the publishers.

Advertisers fear junk content. Flip it around, and quality publishers are racing to prove "ours is clean." South China Morning Post (SCMP) built a product called SCMP Signal that uses AI to vouch for its own ad environments.

How does it work? The same combination as before: natural language processing plus a sentiment lexicon tool like VADER, reading an article's emotion, keywords, and overall tone, then judging what ads the placement is fit to carry.

The "shot" example from earlier is the classic case. A shot on goal in a sports article and a gunshot in a violent news story are the same word; read the context and they split apart at once. What a blacklist can't tell apart, AI can. Quality placements stop getting mistakenly killed, advertisers keep their volume, and publishers keep their revenue.

Does it work? Two numbers.

In 2020, a health-and-fitness campaign that used sentiment targeting improved its results by 35%.

COVID made the contrast even clearer. Back then, many basic blacklists ruled every page with COVID keywords unsafe, and 67% of related pages got blocked. Contextual tools, meanwhile, could read whether a piece of news was explainer or panic, and identified 58.5% of similar pages as actually safe.

Same word. The blacklist sees fear; the AI reads information.

For publishers, this is no longer a selling point so much as a lifeline. Third-party cookies are fading, privacy regulation is tightening, the old road of tracking users is closing, and contextual targeting feels more and more like the natural path. And with partnership campaigns delivering lifts of up to 40%, numbers like that are hard currency with premium advertisers.

Back to My Friend

At the end of that dinner, my friend said his standard is simple now: before a single dollar of budget goes out, ask one question first — will it land next to content that is real, clean, and a match for the brand's tone?

Technology keeps getting stronger, and so do the scams. Tools will iterate, lists will update, but one thing will not change:

Trust is the only universal currency in digital advertising. And AI verification is the craft of making that currency counterfeit-proof.

May every dollar of your ad budget land in the real world.