How AI can Boost Campaign Performance

An Oxford study just confirmed what most of us have been saying all along: AI-generated ads can outperform human-made ones, but only when they don't obviously look AI-generated. The secret? Human refinement. The best marketing campaigns aren't purely AI-driven or entirely human-made. They're like pizza. Dough alone is just bread, toppings alone is chaos. The magic happens when everything works together. Want to collaborate with AI effectively? 1) Use AI for rapid ideation, humans for emotional depth Take your worst-performing ad copy and feed it into ChatGPT or Claude with this prompt: "Rewrite this to evoke [specific emotion: frustration, curiosity, nostalgia]. Use conversational language. Surprise me." It'll give you variations you'd never think of. Then your human brain picks the best concept and refines it until you think: "We'd never have written this ourselves." 2) Let AI spot patterns, humans craft the story AI's really good at combing through customer feedback, support tickets, and social mentions for trends. But humans make those insights into stories that actually matter. Say AI finds that most support tickets mention setup frustration. Humans craft that into: "Setup shouldn't feel like assembling IKEA furniture blindfolded." 3) AI scales the testing, you choose the winners Generate multiple variations with AI, but you decide which ones are worth spending money on. AI can create 50 headlines in minutes, your judgment tells you which 3 are worth testing. 4) You set the rules, AI fills the gaps Define your brand voice, values, and no-go zones. Then let AI work within those boundaries to fill content calendars, generate product descriptions, or create email variations. Platforms are making this easier: - Microsoft’s Ads Studio has AI-powered creative tools built into campaign workflows - Google Cloud rolled out AI marketing tools for personalized experiences - Or start simple with ChatGPT/Claude and the prompt above Stop thinking AI vs. humans. Start thinking AI + humans. Your move: This week, pick your worst-performing content. Run it through AI with a specific emotional prompt. Refine the best result with your gut instinct. That's how you make sure your marketing isn't just dough or just toppings, but complete, irresistible pizza. P.S. I'm team pineapple on pizza 🍕 + 🍍 = 🤤 (Sorry to my Italian friends! At least there's no ketchup involved... 😂) #hicm #AI #AIinAdvertising

Everyone’s trying to use AI to write better ads. But the real unlock isn’t the tools — it’s building a system that turns raw customer data into a machine for scalable ad concepts. At Obvi, we’ve been actively building a system designed to: → Surface deep customer insights → Map them to scalable ad frameworks → Generate ad angles that humans alone couldn’t uncover Not just faster. Fundamentally smarter. Here’s the system so far: Step 1: Fuel the system with raw customer language We collect authentic, emotional data from: - Post-purchase surveys - Product reviews - Customer service tickets - Ad comments - Email replies The more you can educate the machine with real customer feedback, the better. It gives the system deep context to work from. Step 2: Extract personas from the data We use this AI prompt (with uploaded files): "Act as a psychologist and direct response marketer. Identify 5-6 buyer personas based on motivations, pain points, and desires. Use real language from the data." This becomes the emotional blueprint of our system. Step 3: Map product benefits to personas Next, we prompt: *"Match our product features to each persona’s needs. For each: - 3 most compelling benefits - Emotional drivers - Specific pain points solved - Real language patterns to mirror. Now the system can speak to customers in the frame of their wants and needs. Step 4: Analyze top-performing ads We upload screenshots of our best ads and prompt: *"Dissect this ad: - Psychological techniques - Messaging frameworks - Tone, structure, voice - What made it work at a deep level."* This helps the system understand not just what customers want — but how we’ve already succeeded in reaching them. Step 5: Generate new ad creative components For each persona, we prompt AI to create: - 3 headlines (5-7 words) - 3 subheads (10-12 words) - 3 body copy snippets (2-3 sentences) - 3 trust signals - 3 CTAs (matched to tone: authoritative, empathetic, or urgent) Each output is grounded in real customer psychology, not guesswork. Step 6: Humanize and design This is where human judgment takes over. Anything AI generates is conceptual — the final polish and emotional resonance comes from creative teams who know how to craft stories that move people. Early Results: - More diverse, hyper-targeted ad angles at scale - Messaging that resonates with specific personas and pain points - Creative cycles moving with superhuman intelligence Like I said, my goal is not just faster, but smarter ad creation. 🔑This isn’t "using AI to write ads." It’s about building a system that compounds customer learnings in ways people can't — and translating those insights into scalable creative. This system is still evolving.  If you’re working on something similar, I’d love to swap notes. One thing feels clear: The brands that master customer insight systems will dominate the next phase of performance marketing.

