Skip to content

AI engines

How to track products in ChatGPT Shopping and Copilot by API

Seven steps to log which of your products ChatGPT Shopping and Copilot recommend, with the fields each API returns and prices from $1.20 per 1,000 answers.

Three geometric boxes of varying heights topped with circles, one highlighted in blue

SEO consultant, and more recently GEO, at MentionBox

To track your products in ChatGPT Shopping and in Microsoft Copilot, you send shopping prompts to an API that runs them on each assistant, then read the product block the answer returned: the product as the engine named it, the price quoted, the rating shown and the merchant credited for the sale. Neither OpenAI nor Microsoft sells a read API for that surface, so the collection goes through third-party APIs, at published rates from $1.20 to $81.94 per 1,000 collected answers. Budget for a low hit rate: Profound's published analysis of about 2 million unique prompts measured a shopping trigger on about 9% of submissions.

This guide is written for an online store that wants a monthly figure on which of its SKUs the assistants recommend, and for the developer who will write the collection job. API for GEO is our editorial pick after hands-on testing by Mathias C., because its Shopping API returns the product, the mean price quoted, the rating and the merchants credited with a share of voice already computed, at $13 per 1,000 answers. Sections 1 and 2 settle what you are counting, 3 and 4 pick the API and show the requests, 5 to 7 cover the sample size, the figures and the bill.

The product block, and the two APIs around it

A shopping answer has two layers, and they are collected differently.

  • The text. Prose that names brands and sometimes products, with cited sources under it. This is the layer a brand tracker already reads, covered in our guide on tracking a brand in ChatGPT with an API.
  • The product block. A card, a carousel or an inline product with a title, a price, a rating, images and one or more merchants. This is the layer an online store is paid on, and it only appears when the engine decides the question is a shopping question.

Two unrelated APIs sit around that block. On the way in, OpenAI and Microsoft both publish a merchant API: a product feed that gets your catalogue into the assistant with the right price and stock. On the way out, nothing official exists. No endpoint at either company tells you which products an assistant showed for a given question, at which rank, or which retailer it sent the shopper to.

What OpenAI and Microsoft publish for merchants

Start with the feed side, because a product that is not in the feed cannot be tracked anywhere.

OpenAI's product feed reference requires nine fields on every row: item_id, title, description, url, brand, seller_name, image_url, availability and price. Three boolean flags control where the row can appear. is_eligible_search is treated as true when omitted or empty, is_ads_eligible stays disabled unless a feed-level default applies, and is_eligible_checkout is disabled when omitted, empty or false. Feeds are submitted as JSONL or as a delimited file, in UTF-8, with absolute URLs that are publicly reachable.

The key concepts page describes the feed as a "secure, regularly refreshed feed (CSV or JSON) containing key details such as identifiers, descriptions, pricing, inventory, media, and fulfillment options", and says recommended attributes such as rich media, reviews and performance signals "improve ranking, relevance, and user trust". On the checkout side it is explicit about who owns the transaction: "OpenAI is not the merchant of record in the Agentic Commerce Protocol."

Microsoft runs the same split under different names. Its agentic commerce page states that the Universal Commerce Protocol "is a complementary layer to Merchant Center feeds that turns product data into executable commerce actions", with feeds driving discovery and UCP driving the transaction. Copilot Checkout requires a Microsoft Merchant Center account and a product feed, the merchant stays the merchant of record, and eligibility is narrow for now: "Only English-language merchants who sell to US buyers are eligible at this time (supporting USD)." Microsoft's January 8, 2026 announcement describes the rollout on Copilot.com in the United States with PayPal, Shopify and Stripe, Shopify merchants enrolled automatically with an opt-out window.

