This antipattern is subtle because the goal is measurable. It just measures the wrong thing.

The Bad Example

“Student will produce /s/ blends in initial position of words with 80% accuracy during structured therapy activities.”

This is technically measurable. You can take data on it. But it answers only one question: “Can this child say /s/ blends in my therapy room?” It says nothing about the cafeteria, the classroom, the playground, or home, the places where intelligibility actually matters.

Why LLMs Produce This

Models default to the easiest measurement context: structured therapy. It’s the most common goal format in training data and it’s the simplest to write. Generalizing to natural contexts requires clinical reasoning about the child’s environment, something the model has no access to.

The Fix

Add a generalization context to every goal prompt:

Include a generalization condition — where and with whom will this skill be measured beyond the therapy room? Goals should target functional communication in the student's actual daily environments.

The Fixed Version

“In classroom activities and peer interactions, [Student] will produce /s/ blends in initial position of words with 80% accuracy across 3 consecutive observations by SLP, with generalization data collected during one non-therapy observation per month (e.g., classroom group activity, lunch conversation).”

Now the goal lives where the child lives.

The Takeaway

A skill performed only in therapy is a therapy skill, not a communication skill. If your goal doesn’t describe the real world, you’re measuring compliance, not competence.

Pair With

• Pattern: Family-Centered Prompting – naturally targets real-world contexts
• IEP Goal Strengthener prompt – builds in context and generalization
• Articulation domain – where clinic-only goals are most common