What connected AI assistants can do

Once connected, your AI assistant works with the same data and rules as the app itself. You ask in plain language. The assistant reads your real feed data, proposes changes, and applies only what you approve.

What it can read

• Your plan, feature switches and usage against limits.
• Every feed's settings, status and last run result.
• Products in a feed, with their per feed switches and overrides.
• The exact preview output for any product. What it sees is what the channel gets.
• Merchant Center issues across your whole account, grouped worst first.
• Search Console queries for your product pages once search data is synced.
• The optimization playbooks for every feed attribute, from titles and prices to shipping and custom labels.

What it can change

• Feed settings across every editor tab.
• Per product attribute overrides and include or exclude switches.
• Feed regeneration. It can queue a fresh run after changes.

The guided optimization loop

Ask the assistant to optimize your feed and it works through the same sections as the Optimization page. Counts on the page drop as it works.

The assistant does not work from generic advice about writing good copy. Every section carries a playbook written by Digital Darts, a Google Partner agency with over 10 years growing Shopify stores through Google Ads. Your assistant gets the same rules that agency uses on client accounts.

1. Picks the worst problem first

   Rejected products come first, because a rejected product earns nothing. After that the assistant works down by lost impressions. You never have to work out the order yourself.

2. Loads the rules for that section

   Each section has its own rules, and the assistant reads them before writing anything. They cover where the important words go in a title, what Google cuts off, and which fields it matches searches against.

3. Drafts a change from your data

   Every change is built from your live product fields, your Merchant Center issues and the search queries your products already appear for. A new title uses the words real shoppers typed.

4. Waits for your approval, then verifies

   Nothing changes until you approve it. The assistant then opens the feed preview and shows you the result, so you see what Google will receive before Google does.

The same playbooks are on the Optimization page in the app, section by section. Reading them yourself and working by hand gets the identical result. The assistant is the fast way through, not a different set of rules.

There are two ways through a section, and a good assistant picks between them based on the size of your catalog.

Product by product

The assistant fetches the next affected product, reads its live issues and search queries, proposes one change, applies it after your approval, then verifies the result in the preview and moves on. That suits a small catalog and the leftovers after a bigger fix.

Field by field

The assistant reads a page of the products a section flags and looks for what they have in common. Where the fix is a rule, one change to the feed setting fixes the whole field at once. Missing product types filled from your tags is one rule, not thousands of decisions.

Bulk edits where no rule fits

Some fixes are per product by nature and still run in bulk. Titles are the clearest case. Search Console gives each product its own top query and its own missing words, so every product needs a different title and no feed setting can produce that. The assistant reads the search data for a page of products, writes a title for each, shows you the list, and applies up to 200 in one approved step.

By default those overrides apply to every feed at once, so a title pass is done once and every feed gets it. A feed can still carry its own value for a product, and that value wins in that feed only. If your feeds serve more than one language, the assistant writes titles and descriptions per feed.

The tools by name

Assistants list these tools after connecting. If your connector shows a different set, reconnect with a fresh code.

get_plan_status         Your plan, features and usage against limits
list_feeds              Every feed with status and last run
get_feed                One feed's full settings
list_products           Products in a feed with switches and overrides
preview_product         The exact feed output for one product
list_product_issues     Merchant Center issues, grouped worst first
get_search_queries      Search Console queries for a product page
get_optimization_guide  The optimization playbook per section
list_section_products   Products a section flags, in pages of 50
get_mapping_schema      The rules for building a field across the catalog
update_feed_settings    Change feed settings, after your approval
set_product_override    Override one product's attributes, after approval
set_product_overrides   Override many products at once, after approval
set_products_included   Include or exclude products, after approval
run_feed                Queue a fresh feed run
optimize_next           Fetch the next product in the optimization loop

How to ask

Clients that support MCP prompts, like Claude, also get an optimize prompt in their prompt menu. It starts the same guided loop. The identical prompt is copyable from the app's AI assistants settings page, so clients without prompt support can paste it instead.

The wording matters more than most merchants expect. A prompt that names the section, the data to use and how much to change at once gets a far better pass than a one line request. There are prompts to copy and rules for writing your own.

Limits that always apply

• Plan rules bind the assistant exactly as they bind the app. A plan without customization cannot apply overrides through chat either.
• The item id can never be changed. It is how Google identifies each item.
• Assistants never see your Google tokens or other stores' data.
• Revoking the connection in settings stops access immediately.