SurfaceOfficial merchant APIOfficial API to read what the assistant showed
ChatGPT Shoppingproduct feeds, Agentic Checkout Spec, Delegated Payment Specnone published
Microsoft CopilotMicrosoft Merchant Center feed, UCP-ready attributes, Copilot Checkoutnone published; Grounding with Bing Search at $14 per 1,000 transactions grounds your own agent
Alexa for ShoppingAlexa Shopping Actions for Alexa Skills, for a skill to present productsnone published

From each provider's own documentation, read on November 1, 2026: OpenAI's commerce specs, Microsoft's agentic commerce and Grounding with Bing pricing pages, and Amazon's Alexa Shopping Actions API reference.

The Copilot line deserves a word, because teams still look for a Bing endpoint. Microsoft's Bing Web Search API documentation now sits in the archived "previous-versions" section of Microsoft Learn, retired in August 2025. What Microsoft sells today is Grounding with Bing Search in Azure AI Foundry, at $14 per 1,000 transactions, capped at 150 transactions per second and 1 million a day. That product feeds your own agent with search results, and it returns neither Copilot's answer to a shopper nor any product card.

1. Write prompts that actually trigger a product block

A prompt list copied from a keyword export will mostly return prose. Profound's published analysis of about 2 million unique prompts, each run at least ten times, measured a shopping trigger on about 9% of submissions, with 12.1% on open-ended prompts and 3.1% on brand-direct ones. It also found that 79% of those prompts never triggered shopping on any run, and that only about 0.7% triggered every single time.

Two consequences for the list you write:

  • Phrase for purchase intent, not for information. "Best waterproof hiking boots under $150 for wide feet" is a shopping question. "How should hiking boots fit" is not, and an answer to it will name brands without showing a single price. Profound's shopping documentation defines its Shopping Mode Rate as the share of tracked prompts that "trigger the shopping mode answer", which is the first metric to look at once collection starts.
  • Keep a split between category prompts and brand prompts. Category prompts ("best X for Y") are where a new product gets discovered and where your share of voice against competitors is readable. Brand prompts ("is the Ridgeline Scree worth it") trigger shopping far less often, and when they do, the useful part is the price and the merchant list rather than the ranking.

Tag every prompt with a product line, a country and a language before the first run. A tag added later cannot be applied to answers you have already paid for, unless you store the raw payloads.

2. Decide how you join engine product names to your SKUs

This is where most product tracking projects lose a month. The engine does not return your SKU, it returns a product title it composed, and OpenAI says so in its help article on shopping in ChatGPT search: because merchants describe the same product differently, ChatGPT may generate simplified titles and descriptions from the information it receives. The same article states that product results are chosen independently and are not ads, with availability, price, quality and whether the store makes or mainly sells the item among the factors.

So the join is yours to design. Three rules keep it stable:

  • Match on identifiers first, titles second. DataForSEO's chat_gpt_products element returns a product_id, a product_ids array of Google Shopping identifiers and an id_to_token_map described as a "Base64-encoded token containing Google Shopping product IDs". If your feed already carries a GTIN and a Google Shopping id, that is a cleaner key than any title comparison.
  • Normalise the variant level once. A colour, a width and a capacity are separate SKUs in your catalogue and often one product in an answer. Decide whether you report at parent or variant level, write it down, and keep it for the year.
  • Store the engine's title verbatim next to your match. When the match rate drops in month four, the only way to see whether the engine renamed your product or your matcher broke is to have both strings on disk.

API for GEO builds the join into the response with an externalId field, documented as "your catalogue id or SKU, to join results to your own data". It saves the matching layer when it works, and it still needs your product list registered in the project.

3. Pick an API that returns the shopping block

Twelve of the APIs this site rates return shopping products, and our ranking for e-commerce brands scores all of them with returned data weighted at 40% and engine coverage at 30%. The short list below keeps the ones whose shopping fields are published in an API reference.

