AI Shopping Visibility

Ordinary AI visibility asks whether AI mentions a brand. Shopping visibility asks the harder, more commercial question: when a buyer asks AI what to buy, does it recommend you? Open a client's AI Visibility project and go to the Shopping tab. It focuses on purchase-intent questions — the "best X for Y", "top X under $Z", "what X should I buy" style of prompt that people ask when they're ready to spend. Shopping visibility is part of AI Visibility, not a separate product, so it uses the same models, the same daily scans, and the same honest measurement (sample size and margin of error on every figure). Nothing is estimated — every number comes from real scan runs of prompts tagged as shopping-intent.

The headline metric is your shopping recommendation rate: across all the purchase-intent runs, how often did AI actually recommend your brand? It's shown as a percentage with the exact sample size ("27/50 shopping runs recommend you", n = 50 runs) and a ± margin of error, with a low-confidence note when the sample is still small (under ten runs). That keeps a single lucky answer from looking like a trend. Right below it, the "You vs competitors — shopping recommendations" table lines your recommendation rate up against each competitor you track, each with its own rate, margin of error and run count. Alongside the rate, each brand shows its share of recs — of every brand recommendation made across your shopping runs, the fraction that went to that brand (its recommendations ÷ all tracked brands' recommendations). Where the rate answers "how often is this brand recommended?", share of recs answers the competitive question "of every recommendation given, who is winning?" — a true share-of-voice for purchase-intent answers that sums to roughly 100% across the brands shown. Together they tell you whether AI is steering ready-to-buy shoppers toward you or toward a rival. Set a product type / category on the tab and CrunchJunkie seeds sensible purchase-intent prompt templates for it; you can edit or add your own at any time. At the product level, each tracked product also shows a head-to-head win rate alongside its recommendation rate. Where the recommendation rate is over every shopping run, the win rate is over the "decisive" runs only — the runs where AI recommended at least one tracked product (yours or a competitor's) — and is the share of those that included this one. It answers a sharper question: when AI actually picks something in this category, how often is it you? Because it ignores runs where nothing was recommended, it isn't diluted by prompts that produced no pick, and the decisive-run count is shown so the number always carries its sample size.

Beyond the brand, you can track individual products and see the recommendation rate for each one — so you know exactly which SKUs AI puts in front of buyers, not just whether your brand comes up. Add your own products (and competitors' products) and each becomes a tracked item scanned like any other prompt. CrunchJunkie also runs a discovery view: "Products AI recommended that you're not tracking". When a shopping answer names a specific product you haven't added, it surfaces there with how often it appeared (e.g. "9/12 runs"), and a one-click Track button adds it to your list. It's the fastest way to find the products already winning in AI answers — yours or a competitor's — and start measuring them. The flip side is the "Your products AI never recommends" panel: your own tracked products that weren't recommended in a single shopping run over the current sample. These are your clearest opportunities — the SKUs AI is overlooking. Pair it with the Feed AI-readiness panel: fix the feed gaps (GTIN, availability, price, images, reviews) and product-page content, then re-scan to see whether they start getting picked up. It only ever lists your own products and is drawn from real scan data, so read it with the sample size in mind at small run counts. Each tracked SKU generates a purchase-intent prompt (one, occasionally two) and is scanned on your project's schedule. Because scanning costs money, tracking is capped per plan: Solo 25, Starter 100, Pro 500, Agency 2,000, and Scale unlimited tracked SKUs. Only SKUs you actually track are scanned, and the tab always shows the projected checks per month before you commit — so cost is transparent. With your own API keys, per-SKU scanning is free; on Managed AI it's metered by check like the rest of AI Visibility.

You don't have to add products by hand. CrunchJunkie imports your catalog from a standard product feed in the Product feed section of the Shopping tab. Three formats are supported: Google Shopping XML (the RSS/Atom feed with the g: namespace that Merchant Center produces), tab-separated (TSV), and comma-separated (CSV). You can either paste a feed URL for CrunchJunkie to fetch, or upload a file (.xml, .csv, .tsv or .txt). CrunchJunkie reads each product's id/SKU, title, brand, product type/category, link and price, mapping common column names automatically (so "sku", "mpn" or "offerid" all resolve to the id, for example). Products without a title are skipped rather than guessed, and very large feeds are capped at 50,000 items. Imported products don't start scanning on their own — importing only populates the catalog. From there you pick which ones to track, filter by product type, brand or price range, and use "Track all matching" to bulk-track a filtered set. The bulk action shows a preview of the projected added checks and new total first, and never tracks past your plan's cap (SKUs beyond it are skipped, not tracked).

If your catalog already lives in Google Merchant Center, connect it directly instead of exporting a feed. Under Data → Integrations (or the Merchant Center card on the Shopping tab), click Connect and authorize with Google. The connection uses Google's Merchant API and requests the Merchant Center permission (auth/content) — you must leave that permission ticked on Google's consent screen, or the connection can't read your catalog. Access is read-only: CrunchJunkie only reads your product list, never writes to your Merchant Center. After connecting, pick the Merchant Center account to import from (sub-accounts under an MCC are listed too), and click Sync now to pull the catalog. Each product is normalised into the same shape as a feed import, so tracking, filtering and bulk actions all work identically. Syncing is user-initiated — click Sync again whenever your catalog changes to refresh it. You can disconnect at any time from the same panel, and revoke access entirely from your Google account.

