Audit Whether Your Products Appear in AI Shopping Recommendations

Someone asks an assistant for the best wireless headphones under 200 euros and gets a shortlist of products and stores, with no way to know whether your catalog is in it. This workflow reads what was detected in your monitored answers, then sorts the work into tracks.

The GEO problem it solves

Treated as product SEO, this becomes rewrite the page, add markup. But shopping answers are assembled from a shifting mix of retailer listings and editorial reviews that varies by provider and by query, so a flawless page proves nothing.

Observed evidence reverses the order: start from the names detected, keeping apart what a tool returns, what the answer says, and what your catalog says.

How the workflow breaks down

1. Pull the shopping baseline. get_shopping_insights returns detected products and stores with mention counts, ratings, and two flags per store: whether it carries your products, and whether it carries a competitor's. From those flags it returns a ranked opportunities list, very-high for a store carrying competitor products and none of yours, high for one carrying both, drawn from its most-mentioned stores only. It reports detections, not inventory.

2. Sources, competitors, then the answers. get_source_domains ranks cited domains by mentions, get_source_profile details one with its top URLs, and get_source_urls lists cited URLs for an exact domain. Different data from the stores, and only the profile honours the period. get_competitor_position adds coverage, share of voice and a shopping overlap count, which counts product records and never ranks them. get_prompt_answers returns truncated previews; get_answer_details returns one full answer with its mentions in position order and its cited URLs. Which products are named is your reading of the prose.

3. Retain the gaps. get_content_opportunities returns up to ten questions where competitors appear and you do not, plus up to ten low-visibility ones, with at least two answers and under 30 percent brand presence. Neither it nor get_keyword_triggers filters for shopping, so purchase-intent questions are kept by hand.

What to do with the output

Run one track at a time. Contradictions first: a detected rating or availability your feed says is wrong, on a product you sell. It leads because it is the only track you can verify without new answers.

Then the stores ranked very-high, by descending mention count. Then those ranked high, where both appear: consistency work on a listing you control. Then prompt coverage, missed questions first. Publisher outreach last, the outcome you control least.

An empty opportunities list is a normal result, not a failed pull: when a track is empty, move to the next rather than inventing work inside it.

Save the baseline first: summary counts, store list, cited domains and the answers you read. Nothing guarantees a recommendation, so before starting a track, write down which baseline number must move.

The Complete Prompt

I want to audit whether my products appear in monitored AI shopping results. List the brands I monitor and ask which to use if there is more than one. Use the last 30 days, keeping tool fields, answer-text readings and my catalog data in separate columns.

1. Run get_shopping_insights and report the summary counts first. If it returns zero products and zero stores, say no shopping detections exist, retry over 90 days, and if still empty, stop the shopping branch and resume at step 2. Otherwise report the top products with rating and mention count, the top stores with both store flags, and the ranked opportunities, which cover only its most-mentioned stores. Invent no price, provider or purchase link. Then show me the detected names, ask which match my catalog, and carry no unmatched name onward.
2. Run get_source_domains, pick the domains relevant to my category, then get_source_profile on those and get_source_urls on an exact domain, keeping shopping stores and cited sources apart. Only the profile honours it.
3. Run get_competitor_comparison and get_competitor_position on the two competitors closest to my category, reporting brand-level figures only, and note the shopping overlap counts product records, not rankings, and ignores the period. Then select product and category prompts with get_prompts, run get_prompt_answers to shortlist, and get_answer_details for the full text: for each answer, the products named and their order, plus any price or rating stated in the text, labelled as your reading, for me to check.
4. Run get_content_opportunities and get_keyword_triggers, keeping only product discovery, comparison or purchase questions, since neither has a shopping filter. For each gap, give the question, the competitors present where I am absent, and the missing evidence.
5. Build the plan as ordered tracks. One: detected ratings or availability contradicting my feed on products I sell. Two: stores ranked very-high, by mention count. Three: stores ranked high. Four: prompt coverage gaps, missed questions first. Five: publisher outreach. On an empty track, write empty and move on. Give every line its evidence, owner and review date, and for each track the baseline number that must move.

Before track one, ask me what the tools cannot see: my prices and stock by market, and the claims I am allowed to make. Invent no query volume or revenue estimate.