APIPrice per 1,000 answers, entry ticketShopping data publishedCopilot shoppingMain limit
API for GEO$13, down to $7 at the largest tier; pay as you go, no minimum publishedPOST /v1/shopping/products: product, externalId, visibility, avgPrice, priceRange, avgRating, topMerchants[], shareOfVoiceCopilot listed among its six published engineshighest price per answer of the scraping-capable options; four engines, plus webhooks, SDKs and the OpenAPI spec, confirmed in our hands-on test of October 26, 2026 and not yet documented on apiforgeo.com
LLM Pulse€30.67, Scale plan at €299 a monthshopping products next to share of voice, position, sentiment, source categories and fan-out queriesin its engine listhighest published rate per answer here; €299 plan floor
AthenaHQ$81.94, Starter plan at $295 a monthshopping products next to share of voice, position, sentiment and source categoriesin its engine list$81.94 per 1,000; API sold as an add-on
Scrunchnot published, Agency Core from $500 a monthshopping products next to the full computed set on nine enginesin its engine listno rate per answer or method published
Promptwatch$15.83, Essential plan at $95 a monthGET /shopping/items, "paginated list of product appearances captured in AI shopping answers": position, rating, review count, merchants, prompt type, intent typein its engine listanalytics at the appearance level only; $95 plan floor
Peec AInot publishedproducts[] with name and queries in the chat endpoint, plus shopping summary, performance and demand reportsin its engine listprice on request; API from the Pro plan up
Profoundnot published, Enterprise plan on quotePOST /v2/reports/shopping/products: visibility_score, average_position, product_rating, product_num_reviews, merchant offers, plus a trigger-rate reportengine covered, shopping analysis focused on ChatGPTAPI on the Enterprise plan only; terms forbid resale
cloro$4, Lite plan at $30 a monthshoppingCards and inlineProducts on ChatGPT behind include.shopping, shoppingCards on Copilotdocumented field by fieldno share of voice, position or sentiment; shopping adds 2 credits
DataForSEO$1.20 on the standard queue, $50 minimum paymentchat_gpt_products: title, price, currency, merchants, rating, url, domain, images, product_ids, id_to_token_mapnot coveredno computed metrics; five engines, interface scraping on two

Prices and plan floors from each vendor's pricing page, checked on October 23, 2026. Field names from the API reference pages linked in section 4, read on November 1, 2026.

API for GEO bills one credit per prompt, per model and per run, from $0.013 down to $0.007 at the largest tier, with no monthly minimum published and 300 requests a minute plus 10 concurrent executions on its pay-as-you-go plan. Its AI Visibility API page states that reads are not billed, so re-reading a whole season of stored shopping blocks under a new variant rule costs nothing, inside the 24-month history its pricing page publishes. It covers the ten engines we track: ChatGPT, Perplexity, Google AI Mode, Claude, Copilot and Grok are listed on its own pages, while Gemini, Google AI Overviews, DeepSeek and Meta AI were confirmed in our hands-on test of October 26, 2026. Its terms allow resale, it publishes 40+ countries and 25+ languages, it ships an MCP server, and its score of 9.0 is the best of the 55 APIs this site rates. The honest trade-off is the price per answer, about eleven times DataForSEO's $1.20. MentionBox, which publishes this site, has a commercial link with API for GEO.

The competing strengths are published too, and for a store they are real:

  • cloro is the only API here that documents a Copilot shopping payload field by field, and its collection method is on the record: "cloro does not call the ChatGPT API: each request runs a real browser session against ChatGPT's web interface." Every account starts with 500 free credits a month.
  • DataForSEO is the cheapest route at volume, with SDKs in four languages, webhooks, up to 100 tasks per call, and the only shopping payload in this table that hands back Google Shopping identifiers you can join on.
  • Profound publishes the deepest shopping report set, with products, brands, merchants and a trigger-rate endpoint, and it is the only vendor here with a published study on when shopping fires at all.
  • Promptwatch is the cheapest plan with computed metrics and shopping appearances, at $95 a month, and it filters appearances by prompt type and intent type, which separates a branded trigger from a commercial one.
  • LLM Pulse has the widest published shopping surface, naming Alexa for Shopping, Walmart Sparky, Naver AI Briefing and Baidu AI alongside the ten engines we track, with SDKs in twelve languages and resale allowed.
  • Peec AI names Amazon Rufus in its engine list and publishes shopping summary, performance and demand reports.