Once you've imported a catalog (by feed or Merchant Center), the Product feed section shows a Feed AI-readiness score out of 100. It measures how complete your catalog is on the specific fields AI shopping assistants — ChatGPT Shopping, Google AI Mode and Overviews, Perplexity — actually rely on when they decide what to recommend: GTIN/barcode, availability, price, image, product type/category, product link, and a descriptive title. Google Merchant Center is the connective tissue behind all three engines, so how complete your feed is, is the single most controllable lever on AI shopping visibility. The score is weighted field coverage across your imported SKUs — each field is weighted by how much it matters to AI shopping (GTIN, availability and price weigh most), and each row shows the exact coverage ("82% (410/500)") with a green/amber/red status. Nothing is estimated: it's a straight count of which SKUs actually carry each field. A "Biggest win" line calls out the highest-impact gap to fix first. We're honest about what we can't see too — product reviews (review count and rating) are a strong AI-shopping signal that isn't captured in a standard feed import, so we flag it as a note rather than scoring it. Fix the gaps in Merchant Center or your feed, re-import, and the score updates.

When AI answers a purchase-intent question, it usually pulls from live web sources first. The "Where AI looks for shopping answers" panel shows which domains the AI referenced across your shopping runs, ranked by how many runs cited each one, and tags the well-known ones as your own site, a marketplace (Amazon, eBay, Otto, Idealo…), or community/reviews (Reddit, Trustpilot, YouTube…). It's computed from the sources already captured on each scan — no extra scans are run. The headline is your own-site share: how often the AI leaned on your own pages when recommending in your category. A low share means AI is deciding from marketplaces and third-party pages rather than your site — which tells you exactly where to earn citations (your product pages) to influence what it recommends. As with every figure here, it's over the real run count and a domain with zero runs never appears. (Own-site share needs the brand's website set in AI Visibility setup.)

Getting the AI to look at your own product pages is only half the battle — the pages also have to be readable to a shopping agent. AI shopping answers lean heavily on structured data (schema.org JSON-LD), and three signals do the most work: a Product block (name and brand), an Offer (price and availability), and an AggregateRating (review stars and count). Pages that carry all three are far more likely to be picked, and to be quoted with a real price and rating rather than a guess. The Product-page schema panel on the Shopping tab audits exactly this across the product pages in your imported feed. Click "Audit pages" and CrunchJunkie crawls a bounded batch of your product URLs, parses each page's JSON-LD, and reports the share of reachable pages that expose each of the three signals — for example "82% Product, 74% Offer, 41% Review rating" — with the exact page counts behind each figure. It's accuracy-first: a signal is only marked present when the page's markup actually declares it, and pages that can't be fetched are left out of the coverage denominator rather than counted as failing. A "Product pages missing schema" list ranks the weakest pages worst-first (a missing Product block counts most), with a per-page Product / Offer / Rating tick so you can see at a glance what to add. Fix the missing JSON-LD on those pages and re-audit. Results are cached per URL and checked in bounded batches, so the audit stays fast and cheap, and a URL checked for one brand is reused for all.

Good purchase-intent prompts mirror how buyers actually shop. The Shopping tab seeds templates from your product type, and you can also let Crunch write a tailored set: click "Generate with Crunch" and it produces purchase-intent questions written for what you sell, marked with a "Tailored by Crunch" badge. It runs only when you click — nothing is generated automatically on page load — and if no AI model is connected yet, it falls back to sensible static templates and tells you so. Generated prompts are deliberately unbranded category questions (for example "what's the best moisturiser for oily skin under $40?") rather than prompts that name your brand. That's the point: shopping visibility measures whether AI recommends you when the buyer hasn't already picked you — so the prompt must not put your name in the model's mouth. Review, edit and add the generated prompts, then they scan alongside the rest of your shopping prompts. Like the rest of Crunch, generation needs an available AI model (Managed AI or your own key).

Shopping data flows straight into client reports. The AI Visibility report template includes a "Shopping visibility" table and a "Top recommended products" table, both driven by real scan data, so a client sees not just that they're visible in AI but whether AI recommends them and their products for the sale. You can add any of this to a report the same way: drop in a Table widget, choose the AI Visibility source, and set its dimension. Every shopping and product data point on the Shopping tab is available as a widget — Shopping (brand recommendation rate and share of recs vs competitors), Product (top recommended products), SKU (tracked feed products), Product Schema (share of your product pages exposing Product / Offer / review-rating structured data), Feed Readiness (per-field feed coverage), Shopping Sources (the domains AI leans on for shopping answers, with your own-site share), and Source Health (cited pages that are dead or retracted). The Product Schema and Source Health widgets read the cached audit results, so run the audit / source check on the tab first; the others compute from real scan data. As with every AI-visibility figure, the numbers carry their sample size so the measurement is never over-claimed.