Generative AI in performance marketing:
what works and what doesn’t

The consensus in digital marketing from 2024-2026 is that "AI changes everything". What nobody says is what exactly it changes, in which specific operations, and under what conditions. The difference between implementing AI in an acquisition system and merely using AI tools is the difference between measurable impact and technology theatre.

This is what I learned integrating AI into systems running several million pesos a month in digital investment.

What actually works

Creative performance analysis: the real case

When you run 80-120 live creative variations in one account, reviewing each one by hand takes hours and is usually done badly — prioritized by dashboard CPL rather than by analysis of conversion patterns per audience segment.

With an LLM connected to the performance data, that analysis takes minutes and surfaces patterns manual review never finds: which message angle works best in which demographic segment, which formats hold attention better in certain placement contexts, which headlines correlate with higher conversion rates in the CRM.

It is not magic. It is data processing at a scale a human cannot manage in the time available. The value is in the speed and volume of analysis, not in the AI "understanding" the campaign.

Generating copy hypotheses: the right framework

Using AI to write campaign copy directly is a common mistake. AI-generated copy tends to be generic, optimized to read coherently rather than to convert in specific contexts.

The correct use is different: use AI to generate hypotheses about message angles the team would not have considered. Not the final copy — the strategic questions. Which pain points in the 45-55 telecom segment are going unaddressed? Which price objections exist that the current creative doesn't answer?

What AI cannot do

AI cannot define what to optimize. That question requires understanding the business: which product has the highest LTV, where the company sits in its commercial cycle, what the margin constraints are. No language model has that context by default.

AI cannot judge whether a strategic decision is right for the business. It can tell you which ad set has the better CPL. It cannot tell you whether that ad set is producing the kind of customer the business needs right now.

AI amplifies what already exists. If the underlying system is sound — clean data, correct architecture, a defined target CAC — AI accelerates the improvement. If the underlying system is poor, AI accelerates the mistakes.