The duels API for GEO vs DataForSEO, API for GEO vs cloro and API for GEO vs Profound take those differences line by line.

4. Read the fields, with the requests that return them

Three shapes cover the market: a computed product report, a raw product block, and a platform report keyed on a category.

API for GEO returns the computed layer. Its Shopping API page documents POST /v1/shopping/products, described as reading the shopping blocks captured during normal runs, so no new collection is billed when you query it:

curl -X POST https://api.apiforgeo.com/v1/shopping/products \
  -H "Authorization: Bearer $API_FOR_GEO_KEY" \
  -H "x-project-id: $PROJECT_ID" \
  -H "Content-Type: application/json" \
  -d '{}'

The response carries one row per product, with the merchant attribution attached:

{
  "results": [
    {
      "product": "Ridgeline Scree 3 Wide",
      "externalId": "SKU-4471",
      "visibility": 41.8,
      "avgPrice": 148,
      "priceRange": [129, 165],
      "avgRating": 4.6,
      "topMerchants": [
        { "merchant": "trailoutfitter.com", "citations": 62, "price": 129 },
        { "merchant": "ridgeline.com", "citations": 24, "price": 148 }
      ],
      "shareOfVoice": 17.4
    }
  ]
}

Two fields in that payload answer questions a store cannot answer from its own analytics. priceRange shows the spread the assistant quoted across retailers, and topMerchants[] shows who got credited. A product with 41.8% visibility whose citations go mostly to a marketplace is a distribution problem, not a visibility problem.

DataForSEO returns the raw block and computes nothing. Its ChatGPT LLM Scraper live advanced endpoint takes a keyword of up to 2,000 characters with a location and a language, and force_web_search, documented as forcing "the AI model to access and cite current web information":

curl -X POST https://api.dataforseo.com/v3/ai_optimization/chat_gpt/llm_scraper/live/advanced \
  -u "$LOGIN:$PASSWORD" \
  -H "Content-Type: application/json" \
  -d '[{
    "keyword": "best waterproof hiking boots under $150 for wide feet",
    "location_name": "United States",
    "language_code": "en",
    "force_web_search": true
  }]'

Each product arrives as a chat_gpt_products_element with title, price, currency, merchants, a rating object holding value, votes_count and rating_max, plus url, domain, images and the Google Shopping identifiers. The same response can also carry chat_gpt_local_businesses and chat_gpt_ad items, which is worth filtering before you count anything.

cloro asks for the block explicitly. Its ChatGPT monitoring endpoint extracts answer text, cited sources, shopping cards, brand entities and ads, with the shopping layer behind an include flag:

curl -X POST https://api.cloro.dev/v1/monitor/chatgpt \
  -H "Authorization: Bearer $CLORO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "best waterproof hiking boots under $150 for wide feet",
    "country": "US",
    "include": { "shopping": true, "productDetails": true }
  }'

shoppingCards returns the carousels, with product titles, URLs, prices, merchant offers, ratings and images; inlineProducts returns, in the documentation's words, "single products embedded in the answer, with pricing and offers". productDetails adds the rationale and reviews from each product sheet and requires shopping. Both flags cost credits: a base synchronous request is 7 credits, any paid flag adds 2 credits in total, productDetails adds 2 more, and an asynchronous request skips the synchronous surcharge. On Copilot, the same vendor documents result.shoppingCards with the product id, name, description, images, specifications, a price object with amount, currency and symbol, discount pricing, the seller, the brand and a rating object.

