Current language en no

Trickle challenged AI head-to-head

In late Autumn 2022, ChatGPT made an entrance into the world. People used it for writing news articles, code debugging and simply for the joy of exploiting its imperfection. Being a digital marketing agency, Trickle wanted to try its competence within paid social.

The objective was very simple: push as many people as possible to a Looker Studio Dashboard at the lowest possible cost. The target audience was very, very broad, and consisted of people who have shown interest in digital marketing.

We instructed the AI to create three different ad captions, and three different images, with that purpose. We gave ourselves the same mission, and that gave us in total:

  • 3 AI generated captions (TEXT_AI_1, TEXT_AI_2, TEXT_AI_3)
  • 3 human generated captions (TEXT_HUMAN_1, TEXT_HUMAN_2, TEXT_HUMAN_3)
  • 3 AI generated images (IMAGE_AI_1, IMAGE_AI_2, IMAGE_AI_3)
  • 3 human generated images (IMAGE_HUMAN_1, IMAGE_HUMAN_2, IMAGE_HUMAN_3)

We decided to mix and match those combinations in all possible ways. That gave us:

  • 9 different ads with only AI material
  • 9 different ads with only human material
  • 18 different ads with a combination of AI material and human material
  • A total of 36 unique ads.

We threw some dineros on those ads, pushed “Action!” and sat down, and started watching.  The actual competition, which was displayed in the dashboard, compared the cost per click (CPC)  for the “AI only ads” and the CPC for the “human only ads”. 

Now, normally, we never just sit back and watch how a campaign performs. We take great pride in our optimisation work, so it did feel strange to just leave everything untouched.  After a few weeks, and US$650 in ad spend, the result was finally in. You can find the total list here.

Reflection and what we learnt

Of the top 10 performing ads, 6 were all AI made, 4 were based on a combination. 0 (zero) of them were entirely human made. The first entirely human made ad came at position 19. The data gets even more interesting when we break it down a little. Based on who created the caption, the difference is almost insignificant:

  • Text by AI, CPC $0.14
  • Text by human, CPC $0.15

But if we look at the imagery, the difference is more tangible:

  • Image by AI, CPC $0.12
  • Image by human, CPC $0.17

Combined, those numbers give an impressive victory to the AI. 

So – what are the reflections of this campaign? First, it’s hard to not be impressed by the AI. It generates ad captions at the perfect text length. Even if it, at a first glance, generates a bit low quality images, they obviously deliver as intended.  The AI is, already today, good enough for assisting you in creating some kinds of content. (Just be aware of the copyright issues.) 

As stated previously, this campaign goes against the way Trickle works in so many ways. Primarily: the sit back and relax methodology, yikes. Secondly: the target audience (over 200 million people), talk about “Spraying, don’t Trickle”. What we really are waiting for is when AI can work independently as a paid social specialist. Based on just a few basic instructions, it creates content, landing pages, creates campaigns,  installs tracking code and generates target audiences. It optimizes the campaign from different perspectives hundreds of times every second. It reacts to patterns impossible to discover for the naked eye, and will eventually learn so much that there will be need for optimisation.  

Does this mean AI can work independently as a specialist? Create content, landing pages, campaigns,  installs tracking code and generate target audiences based on just a few basic instructions? Not yet. But maybe… within 10 years? 2 years? Time will tell.

Rest assured, we will monitor it closely.

Examples of AI generated content

Examples of human generated content