Adzviser Merchant Center

View saved version →

---
name: adzviser-data-mode
description: Retrieve and analyze model-visible Merchant Center reporting data with Adzviser. Use for product impressions, clicks, CTR, product rankings, customer countries, ads versus organic performance, date comparisons, and feed diagnostics. Do not use when the user explicitly wants only an interactive report preview.
---

# Merchant Center Data Mode

Use `retrieve_reporting_data` and answer from actual reporting rows.
Use only `assorted_requests.merchant_center_request` for this plugin.

1. Call `list_workspace` and choose the named workspace with Merchant Center
   access. If none is named, choose automatically only when one suitable
   workspace exists. Explain missing access without substituting accounts.
2. Call `list_metrics_and_breakdowns_merchant_center` for exact field names.
   Request only the metrics and breakdowns necessary to answer the question.
3. Resolve dates from the user's current date and timezone. Send inclusive ISO
   pairs. For N days ending two days ago, use `[D-(N+1), D-2]`; recalculate
   rather than reusing preparation dates. Use separate pairs for comparisons.
4. Build one `adzviser_request` with `workspace_name`, `date_ranges`,
   `assorted_requests.merchant_center_request` containing `metrics` and optional
   `breakdowns`, and optional `time_granularity`. Date/Week/Month/Quarter belong
   in time_granularity, not breakdowns.
5. Read model-visible `structuredContent.reportingData.MerchantCenter.data`,
   its table names, headers, date segments and `note_of_assumption`.
   Match columns by header, not assumed positions.

## Interpret results

- Totals without breakdowns appear in `Combined`. Segmented performance appears
  in `ProductPerformance`. Current details and diagnostics can appear in
  `ProductDetails&Issues`; keep them separate from historical performance.
- Clicks and impressions may be numeric strings. Convert them to numbers for
  calculations, but preserve product IDs as text, including leading zeros.
- CTR is a fractional ratio. Multiply by 100 once for percentage display.
  Calculate overall CTR from compatible click and impression totals; do not
  sum or average row CTR values. A calculated ratio with zero denominator is
  undefined. Rate fields are not additive.
- For product rankings, request Product ID and Title. Aggregate additive
  counts across title variants with the same nonblank ID within the account
  and period. Keep different IDs separate even when titles match. Use returned
  titles without claiming a variant is the latest. Rank after aggregation.
  Disclose blank-ID traffic separately instead of naming an invented product.
- Product attributes in performance reports describe the product at the time
  of its impression. Current feed attributes can differ. Do not join the two
  reports by row position or assume their product-ID formats are interchangeable.
- Marketing method separates ADS from ORGANIC. Preserve other or unknown
  methods if returned. Merchant Center performance is not Google Ads spend.
  Customer country code describes the customer, not the product's target country.
- Segmented impressions can differ from unsegmented account totals. Use the
  appropriate returned scope and disclose differences when comparing totals.
  Do not invent missing countries or redistribute unattributed impressions.
  A top-products list is not a complete account total.
- Compare periods separately. Show recent minus earlier, and divide the
  difference by earlier for percentage change when earlier is nonzero.
  State the dates. Missing values are not automatically zero.
- Product details, availability and issue records are current snapshots;
  they are not filtered to the requested period by this connector. Do not
  describe issue counts as newly occurring within that date range.
- Product-details output can repeat a product for multiple issue codes.
  Count unique product IDs for affected products; count distinct product/issue
  pairs for issue

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: adzviser-preview-ui-mode
description: Open an interactive Merchant Center report preview in ChatGPT so the user can inspect, edit, and run it. Use only for an explicit Preview UI, UI Mode, embedded report, or interactive report request. Do not use for raw rows, calculations, rankings, comparisons, audits, or ordinary analysis.
---

# Merchant Center Preview UI

Use only for explicit interactive-preview requests. Use `adzviser-data-mode`
when the answer requires actual reporting rows.

1. Call `list_workspace` and choose the named workspace with Merchant Center
   access. Ask when none is named and multiple suitable workspaces remain.
2. Call `list_metrics_and_breakdowns_merchant_center` for exact fields.
   Request only necessary fields. Keep product detail and issue reports
   separate from performance analysis.
3. Resolve inclusive dates using the current date and user's timezone. For
   N days ending two days ago, use `[D-(N+1), D-2]`.
4. Build `adzviser_request` with `workspace_name`, `date_ranges`, optional
   `time_granularity`, and `assorted_requests.merchant_center_request`
   containing `metrics` and optional `breakdowns`.
5. Call `preview_and_retrieve_reporting_data` once. Let the user edit and run
   it. A preview alone contains no final reporting rows; do not invent values
   or claim to analyze hidden component data.

CTR is a fractional ratio, displayed as a percentage after multiplying by 100.
Numeric counts may be strings; preserve Product ID as text. Current product
details and issues are not historical date-filtered performance. Repeated issue
rows do not represent additional products. Customer country and target country
are different fields. Keep currencies and reporting scopes separate.

Never send `client_id`, `reporting_mode`, `_widget_action`, conversation history
or unrelated context. Never call `set_reporting_mode` from the conversation.
Date/Week/Month/Quarter belong in time_granularity, not breakdowns. Do not add
unsupported filters or other sources.

Let the component handle Run and mode controls. Analyze only model-visible rows.
If the user switches to analysis and rows are absent, use `retrieve_reporting_data`
through Data Mode. Explain errors and missing access accurately. This plugin
cannot edit or delete products, fix disapprovals, or change ad bids and budgets.

Referenced files: 1 — file references are metadata, not package contents.