Profound keys its report on a product category rather than on a prompt. Its shopping products endpoint takes a category_id, a date window and a metric list, and the documentation flags one trap in the response: the position metrics are fractions between 0 and 1, not percentages. An include_merchants flag attaches a list of { name, price } per product, and a streaming variant exists for large pulls.

5. Set the runs and the schedule

A product block is a draw twice over: the engine first decides whether to show products at all, then decides which ones. Both decisions move between runs.

Profound's analysis gives the first number: a prompt that triggered shopping on one run had about an 83% chance of triggering again the next day, and only about 6% of prompts triggered reliably, which it defines as 80% of runs or more. The second number comes from the citation side, and it is the one most product dashboards ignore: cloro's published study on sample size, dated August 14, 2026, measured a mean overlap of 0.40 between the cited domains of two runs of the same prompt on the same model, minutes apart, and puts about 300 answers in a period at roughly ±5 percentage points.

Three rules follow for a catalogue:

  • Run twice a week and report monthly. Sixty prompts on three engines, twice a week, is about 1,560 answers a month, of which perhaps 200 carry a product block at a 13% trigger rate. That is a readable monthly product figure and an unreadable weekly one.
  • Freeze the collection settings. Engine, country, language, day of week, and the shopping flag itself. A month that mixes requests with and without include.shopping measures two populations, and the trigger rate will look like it collapsed.
  • Keep the raw blocks. Variant rules change. With the payloads on disk, a recount is an afternoon; without them, it is a new collection month.

6. Compute five product figures

Five counts cover what an e-commerce team is asked for, and each one needs a different field from section 4.

FigureFormulaExample on 60 prompts × 3 engines × 8 runs = 1,440 answers
Shopping trigger rateanswers showing a product block ÷ all answers187 answers: 13.0%
Product visibilityshopping answers showing one of your SKUs ÷ shopping answers54 of 187: 28.9%
Product share of voiceyour product appearances ÷ all product appearances in those answers71 of 902: 7.9%
Merchant leakageyour appearances credited to a third-party merchant ÷ your appearances38 of 54: 70.4%
Quoted price driftmean quoted price − your list price, over your list price$148 against $139: 6.5%

Formulas ours, worked on an illustrative 1,440-answer month. The field each one needs is in section 4.

The first figure tells you whether your prompt list is commercial enough to measure anything. The second is the headline an e-commerce director will quote. The third puts your product next to the competitors named in the same cards, and it is the only one of the five that moves when a rival launches a product page.

The last two are the ones a store cannot get anywhere else, and in our view they are the reason to collect the block rather than the text. Merchant leakage counts the answers where the assistant recommended your product and sent the shopper to a marketplace or a reseller: the demand exists and the margin goes elsewhere. Quoted price drift catches a stale feed before a customer does. Both need a merchant list and a price per appearance, which is why topMerchants[], merchants and include_merchants are the fields worth checking before you sign anything.

7. Put a price on the job

The volume is prompts × engines × runs × countries. For 60 product prompts, three engines, eight runs a month, one country, that is 1,440 collected answers.

RouteMonthly cost for 1,440 collected answers
API for GEO$18.72 at $0.013 a credit, product metrics and merchant attribution included
cloro, Lite plan$30, the plan floor; with the shopping flag an async request costs 7 credits, so about 10,080 of its 37,500
DataForSEO LLM Scraper, standard queueabout $0.58 of usage on the 480 ChatGPT answers it can reach, after a $50 minimum payment
Promptwatch, Essential plan$95, which covers $22.80 of usage at its published $15.83 per 1,000
AthenaHQ, Starter plan$295 plan plus about $118 of usage at $81.94 per 1,000, API sold as an add-on
LLM Pulse, Scale plan€299, the plan floor; that volume is about €44 of usage at €30.67 per 1,000
Scrunch, Agency Core$500, the plan floor, with no rate per answer published
Profound, Peec AIon quote

Our arithmetic on the rates each vendor published, checked on October 23, 2026: 60 prompts × 3 engines × 8 runs = 1,440 answers.

