The dos and don'ts of ai-generated ad creatives

The general public’s opinion on generative AI is incredibly polarized. Some see it as a miraculous tool that will solve most of humanity’s problems while others view it as a threat to livelihoods and genuine creativity. The problem this polarization poses for affiliate marketers is that while we can’t afford to ignore the audience that refuses to engage with anything remotely resembling the product of an AI, we also can’t afford to miss out on the massive productivity boost that AI provides.

This list of dos and don’ts is put together to help you mitigate the risks involved in running AI-generated ads while maximizing the value of the technology in your performance marketing campaigns. Let’s get into it.

DO fuel your A/B testing

The most enticing value proposition AI has for performance marketers is providing tons of fuel for A/B testing in record time. If you rented out an entire content mill and tasked every writer with writing ads for your campaign, they still wouldn’t be able to touch your average generative engine in terms of volume. That’s why most content mills edit AI-generated content nowadays.

With the current capabilities of major models, affiliate marketers can generate countless iterations of ads, adjusting them down to individual words, and immediately turn them into actual creatives, which are also endlessly iterable. For zero dollars. If you are not doing that, you are not exploiting one of the main advantages AI has to offer.

DON’T skimp on voice generation tools

The overuse of cheap AI-generated voiceovers in TikTok videos, Instagram Reels, and YouTube Shorts has spawned a generation that has a negative kneejerk reaction to even slightly robotic-sounding voices in videos. If you plan to produce video ads with AI voiceovers at scale, get a paid plan from a specialized platform like ElevenLabs or Jammable if you don’t want to tank your conversion rate.

DO color-process generated images

Gone are the days when an AI-generated image could be discerned by an abnormal amount of fingers or legs in the picture. Some structural anomalies still occasionally occur with complex shapes, but engines have gotten exponentially better at creating believable images from even a year ago.

The features that still immediately scream “AI image” are glossiness and color saturation. Most images produced by popular engines like DALL-E and Midjourney are more vibrant than a Disney poster. Some extra prompting or post-processing can make your ads look magnitudes more human.

DON’T let AI generate text-based objects within images

Whether it’s a speech bubble in a comic strip or a shop sign in a generated photo, current models have difficulties creating text-based objects within a picture. Occasionally you can get an eligible word, but anything beyond that devolves into a jumbled mess of letters. Putting an image like that in your ad creatives is a dead giveaway that it was AI-generated and it won’t reflect well on their performance.

DO include exact specifications in your prompts

“Prompt Engineer” is an actual job title for a reason. There is a world of difference between writing “Create an ad for a fitness app” into ChatGPT and running “Create a mobile interstitial ad in a minimalistic visual style with pastel colors for a fitness app aimed at middle-aged adults offering time-efficient workouts for busy people” through multiple engines, picking the winners, and iterating from there. Excruciating detail is your friend when dealing with generative AI.

DON’T let AI bias leak into your content

Remember the controversy when Google Gemini’s image generator couldn’t generate white people? That is a prime example of an AI bias. They are still present in modern models and most aren’t as readily apparent as the Gemini one because they may present themselves in text generation about niche topics. For example, ChatGPT can start pushing veganism when discussing plant-based diets, likely connecting topics that are often adjacent in its training data. Double-check that AI biases don’t find their way into your generated content.

DO fact-check AI content

Like any copywriter forced to churn out hundreds of iterations in a matter of minutes, generative AI tends to hallucinate. The more fact-based the content, the higher the likelihood of an AI to authoritatively state something not at all based in reality.

Luckily, there is no need to scour Google Scholar or Statista to confirm every point a generative engine makes. Whenever a statement looks suspicious, ask the model to source the claim. It will correct itself and even apologize to you.

DON’T expect cultural nuance from generated content

AI can’t meme. Ask any model to draw you a meme and the results will range from amusingly broken to deeply horrifying. It can generate serviceable puns, but anything with more layers of cultural context won’t make much sense to a human being.

Its lack of cultural nuance goes beyond humor. Writers immediately know to take a different approach to Western Easter as opposed to Orthodox Easter. AI models don’t, and a mistake like that immediately breaks a campaign.

DO leave simple front-end work to AI

Coding and designing simple one-page prelanders is precisely the type of work you should be leaving to AI. Conversion rates will still ultimately depend on the content, and most of your efforts should be directed towards optimizing the ad copy, but if you need a simple one-pager without extensive interactivity, AI will let you cut costs on putting it together. What is the last time you have seen a webpage and thought “that looks AI-generated”? Exactly.

DON’T let AI have the final draft

When we automate advertising to the extent that an AI model generates final drafts of ad creatives, tests them against one another in a live campaign, optimizes, and repeats the process, the products will likely be purchased by another AI.
Every time AI-generated copy was split-tested against ads created by copywriters, conversion rates significantly favored ads written by humans. As long as you are targeting people, a human touch remains essential for effective advertising.

Conclusion

The key to creating high-converting AI-generated ads is to work on them so much that they might as well not be AI-generated. It might sound like that defeats the whole purpose, but the costs saved on visual design, webpage coding, voiceover work, video montage, and copywriting massively outweigh the extra time spent formulating prompts. We aren’t at full automation yet, and we may never get there, but the tools available now make it much easier to scale as a solo affiliate as long as one is familiar with the intricacies and limitations of modern generative models.


Want more insights? Register here https://www.clickdealer.com/signup/?s1=68913