Lighthouse › Settings

Products table columns

Checked columns are visible on the Products page. These settings are saved in this browser.

Adjusted brand names

Left: brand on the Amazon listing. Right: name people type when they search the brand. We use the right-hand name for Brand Analytics search frequency. Each country has its own list. Click Save after you add a row.

Listing brand People search

AI prompts

Prompt text comes from deployed code (index.js) — edit there and deploy. Here you pick the model and which data blocks go in (saved on Signal, shared with the team).

How this works

What these prompts do. When you open a product page and click an analysis button, Lighthouse sends one of these prompts to an AI model. You get back the report you see on the page.

The blocks:

  • Market Snapshot — runs once per product when you click Run market snapshot. Reads up to 104 weeks of weekly tables for our product and every Tracked competitor (price, BSR, rank, coupon, SQP share, traffic) plus our economics and Product Events. Returns a markdown read of the market. Versions can pick which data tables go into the prompt.
  • Weekly AI Advisor — weekly call prep: last week's metrics checked against Product Brief and Market Brief ({{OUR_BRIEF}}, {{COMPETITORS_BRIEF}}). Same weekly data tables as Market Snapshot.
  • Ask Lighthouse — the chat. Runs every time someone asks a question. Gets the question, the products and weeks picked for it, the data blocks, the Brief and Events, and recent chat. Returns a short markdown answer with the numbers used.
  • PPC Snapshot — runs once per product when you click Run PPC analysis. Reads 13 weeks of ad spend, sales, ACoS, TACoS, CVR, CTR, CPC, CPR. Returns a 5-box report (Verdict / Waste / Wins / Targets / Alerts). Requires a PPC Goal saved on the product. Versions can pick which data tables go into the prompt, same as Market Snapshot.
  • Common analysis guidelines — shared text in code, inserted into Market Snapshot, Weekly AI Advisor, and Ask Lighthouse at {{COMMON_ANALYSIS_GUIDELINES}}. Change in index.js and deploy.

Placeholders. The {{LIKE_THIS}} bits inside a prompt are filled in at runtime with real data (ASIN, brand, price tables, etc.). Don't remove them unless you know what you're doing.

Model picker. Each prompt has a Model dropdown: Claude Opus 5 (default), Fable 5.1, Opus 4.8 (faster legacy), or Gemini Flash Lite.

When edits take effect. Save updates the database immediately. New AI runs pick up the new wording or model within ~60 seconds (a small cache on Lighthouse). Already-cached reports on product pages do not auto-refresh. To see the new prompt in action, open a product page and hit Re-run.

Versions. Each prompt can have several saved versions. Duplicate to try a new template without losing the old one. Pick Use this version to make it the one AI runs use. Only the version marked ACTIVE is sent to the model.

Loading prompts…