AI in marketing has come a LONG way. Here are 8 ways we use AI in marketing at Avenue Z: ✅ #1 - Market Research & Persona Development ChatGPT is phenomenal at competitor research, review-scraping to uncover reasons why people by specific products, and can develop specific personas/funnel ideas to fuel customer acquisition. ✅ #2 - Creative Our AI-powered research fuels our creative briefs & angles. Best use case for AI-driven creative is background swapping of product images, AI voiceovers & video cutdowns. Still aren’t there yet on HQ/usable GenAI creative from a simple prompt. ✅ #3 - Search Engine STILL the #1 way content is found online. AI Overviews appear above organic search results. We craft on-site content designed to rank not just on org search, but in AI search results too. Showcased an example of this a few months ago, where we consistently appear in AIO for a service of ours. ✅ #4 - UGC Playing around w/ Icon, a platform enabling AI-driven influencer content. If an influencer shoots a 2-min video of your product, you can use their likeness to create endless iterations of the content. Swap the hook, talking points, language or B-roll at scale. Have a winning ad? AI deconstructs the vid & enables its recreation using other influencers. Easy way to test diff demographics or languages w/o having to source specific influencers. Game-changing. ✅ #5 - Ad Copy AI is much better at written content than visual content. Platforms like Meta provide copy iterations that you can test seamlessly. ChatGPT is solid for rewriting content to give a diff tone, look, or feel. Build custom GPTs to train the LLM on brand voice or other examples of copy you like/have performed well. ✅ #6 - Landing Pages We built a custom GPT that generates specific landing page frameworks for any brand/offer/product based on our LP best practices. Eventually we'll see AI create personalized landing pages based on a user’s attributes. Ex: returning customer sees diff website experience than new website visitor. Highlights the importance of collecting ample 1P data in today’s world. ✅ #7 - Measurement For larger brands w/ many different traffic sources & sales channels, attribution can be a futile effort. Multi-touch attribution (MTA) can fall short. This is where marketing mix modeling (MMM) comes in. Prescient AI modernized MMM by powering it with AI/ML to give objective recommendations on budget/allocations based on audience saturation & halo effects your marketing has on other ad platforms & sales channels. ✅ #8 - Targeting/Algorithm Training Ad platform algorithms already use AI in targeting which is why they advocate so heavily for broad targeting. BUT- what’s really cool is the concept of using predictive analytics to make the algorithm/AI smarter. Angler AI doing cool things in this area. Especially effective in niche product categories or brands w/ high AOV. Reminder: you're only as effective as your utilization of the tools around you!

AI large language models are phenomenally good at understanding human psychology, sociology, and behavior. I think most marketers dramatically underestimate how good these tools are at this when they're prompted well. I've seen a lot of company-generated customer persona and ICP documents, and I've seen a lot of personas and ICPs that were created by ChatGPT, Claude, and Gemini. And 9 times out of 10, the well-prompted AI version wipes the floor with the human-created version. (I'm sorry -- I know saying that is going to ruffle a lot of feathers -- but this is just something that, in general, generative AI is much better at creating than humans are, especially if your company not spending insane amounts of money on market research). I just read a study from NYU and Stanford that was published on August 8th that demonstrated how insanely good these models are at predicting outcomes of social science experiments and human behavior. This got me thinking... what if we applied a very similar methodology to the one that yielded high accuracy in predicting these sociology experiment results to our marketing A/B testing? Imagine being able to actually simulate A/B tests before running them, using GPT-4o or Claude Sonnet 3.5, and being able to forecast which ad copy or landing page copy would be more likely to perform better with that campaign's target audience. This method could optimize our testing processes, allowing us to focus on running our tests with the most promising variations and potentially speed up results. And it would be possible at a very low cost. By tapping into the large language model APIs with a well built automation, I think you could probably predict test A/B results before running them—not to an exact percentile, but with a decent degree of confidence unless the test would be a razor thin margin. You could actually probably even do this (with a bit more work on every test) in ChatGPT! And I'm not talking about future capabilities, I'm talking about what they're probably capable of right now. To be clear, this wouldn't replace actual A/B testing. Instead, it would help us prioritize which variants to A/B test, potentially saving time and resources and accelerating how fast we get results and lower our CPAs. Our Demand Gen teams could quickly identify the most promising options and focus their efforts and budgets there. I come from a performance marketing background in my earlier marketing life (before I was obsessively focused on all things GenAI) so this gets me really excited. I'm really curious to hear your thoughts on how this could fit into your current marketing strategies and if it's something you'd actually be interested in. Could this approach help you make more informed decisions about which ad creatives to run or which landing page variations to test?