If your therapy sessions look like structured trials with clear antecedents, target responses, and consequences, your LLM prompts need to speak that language. The model will default to vague, narrative-style output unless you set explicit expectations for data precision.

The Core Pattern

I use a structured, discrete-trial approach. My documentation should reflect specific trial counts, accuracy percentages, cueing hierarchies, and criterion-referenced progress. Frame all goals and notes in terms of measurable behaviors, not narrative descriptions.

How This Changes Output

Without the pattern:

“Student is making good progress with articulation and is becoming more intelligible in conversation.”

With the pattern:

“Across 40 trials targeting /r/ in initial position of words, student achieved 75% accuracy (30/40) at the word level with a visual model. This represents an increase from 60% (24/40) in the previous session. Cueing hierarchy: independent > visual model > auditory model > simultaneous production.”

The second version gives you data you can chart, compare, and defend.

Prompt Modifiers for Discrete Trial Work

• “Include trial counts and accuracy percentages” – forces quantitative output
• “Describe the cueing hierarchy used” – captures the support structure
• “Organize by target and level (word, phrase, sentence, conversation)” – matches your data tracking
• “Include criterion for advancement” – specifies when to move to the next level
• “Note any error patterns observed” – captures clinical detail for treatment planning

Pair With

• Progress Note Organizer prompt – for session-to-note workflow
• Articulation domain – where discrete trial is most common
• Pattern: Family-Centered Prompting – the contrasting approach