More views, fewer clicks: what a study on AI avatars teaches us about measuring ads

-

More views. Fewer clicks. Human presenter vs. AI avatar

An ad can succeed in views and still not justify a bigger budget. A study comparing a human presenter with an AI avatar is a chance to look at how we measure creative and what counts as success.

Imagine you get two ads to approve. In one, a person speaks to the camera. In the other, a character created with artificial intelligence. It looks good, speaks well, and the result is impressive enough that someone in the office says: “Wait, all of this is AI?”

The campaign goes live. You open the report and see that the AI version got more views. The obvious move is to pick it and shift the budget to it.

But before you declare a winner, keep reading the report. You may have chosen the ad that wins on one metric, while your actual goal is further down the road.

What they tested and what they found

A study published on 23 September 2026 compared a human presenter with an AI avatar in a Facebook campaign for a university in Colombia. The researchers analysed 384 ad runs, and also tested disclosing how the creative was made.

In the main model, the avatar was linked to 25% more video plays that reached a quarter of the video, 38% fewer clicks and 13% fewer people reached. No significant difference was found in the number of messages. This refers to all clicks, not specifically visits to a website; video plays also include autoplay and replays.

The test covered one campaign and two specific presenters, and the exposure was split by the algorithm. It doesn’t prove that people sell better or that AI doesn’t work. And the absence of a significant difference in messages isn’t proof that they’re equal. [1]

What should a business take from this?

For me, the interesting question is how we decide that an ad has succeeded.

“It looks amazing” is a reaction to the creative. “It brought the right customers at a price that works for the business” is a business conclusion. To get from the first to the second, you need to decide in advance what you’re testing and connect the data to the customer’s journey.

If the goal is to introduce a new brand, reach and views have a clear role. If the goal is to get enquiries for a specific service, we also want to know how many people got in touch, how many of them were a good fit and how many moved on to a conversation or a deal. You can appreciate creativity and, at the same time, require it to prove that it serves the goal.

This applies to every ad: a filmed video, an animation, a product image or an AI character. The technology behind the creative doesn’t choose the success metric for us.

Before producing another video, define what success looks like

My recommendation is to start with one sentence: “This ad should get the right person to ______”.

Get to know the business? Visit a product page? Request a quote? Book a place? Once you fill in the blank, it’s much easier to choose what should appear in the ad and what to check afterwards.

For example, a business that sells a service through a consultation call can define a qualified enquiry by need, service area and budget. That way it doesn’t settle for counting forms. It checks whether the forms represent people it can really help.

It’s also worth deciding when to review the result. In a service with a long sales process, reviewing too early can leave out deals that are still in progress.

How I would test AI creative in a business

I would start with a focused comparison around one offer. As far as possible, keep the message, the call to action and the destination similar, and change the way it’s presented. If you change the presenter, the price, the script and the landing page all at once, it’s hard to tell what drove the result.

In advance, I would define one main metric that fits the goal, alongside a small number of metrics that help explain what happened along the way. In a lead campaign, for example: cost per qualified enquiry as the main metric, with views, page visits and completed enquiries as supporting metrics. This is a professional working suggestion, not a method the study itself tested.

When there are views without progress, I would check whether the offer is clear and whether there’s a reason to continue. When there are visits without enquiries, I would check how well the promise in the ad matches the page. When the enquiries aren’t a good fit, I would go back to the messages and the expectations they create.

These are directions to investigate, not a diagnosis you can draw from one column in a report. And before any significant budget change, make sure enquiry tracking works and that you have enough information to make a decision.

And what about the time and money we saved on production?

Production cost belongs in the calculation too. If AI lets a business test an idea on a budget it couldn’t otherwise afford, that has value. But you also need to include the time for revisions, quality checks and brand fit.

I would look at the total cost, production, working time and media, against the result the business wanted to achieve. Saving on production is part of the picture; it doesn’t end the test.

The same goes for a brand character. You can use one to build recognition and a consistent language, and let the business owner appear where their experience and voice matter. Choose each one’s role according to the message and the audience, and then test it in practice.

The moral: choose the metric before you choose the winner

AI gives us more ways to create. The responsibility for deciding what to create, for whom, and what we want to happen next stays with us.

So the next time an ad impresses you, enjoy it. Then ask: what is it supposed to do for the business, and where can you see that it actually happened?

Because it’s easy to fall in love with creative. Split the budget after you’ve checked the results.

Want to understand which ads really move your business forward? At Barak Digital we connect creative, campaigns and measurement, so you can make decisions based on the result that matters to you.

Source

[1] Arango-Espinal, E.; Rodríguez Orejuela, A.; López-Herrera, P. A. (2026). Human Spokesperson Versus Artificial Intelligence Avatar and Authorship Disclosure in Digital Advertising: Effects on Performance Metrics. Journal of Theoretical and Applied Electronic Commerce Research, 21(10), 337.

https://doi.org/10.3390/jtaer21100337

This article is an independent professional adaptation and interpretation by Barak Digital, not a full translation. The recommendations for applying it in a business are Barak Digital’s.