If a bot can fake it, we won't buy it.

By Kevin Veitia, founder of OuiGrowth

We don't run link-click campaigns, engagement campaigns or boosted posts. Each one asks the ad platform for an action a bot or a click farm can perform as easily as a person, and the platform gets paid either way. Here's how we learned that, and what we run instead.

It started with likes

Kevin started running Facebook ads in 2012. One of his early campaigns promoted a page for likes, and the likes poured in from profiles that made no sense. A typical one: an account in Nigeria with a Chinese profile photo, interests in Guatemala, friends in Estonia, and schools listed in both Pakistan and Australia. Almost every new like looked like that.

He wasn't the only one seeing it. In 2012 the BBC promoted a fake company page, Virtual Bagel, and collected thousands of the same kind of likes. In 2014, Derek Muller of the Veritasium YouTube channel showed in a video called Facebook Fraud that the likes he had paid Facebook to deliver, through its own ads and never a click farm, came largely from accounts that liked everything and engaged with nothing.

The test that settled it

Years later, on Yizzam, the question came back from the other direction. Yizzam sold all-over-print shirts, and its Meta campaigns had found a reliable buyer: art teachers. Its owner asked a reasonable question. If the audience was proven, why pay Meta to find purchases? Optimize for link clicks instead, get cheaper traffic, and the same people should buy at the same rate.

We cloned the campaign and ran the two versions side by side as an A/B test. Same audience, same ads. The only difference was the goal Meta was told to optimize for.

A 99.99% bounce rate isn't a weak landing page. Even terrible traffic leaves a few people who scroll. It's what happens when an algorithm is asked for clicks and finds the cheapest clicks available, whoever or whatever makes them. Nobody needed a second test. The Yizzam case study has the rest of that account's story.

When a client wants clicks anyway

PASCO Scientific, an education client we worked with for six years, promoted in-person teacher events with link-click campaigns. We advised against it; it was their call. Without a way to count who showed up, nobody could say whether those clicks put a single teacher in front of a booth. A click proves that someone, or something, tapped an ad. It doesn't prove a person arrived. The PASCO case study tells the whole story, including the parts that worked very well.

Why the platforms won't fix this for you

We're not saying Meta runs click farms. We don't believe it, and nobody has shown it. Our point is simpler, and it's about incentives. Meta gets paid when a click or an impression happens. Bots and click farms produce clicks and impressions. A company that exists to make money has no strong reason to work hard at turning that revenue away.

The public record doesn't make the platform's own numbers look any better:

• Video metrics. In 2016, Facebook acknowledged it had overstated how long people watched video ads; the Wall Street Journal reported the average viewing time was inflated by 60 to 80%. Advertisers sued, alleging the real inflation ran as high as 900% and that Facebook was slow to correct it. In June 2020, a federal judge gave final approval to a $40 million settlement in LLE One v. Facebook. Facebook maintained the suit was without merit.
• Potential reach. In DZ Reserve v. Meta Platforms, advertisers allege that Meta's potential reach estimate counted accounts, including duplicate and fake ones, rather than people, inflating it by as much as 400%, and that senior executives knew and took steps to cover it up. The Ninth Circuit let the case proceed as a class action in March 2024, and in January 2025 the Supreme Court declined to hear Meta's appeal. Advertisers estimate damages above $7 billion. Meta denies the claims and says advertisers were never charged based on potential reach.

None of this proves your clicks are fake. It shows that a platform's numbers are a starting point, not a verdict, and that the only results worth optimizing for are ones you can check yourself.

Our rule

If it can be spoofed, we don't optimize for it.

What about brand awareness?

It's a fair question, and we've run an awareness campaign ourselves. For Starbucks in 2018, we optimized for engagement and got it down to $0.01 per engagement. It made sense for one reason: Starbucks measured lift in its own studies, so engagement was a cost control, not the proof that the campaign worked. Without an independent check like that, we wouldn't run an engagement campaign today. The Starbucks case study explains how it was set up.

Check your own account

You don't need us to spot this. Ask whoever runs your ads, or check it yourself:

• What does each campaign optimize for? If the answer is clicks, engagement or reach, ask how they'll prove those were people.
• Compare clicks in the ad platform with sessions in your analytics. A large gap is a warning.
• Look at bounce rate and time on page for each campaign. Near 100% and near zero means you're paying for something other than people.
• Break results down by placement. Clicks piling up in placements you'd never choose by hand deserve a hard look.
• Judge every campaign on a number in your own systems: qualified leads, purchases, revenue.