What AI literacy actually means for A GC

CHELSEA WAGNER, MS CGC | July 7th, 2026

You might have heard a colleague say recently that they are "pretty good with AI." But if you follow up and ask what they mean, they might say they use ChatGPT a few times a week to help draft emails.

It’s definitely encouraging to hear genetic counselors using AI in their day-to-day workflows, but the key word in this scenario is use. AI use is not the same as AI fluency. The gap between use, literacy, and fluency is where most of the risk in our profession is hiding now. 

Here's the distinction. Use is knowing which buttons to press. Literacy is knowing what's happening when you press them, and fluency is knowing when not to. A GC who uses AI can generate an introduction to a new publication in a matter of seconds. A GC who's literate about AI knows why that draft confidently cited a study that doesn't exist and catches it. A GC who is AI fluent understands which tasks are safe to hand over in the first place.

You can be a heavy AI user and completely illiterate. That combination is the one with the biggest implications for our profession. 

Think about how we already work. When a lab report comes back, you don't just read the classification and move on. You know how the classification was made. You know ACMG criteria, you know what a PS3 versus a PP3 rests on, you know which labs you trust for which variants and why. That knowledge is what lets you identify a solid interpretation from a shaky one. You'd never counsel a patient on a result you couldn't evaluate or investigate. AI outputs deserve the exact same treatment. And most of us don't have the equivalent knowledge yet.

So what does literacy actually cover? I'd put it in four buckets.

How the thing works. Not at an engineering level. At the level where you understand that a large language model predicts likely text, it doesn't retrieve verified facts. Once you really absorb that, a lot of behavior stops being surprising. The confident fake citation makes sense. The plausible-but-wrong inheritance pattern makes sense. You stop trusting fluency as a signal of accuracy, which is the single most important habit to build.

Where it fails, and why. Hallucinations aren't random glitches. They're more likely in specific, predictable situations: niche clinical topics with thin training data, requests for citations, anything where the true answer is "there isn't good evidence yet." If you know the failure patterns, you know when to slow down.

What it does to your data. When you paste a clinical note into a consumer AI tool, where does that text go? Is it training the next model? Does your institution's agreement cover it? Call it privacy literacy: knowing the difference between a HIPAA-covered enterprise tool and the free tab you opened on your personal laptop.

When the answer is no. This is the one people skip, and it's the most important. Literacy includes a clear sense of the tasks AI has no business touching. Psychosocial assessment. Breaking bad news. Any moment where the value is the human sitting in the room. Knowing where the line is matters more than knowing any prompt.

None of this requires a technical background. I want to be clear about that, because "AI literacy" can sound like it belongs to people who code. It doesn't. The GCs who evaluate AI best are the ones who already think critically about evidence, bias, and uncertainty. That's most of us. It's the core of the training. We just haven't pointed those instincts at this particular technology yet.

That's the whole reason we’re building GC x. Not to turn genetic counselors into technologists and coders. To give the profession the same rigor about AI that we already bring to a variant classification or a risk assessment.

If you take one thing from this: stop measuring your AI skill by how often you use it. Start measuring it by how well you can judge what comes back. The counselor who uses AI twice a month but catches every error is in far better shape than the one who uses it daily and trusts it blindly.

We'll go deeper on the failure patterns, the privacy questions, and the clinical use cases over the next few weeks. For now, the reframe is enough: literacy first, use second. That order protects your patients, and it protects you.