Industry Changes is the Lantern weekly diff: a Monday-morning readout comparing this week's AI answers against last week's, across your category, your tracked competitors, and the prompts your buyers actually use. It is the operator's early-warning surface.
What the diff compares
- Category answers — the set of answers returned for the prompts in your tracked set this week vs the same set last week.
- Competitor answers — the same comparison, but scoped to tracked competitor brand names. New competitor entrants to the citation set surface here first.
- Prompt-set drift — whether tracked prompts are producing materially different answer shapes than they did the previous week.
Shift detection thresholds
| Signal | Movement threshold | Confidence minimum | Severity |
|---|---|---|---|
| Brand Visibility (one provider) | ≥ 4 pts WoW | medium (n ≥ 8) | Watch |
| Brand Visibility (aggregate) | ≥ 3 pts WoW | medium | Notable |
| Citation type mix (Social share) | ≥ 8 pts WoW | high (n ≥ 25) | Notable |
| Descriptor cloud (new "negative") | ≥ 2 distinct mentions | medium | Watch |
| New competitor entrant | 1st citation | medium | Notable |
| Lost top-cluster URL | URL drops out of cited set | high | Major |
| Brand AI Health composite | ≥ 1.5 pts WoW | medium | Notable |
False-positive handling
Three suppression mechanisms keep the diff honest. First, confidence tiering. Second, descriptor-cloud lag — the descriptor cloud is computed on a 14-day trailing window. Third, the volatility floor: categories with structurally high week-over-week swing have higher thresholds applied automatically, learned per-prompt from the trailing 60 days.
Methodology
Snapshots taken at 23:59 UTC on Sunday across all five tracked providers. Diff computation is deterministic given a snapshot pair: there is no ML model in the loop, only threshold rules and family-attribution rules.