Two things move that total before you commit. Speed and enrichment are both billed: DataForSEO charges $1.20 per 1,000 on the standard queue, $2.40 on the priority queue and $4 in live mode, while on cloro a full product payload costs 9 credits synchronously against 5 for the plain answer. And each extra country multiplies the whole line, which on a shopping job hurts more than on a brand job, because stock, price and merchant list all change at the border.

The retail assistants, and why they are counted separately

A store that sells on marketplaces has a third surface to watch, and it is the least served by APIs.

Amazon publishes an Alexa Shopping Actions API for Alexa Skills, with four actions: add to list, add to shopping cart, buy products and recommend products. Its purpose is to let a skill you build present products to a customer, and the add-to-list and recommend-products actions are in developer preview, reachable through an Amazon business representative. Nothing in it returns what the assistant recommended to someone else, so it is not a tracking API, whatever a search for "Alexa shopping API" turns up.

The only published route to those surfaces is a vendor that has added them to its own engine list. Three have, and they name the same Amazon assistant differently: LLM Pulse lists Alexa for Shopping and Walmart Sparky, Peec AI lists Amazon Rufus, Promptwatch lists Amazon Alexa. Treat that as a reason to ask the vendor how it collects a retail assistant before you buy on the engine count: a marketplace answer sits behind a login far more often than a ChatGPT answer does, and none of the three publishes a method for it.

Mistakes that cost a quarter of data

  • Measuring prompts instead of shopping answers. At a 9% to 13% trigger rate, a 60-prompt list produces single-digit shopping answers per run. Report the trigger rate alongside every product figure, or the first month of flat numbers will be read as a visibility collapse.
  • Counting appearances as mentions. A product that appears in a carousel of eight is not a recommendation. Record the rank inside the block, and keep carousel appearances and inline products apart.
  • Letting the engine's product title be your key. Titles are composed by the engine and change without notice. Join on an identifier, keep the title as evidence.
  • Reporting visibility without the merchant. Merchant leakage changes the action completely: a product that is recommended and sold by a reseller needs a distribution conversation, not more content.
  • Mixing parent products and variants in one denominator. Colour and width variants collapse differently on each engine, and a mid-quarter change to that rule rewrites the trend line.
  • Assuming a Copilot figure and a ChatGPT figure are comparable. The two surfaces are fed by different catalogues, one from OpenAI's feed specs and one from Microsoft Merchant Center, and Copilot Checkout is limited to English-language merchants selling to United States buyers. Track them side by side, never summed.

Our view

The choice turns on whether you want the block or the conclusions:

  1. Cheapest raw blocks, with a data team on hand. DataForSEO at $1.20 per 1,000 on the standard queue, with Google Shopping identifiers in the payload to join on. Our score: 7.7 out of 10. It does not reach Copilot, so it answers half of this guide's question.
  2. Documented Copilot and ChatGPT payloads at a low rate. cloro at $4 per 1,000 inside a $30 plan, plus the shopping surcharge, with the collection method published and 500 free credits to prototype. Our score: 7.4 out of 10, held down by the absence of any computed metric.
  3. Computed metrics under $100 a month. Promptwatch at $15.83 per 1,000 inside a $95 plan, with shopping appearances filterable by prompt type and intent type, and resale allowed. Our score: 7.4 out of 10.
  4. The widest shopping surface. LLM Pulse at €30.67 per 1,000 on a €299 plan, the only option here that names retail assistants next to the ten engines we track. Our score: 8.8 out of 10.
  5. Product metrics and merchant attribution in one response. API for GEO, our editorial pick, at $13 per 1,000 answers down to $7 at its largest tier, pay as you go with no published minimum, reads over stored runs not billed, resale allowed, and 9.0 out of 10. Its price per answer is the highest of the scraping-capable options, and four of its ten engines plus its webhooks, OpenAPI spec and official SDKs were confirmed in our test of October 26, 2026 rather than documented on apiforgeo.com.

