The Prompt Lab — Modality Switching

Modality Switching

The Technique

Modality Switching instructs the model to translate information from one representational format into a structurally different one — prose to table, narrative to flowchart logic, bullets to decision tree. It works because forcing a format change requires the model to actively re-parse the source material rather than paraphrase it, surfacing structure that was latent or hidden in the original.

The Naive Prompt

Summarize this refund policy for our support team:

"Customers may request a refund within 30 days of purchase 
for unused software licenses. Refunds are not available for 
annual plans after the first 7 days, or for any add-ons 
purchased separately. Exceptions may be granted at manager 
discretion for extenuating circumstances. Refunds typically 
process in 5–10 business days via the original payment method."

Why It Falls Short

The model will almost certainly return a shorter version of the same prose — grammatically clean, but structurally identical to what a support rep already has to read. The conditions, exceptions, and timelines are still tangled together, which means a rep handling a live customer still has to do interpretive work to figure out what applies to this situation.

The Improved Prompt

Take the refund policy below and convert it into a decision 
tree a support rep can follow in real time. Use IF/THEN/ELSE 
logic. Each branch should terminate in a clear action 
("Approve refund," "Deny refund," "Escalate to manager"). 
Do not add information not in the policy.

Policy:
"Customers may request a refund within 30 days of purchase 
for unused software licenses. Refunds are not available for 
annual plans after the first 7 days, or for any add-ons 
purchased separately. Exceptions may be granted at manager 
discretion for extenuating circumstances. Refunds typically 
process in 5–10 business days via the original payment method."

Why It Works

Specifying IF/THEN/ELSE forces the model to identify every conditional in the policy and assign it a branch, making gaps and ambiguities visible (e.g., the policy never defines “extenuating circumstances,” so the escalation branch surfaces naturally). The terminal action requirement means every path resolves to something a rep can do, not just something they can read. The injected constraint — “do not add information not in the policy” — prevents the model from papering over real ambiguity with invented rules.

When to Use This

• Operational documentation: When policy, legal, or technical prose needs to become a usable artifact — checklist, decision tree, comparison table — for someone working under time pressure.
• Stress-testing source material: The act of switching formats exposes missing conditions, contradictions, or undefined edge cases that read fine in prose but break down when logic is made explicit. Use it on Claude Fable 5.1 or GPT-6 Astra when reviewing contracts or compliance docs.
• Audience translation: When the same underlying information needs two different deliverables — say, a narrative for a stakeholder memo and a table for an engineering sprint — Modality Switching generates both from one source without manual reformatting.