Setting up your prompt set: a category-by-category playbook

The prompt set is upstream of every other Lantern surface. If the Query agent is tracking the wrong prompts, every downstream metric (Visibility, Favorability, the Industry Changes diff) optimises for the wrong shelf. This piece is a category-by-category playbook for picking the prompts to track.

The four patterns

• Buyer-intent: the queries actual shoppers run. "Best subscription coffee 2026", "vitamin C serum for sensitive skin". Should be the majority of the tracked set (≥ 60%).
• Branded: your brand and direct variations. "Is Wildgrain Roasters good", "Wildgrain reviews". 10 to 20% of the set.
• Category: definitional and educational. "What is direct-trade coffee". 10 to 20% of the set.
• Comparison: "X vs Y" prompts where you appear or where a competitor does. 10 to 15% of the set.

Coffee

prompts
# Buyer-intent (sample of 8 of 30)
best subscription coffee 2026
best pour-over coffee beans
best Ethiopian Yirgacheffe single-origin
how to brew pour-over at home
best direct-trade coffee subscription
specialty coffee subscription with origin information
gift subscription for a coffee enthusiast
best decaf subscription coffee

# Branded (all 5)
is wildgrain.co good
Wildgrain Roasters reviews
Wildgrain Ethiopia Yirgacheffe tasting notes
Wildgrain subscription cancel policy
Wildgrain vs grocery-store coffee

# Category / definitional (all 5)
what is direct-trade coffee
what is washed-process coffee
how does coffee subscription work
what is single-origin coffee
how long does roasted coffee stay fresh

# Comparison (all 5)
Wildgrain vs Atlas Coffee Club
Wildgrain vs Trade Coffee
Atlas Coffee Club vs Trade Coffee
best subscription coffee for pour-over
specialty coffee subscriptions compared

Beauty and personal care

Beauty's prompt set should bias slightly more toward branded and comparison than coffee. The descriptor cloud carries more weight in this category. Nova Skincare's set is 40 prompts: 22 buyer-intent, 6 branded, 6 category, 6 comparison.

Pruning the underperformers

• A prompt that has been at 0% Visibility for 60 consecutive days, with no comparison or Citations-surface signal, is dead weight. Prune it.
• A prompt where Visibility has stayed perfectly flat for 90 days is probably mis-shaped.
• A new prompt added more than once per month signals the prompt set is not stable yet.