Get the Best Out of AI—Without Bias

In the first, second, and third parts of copywriter Chris Silvestri’s series on empathy engineering—his framework for understanding customers using AI personas—he wrote about the process of gathering and analyzing data, making AI customer personas, and testing his marketing messages against his new AI-formulated customers. In the fourth and final part of the series, Chris outlines the different cognitive traps that marketers can fall into when communicating with AI and trying to understand their customers.—Kate Lee

Was this newsletter forwarded to you? Sign up to get it in your inbox.

---

Right now, your potential customers are making important decisions about whether to buy your product—or not. They're scrolling through your and your competitors’ websites, comparing features, and feeling frustrated or confused. And you? You're trying to read their minds through spreadsheets and surveys, tailoring your messaging to what you think will resonate with them.

So far in this series exploring the empathy engineering framework, we’ve gathered foundational real-world data, built our AI personas, and started testing our messaging. In step four, the final part of this series, we’ll refine our AI personas and prompts to ensure our marketing strategy is aligned with the constantly changing realities of our target audience.

Traditional data sources give you part of the picture, but they’re best at understanding the individual, not the whole. So, what if you could actually understand all of your customers—not just one, but thousands of them?

That's what makes artificial intelligence so powerful. It helps us see these human patterns at scale. But here's the catch: AI learns from us, and we humans are full of biases and mental shortcuts that can mess with our judgment. How do we get the best out of AI without letting our human quirks get in the way?

Four cognitive traps to avoid

When we start working with AI, cognitive biases can creep in. These biases—mental patterns that influence how we perceive and interpret information, often leading us to make irrational or flawed decisions—may cause us to misinterpret its insights, overlook important data, or become overly reliant on the machine's “wisdom.”

Let’s consider four cognitive biases that have the biggest impact on how we understand our customers through AI. By understanding overreliance bias, anchoring bias, confirmation bias, and availability bias, we can avoid simple mistakes and better tailor our messaging to our customers’ needs.

1. Overreliance bias: AI knows best—or does it?

We’ve all been there: Confronted with a technical problem, we Google the solution, blindly follow the first set of instructions we find, and then wonder why our computer is making strange noises. It's easy to fall into this trap with empathy engineering, too. Overreliance bias is our tendency to trust AI's recommendations too much, especially when the task is complex or the AI’s reasoning is unclear.

Imagine that I want to test a new landing page headline for our fictional software company TeamFlow. So I prompt our AI persona, “Jane, the efficiency-driven project manager”:

"Hey Jane, take a look at this headline for our landing page:

“TeamFlow: Project management zen. Get your inner peace back (and hit every deadline).” How effective do you think it is at grabbing your attention and making you want to learn more?”

Jane gets back to me with her feedback, and her inner monologue, which you can read between the “<thoughts>” tags:

AI (Jane): <thoughts> "’Project management zen’ and ‘inner peace’—those are definitely attention-grabbing, especially in my world of looming deadlines and constant juggling. </thoughts> I actually really like it! It's different from the typical headlines I see for project management tools, and it definitely piqued my interest.”

I’m thrilled! Jane's response is exactly what I was hoping for. But what if:

• The AI misunderstood the tone? Perhaps the AI missed the slightly ironic tone of the headline and interpreted “project management zen” literally, leading to an overly positive evaluation.
• The AI's data is outdated? Maybe the AI's training data doesn't reflect the latest trends in project management, and the concept of “zen” feels outdated or clichéd to real project managers.
• The headline is actually confusing? “Inner peace” might not be a tangible benefit that resonates with project managers who, lately, are more focused on measurable results.

It’s here that I should question both my own assumptions and those of the AI. I return to my TeamFlow chat, using Google AI Studio with the Gemini 1.5 model, and ask questions to validate or disprove my assumptions, specifically about the word “zen” and on whether or not we need to focus more on tangible benefits. This screenshot shows where I share Jane’s feedback with my TeamFlow AI chat and ask probing questions:

The AI replied with potential issues with the headline based on its ideal customer persona (ICP) data and offered a few suggestions with different angles to test.Source: Screenshots from Google Gemini 1.5 Pro.

The alternative headlines use different lenses—like control and automation over project management chaos—to speak to customers.

Next, I wanted to see what Jane thought about these alternatives. Do they resonate more with her? Are they more effective than the original one? I fed these options into the ICP chat, and let Jane walk me through her “thoughts”:Out of all the options we shared with Jane, a clear winner emerged, with the headline that focused more on showing ROI for project managers, like hitting deadlines, delighting clients, and getting time back:We could take this refined headline option and further iterate on it, and test it with real customers on a landing page.

---

Become a paid subscriber to Every to unlock this piece and learn about:

• The three other cognitive traps to avoid in using AI for marketing
• The framework for constructing a continuous feedback loop between human insight and AI analysis
• How empathy engineering transforms marketing from demographic guesswork to genuine human connection