Prompt Runs (Monitor)

Prompt Runs is eCommerce Insights's side-by-side AI engine monitor. You configure a prompt set; eCommerce Insights fans it out to ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews in parallel. The result panel shows mention detection, citation extraction, competitor rivalry, and cost — one row per (engine, prompt) pair.

What it does

Prompt Runs is the closest thing eCommerce Insights has to a measurement loop. You define a prompt set. eCommerce Insights runs each prompt across five engines in parallel. Results stream back into a side-by-side panel. The pattern of who-cites-whom is what tells you whether your visibility work is moving the needle.

Neutral vs persona-segmented runs

Two run modes.

Neutral

The prompt is sent as-is. Useful for category-defining queries where you want to see what a "default user" gets.

Persona-segmented

The prompt is prefixed with a persona context drawn from Personas. The same query, asked from a "first-time apartment dweller" vs a "long-time enthusiast", often surfaces different SKUs. This is where most teams find their hidden gaps.

Cost preview

Before submit, eCommerce Insights shows an estimated cost in tokens and dollars based on prompt length, engine count, and persona count. The cost preview is helpful for sessions with many prompts or many personas; you can drop engines or personas to fit a budget before running.

How mention detection works

Mention detection runs in two passes. The first pass parses the raw engine response and extracts named citations (URLs, source titles). The second pass runs an LLM comparison between the response text and your tracked brand and SKU list, with synonym tolerance.

For each (engine, prompt) cell, you get five counts: brand mentions, SKU mentions (broken down by SKU), competitor mentions, neutral citations (review sites, comparison articles), and the engine's verdict on whether the answer is "complete" (cited 3+ products) or "thin" (cited 0-1).

Outputs

Per cell, six fields.

• Raw answer — the engine's full response, expandable.
• Cited sources — URLs the engine pointed at, with domain authority context.
• Brand mentions — count and quoted excerpts.
• SKU mentions — count per SKU, sortable.
• Competitor mentions — count per tracked competitor; new competitors flagged for review.
• Verdict — one of: You cited, Competitor cited first, Category answered without you, Engine declined to answer.

Scheduling and history

A prompt set can be saved and re-run on a schedule (daily, weekly, monthly). See Scheduler. The History view shows every past session with a delta highlight against the previous run for the same prompt.

Drift is what to look for in History. A SKU that was cited last week but is silent this week is the signal — pair it with a fresh PDP Score on the same SKU to see if a content change is implicated.

Common questions