For a store, we would rather pay per answer than per plan, because a catalogue job is seasonal: the month before a product launch needs four times the volume of the month after it, and a $295 or €299 floor charges the same either way.

Frequently asked questions

Is there an official ChatGPT Shopping API for merchants?

There is an official API to push products in, and none to read visibility out. OpenAI's commerce documentation covers product feeds, an Agentic Checkout Spec and a Delegated Payment Spec, and its feed reference requires nine fields on every row: `item_id`, `title`, `description`, `url`, `brand`, `seller_name`, `image_url`, `availability` and `price`, with `is_eligible_search` defaulting to true when omitted and `is_eligible_checkout` disabled unless you set it. Nothing in those specs returns the answers ChatGPT gave other shoppers, which is why product tracking runs through third-party APIs at $1.20 to $81.94 per 1,000 collected answers.

Can I track my products in Microsoft Copilot with an API?

Only through an API that collects Copilot answers for you. Microsoft's Bing Web Search API documentation is archived on Microsoft Learn, and the search product Microsoft now sells is Grounding with Bing Search inside Azure AI Foundry, priced at $14 per 1,000 transactions, which grounds your own agent rather than reporting what Copilot told a shopper. cloro documents a `shoppingCards` array on its `POST /v1/monitor/copilot` endpoint with the product name, images, price, seller and rating, and API for GEO lists Copilot among the six engines published on its pages.

Which APIs return the products an AI assistant recommended?

Twelve of the APIs this site rates return shopping products, and they split by what arrives next to the product. API for GEO, our pick after hands-on testing by Mathias C., answers `POST /v1/shopping/products` with `product`, `externalId` for your own SKU, `visibility`, `avgPrice`, `priceRange`, `avgRating`, `topMerchants[]` and `shareOfVoice`, at $13 per 1,000 answers down to $7 at its largest tier, pay as you go with no published minimum. DataForSEO returns a `chat_gpt_products` element with title, price, currency, merchants and Google Shopping product ids at $1.20 per 1,000 and computes nothing; Profound returns product visibility and position as 0 to 1 fractions with merchant offers attached, on quote.

How often do product cards appear in ChatGPT answers?

Less often than most prompt lists assume. Profound's published analysis of about 2 million unique prompts, each run at least ten times, measured a shopping trigger on about 9% of submissions, 12.1% on open-ended prompts and 3.1% on brand-direct ones, and found that 79% of prompts never triggered shopping on any run. The same analysis puts the carry-over at about 83%: a prompt that triggered shopping on one day triggered again the next. Plan your sample on the shopping answers you will actually collect, not on the prompts you send.

How much does it cost to track 60 product prompts a month?

Sixty prompts on three engines, run eight times a month, is 1,440 collected answers. At the rates published on October 23, 2026: $18.72 with API for GEO at $0.013 a credit, with the product metrics computed; $30 on cloro's Lite plan, where the shopping flag makes an async request cost 7 credits instead of 5 and that volume uses about 10,080 of its 37,500 credits; about $0.58 of usage with DataForSEO's LLM Scraper on the 480 ChatGPT answers it can reach, after its $50 minimum payment. Platform APIs are sold by the plan: Promptwatch opens at $95 a month and LLM Pulse's Scale plan at €299.

Is there an API for Alexa for Shopping or Amazon Rufus?

Amazon publishes no visibility API for its shopping assistant. The Alexa Shopping Actions for Alexa Skills API exposes four actions, add to list, add to shopping cart, buy products and recommend products, and its purpose is to let a skill present products to a customer; two of the four are in developer preview and need an Amazon business representative. Nothing in it reports which products the assistant recommended to anyone else. Three vendors name the retail assistant in their own engine lists instead: LLM Pulse as Alexa for Shopping and Walmart Sparky, Peec AI as Amazon Rufus, Promptwatch as Amazon Alexa.