For e-commerce & shopping brands
See where your products land on the AI shelf. The carousels ChatGPT and Google AI answers show, matched against your own catalog, with imports from Shopify and Google Merchant Center, demand momentum, and honest engine coverage.
When a shopper asks an AI assistant "what's the best [product] under [price]?", the answer often is not a paragraph. It is a product carousel. Those shelves are the new endcap, and most brands have no idea whether their products are on them.
MentionFlow captures the carousels AI engines show on your tracked prompts and matches every product against your own catalog. You see where you land, who beats you to slot one, and which shopping demand is rising. This guide is also honest about which engines carry shelf data today. A shelf metric you cannot trust is worse than none.
The vocabulary, fixed up front
- A placement is one product in one slot of one carousel.
- A carousel is the set of placements from a single run. It is the shelf as the engine showed it that time.
- Placements are matched against your uploaded catalog. The numbers describe your products, not the shelf in general.
Shopping metrics run over a rolling 7-day window and count active-prompt runs only. A paused prompt's old carousels do not prop up today's numbers. Full model in Shopping overview.
Step 1 — Load your catalog
Your catalog is the bridge between "products I sell" and "products the AI shelf featured". There are three ways in. All preview first, all are re-checked on the server, and all are idempotent, so re-running corrects rather than duplicates:
| Source | Limits | Notes |
|---|---|---|
| CSV | up to 1,000 rows / 5 MB | every row is checked field by field |
| Shopify | up to 1,000 products | reads your store's public products.json; currency is left blank because that endpoint omits it |
| Google Merchant Center | RSS, Atom, or TSV | forgiving parsing of the feed |
See Product imports, plus the guides for Shopify and Google Merchant Center.
Catalog-to-placement matching is strict on purpose. A product links to a placement only with at least 80% title-token containment, at least two shared tokens, and a brand-or-URL-host gate. It is built to under-match rather than guess. A missed link can be fixed. A wrong link quietly corrupts your numbers. Importing runs a 30-day backfill over previously unattributed placements, so history fills in without stealing a placement already matched to another product. Deleting a product only unlinks it. The captured shelf data survives. Details in Catalog.
Step 2 — Read the shelf
Shopping analytics is the home screen: four summary tiles, a presence trend, a top-30 product leaderboard, the 40 most recent carousels to audit, and a CSV export. Zoom into any product on its product-detail page:
- Win rate = slot-#1 placements divided by total appearances.
- Visibility share = distinct carousels featuring the product divided by distinct carousels in the window. Holding two slots of one carousel is not double-counted.
- Average slot = mean position. It is an honest em-dash when the product never appeared, because there is no average of nothing.
- Price vs shelf compares your price to the shelf's typical price, the median of per-carousel medians, so one 12-slot carousel cannot outweigh three small ones. A delta shows only when both sides share a known currency. Otherwise it is an em-dash with the reason stated.
A shopping-gap recommendation fires in Actions when there were at least three carousels and you had zero own placements. A shelf exists and you are missing from it.
Step 3 — Track rising demand
Shopping demand is your early-warning system. It takes the real web searches assistants ran while building a carousel on your prompts, lifted word for word from the stored answer and never simulated, and ranks them four ways:
- Top: highest answer count now. The only mode that works on day one.
- Trending: growing most versus the prior window.
- Losing: declining most. A query that vanished entirely counts as the strongest decline.
- New: ran this window, absent from the prior one.
Turn any rising query into something you monitor with one Track click.
Step 4 — See how AI describes you
Product attributes reads how AI answers describe products: the material, specs, and ratings the text literally states. It lines your product up against the brands winning the most shelf space. Nothing is inferred. An attribute appears only when an answer states it word for word, ratings must be written with digits, and every value keeps its evidence quote.
Attribute extraction is rolling out. The pass is flag-gated and currently off on production, so most workspaces see an honest "not enabled yet" state rather than the comparison matrix. Nothing is fabricated to fill it in the meantime.
Which engines carry the shelf — honestly
Read this before you quote a shelf number. Coverage varies by engine, and MentionFlow tells "not sampled" apart from "awaiting vendor data" apart from a real measured zero:
| Engine | Shopping status |
|---|---|
| ChatGPT | Collecting today |
| Google AI Overviews | Collecting today |
| Google AI Mode | Collecting today |
| Copilot | Collecting today |
| Perplexity | Wired, awaiting vendor data |
| Gemini | Wired, awaiting vendor data |
Claude, DeepSeek, Grok, and Mistral answer through model APIs with no shopping surface, so they never appear. All shopping metrics blend every collecting engine into one view. The Engine coverage strip on /shopping is the only per-engine breakdown.
Two honesty gates keep multi-engine shelves clean. Google products count only when the result actually showed an AI Overview, so a plain Shopping module on an answerless page is excluded. Copilot cards count only with real commerce evidence, a price or a rating, so a list of name-drops cannot fabricate a shelf. Full detail in Engine coverage.
Shopping demand is effectively ChatGPT-sourced today. A query needs both an engine that exposes the searches it ran (ChatGPT and Perplexity) and a captured carousel on the same answer. ChatGPT is the only engine delivering both halves. Estimated volumes need DataForSEO credentials. Without them each volume is an em-dash, never a guess.
What this adds up to
You upload the catalog once. MentionFlow watches the AI shelves your shopping prompts trigger, tells you your win rate, visibility share, and average slot per product, and flags the shelves you are absent from. The demand view tells you what shoppers are asking AI right now. The coverage strip keeps you honest about which engines those numbers represent. Shopping analytics is not plan-gated and runs on every live workspace. The demo shows labelled sample data.
Related
- Shopping overview: how carousels are captured and matched.
- Product catalog: the matching gate and backfill.
- Product imports: CSV, Shopify, and Google Merchant Center.
- Shopping demand: the four momentum modes.
- Engine coverage: which engines carry the shelf today.
- Product attributes: how AI describes your product (rolling out).