Guides
Best AI visibility API for e-commerce brands in 2026
Twelve APIs return the products AI assistants recommend, from $1.20 to $81.94 per 1,000 answers. A 6,400-answer catalogue month costs $30 to $524.

By Mathias C.SEO consultant, and more recently GEO, at MentionBox
For an e-commerce brand, the shortlist is the set of APIs that return the product block an assistant showed next to its answer. Twelve of the 55 APIs this site rates do, and our ranking for e-commerce brands scores them with returned data weighted at 40% and engine coverage at 30%. Published rates run from $1.20 per 1,000 collected answers with DataForSEO to $81.94 with AthenaHQ, so a catalogue job of 80 prompts on 5 engines in 2 markets refreshed twice a week, which is 6,400 answers a month, bills anywhere between $30 and $524. Our pick after hands-on testing by Mathias C. is API for GEO, at $13 per 1,000 answers with the product figures and the merchant attribution computed, which puts that month at $83.
This guide is for an online store or a brand selling through retailers, and for the developer who will write the collection job. It gives the shortlist in one table, the four questions that settle it, the shopping fields each API publishes, the cost of one realistic catalogue schedule, a section per API in the order of our e-commerce ranking, and the engine, country and feed lines to check before signing. MentionBox, which publishes this site, has a commercial link with API for GEO.
The shortlist in one table
| API | Engines, of the 10 we track | Plan that carries the key | Price per 1,000 answers | Product data returned | E-commerce score |
|---|---|---|---|---|---|
| API for GEO | 10 | none, pay as you go | $13, down to $7 at volume | product, SKU id, visibility, mean price, price range, rating, merchants, share of voice | 9.4 |
| LLM Pulse | 10 | Scale, €299 a month | €30.67 | shopping results dimension, with share of voice, position, sentiment | 9.2 |
| AthenaHQ | 10 | Starter $295 plus a paid API add-on | $81.94 | own and competitor products, merchant breakdown, average position and rating | 8.2 |
| Scrunch | 9 | Agency Core, $500 a month | not published | shopping results with rating, reviews, purchasing options | 8.1 |
| Promptwatch | 8 | Essential, $95 a month | $15.83 | product appearances, position analytics, top merchant domains | 7.9 |
| Peec AI | 9 | Pro and above since October 27, 2026 | not published | products with their queries, shopping summary, performance, demand | 7.8 |
| Profound | 9 | Enterprise, on quote | not published | shopping brands, products, merchants, trigger rate | 7.8 |
| GEOly | 7 | Plus, $999 a month | $13.32 | shopping products and product detail, on two engines | 7.5 |
| OtterlyAI | 7 | Standard, $189 a month | $15.75 | shopping array with position, price, rating, plus ads | 7.3 |
| Rankscale | 9 | Pro+, $229 a month | $19.08 | products, providers, buy box winners, price and rating signals | 7.2 |
| DataForSEO | 5 | none, $50 minimum payment | $1.20 | ChatGPT products with price, currency, merchants, Google Shopping ids | 6.7 |
| cloro | 6 | Lite, $30 a month | $4 | shopping cards and inline products, with offers and ratings | 6.7 |
Published by each vendor on its pricing page and in its API reference, checked on October 23, 2026. Scores come from our e-commerce ranking, which weights returned data at 40%, engine coverage at 30%, and price and integration at 15% each. For API for GEO, four engines, Gemini, Google AI Overviews, DeepSeek and Meta AI, plus the webhooks, the OpenAPI spec and the official SDKs, were confirmed in our hands-on test on October 26, 2026 and are not yet documented on apiforgeo.com.
Four questions that settle a catalogue shortlist
Most comparisons of these products compare dashboards. A store buying the API has a different set of questions, and four of them can be answered from public pages in an afternoon.
- How many answers will the catalogue buy a month? One answer is one prompt, on one engine, in one market, run once. Eighty prompts on 5 engines in 2 markets, twice a week, is 80 × 5 × 2 × 8, or 6,400 answers. Two caps then apply separately: the answer allowance, and the prompt count, which stops at 50 on Promptwatch's $95 Essential plan and 500 on GEOly's $999 Plus plan.
- Does the payload name your product or only your brand? A brand tracker counts mentions in prose. A store needs the card: the title the engine composed, the price it quoted, the rating it showed, the merchant it credited. Eight of the twelve compute brand metrics on the same answer; DataForSEO and cloro return the raw block with nothing computed.
- Does the shopping coverage match the engine list? GEOly covers seven engines and publishes its shopping surface for ChatGPT and Google AI Mode only; DataForSEO reaches five engines and neither Google AI Mode nor Copilot; cloro has never covered Claude and dropped Grok on October 15, 2026.
- Can an answer be joined to a SKU? The engine returns a title it composed, never your product id. Three vendors publish the field that closes the gap: API for GEO's
externalId, documented as your catalogue id or SKU, Promptwatch'sforeignId, the product id in the source answer engine, and DataForSEO'sproduct_idsarray of Google Shopping identifiers.
What a store buys on top of brand tracking
Adobe Analytics, over more than one trillion visits to US retail sites, put AI-driven traffic to retail up 693% year over year across November and December 2025, with AI referrals converting 31% more than non-AI sources. That traffic arrives after an assistant has already named a short set of products, and a shopping answer carries two layers that are collected separately.
- The text. Prose naming brands, with cited sources under it. Our guide on tracking a brand in ChatGPT with an API covers that layer.
- The product block. A card, a carousel or an inline product with a title, a price, a rating, images and one or more merchants. It appears only when the engine treats the question as a shopping question, and Profound's published analysis of about 2 million unique prompts measured that trigger on about 9% of submissions, 12.1% on open-ended prompts against 3.1% on brand-direct ones.
Size the job on that 9%: an 80-prompt list produces a few dozen product blocks per run, which reports monthly rather than weekly. The collection walkthrough sits in our guide on tracking products in ChatGPT Shopping and Copilot.
The shopping fields each API publishes
This is the column the dashboard comparisons skip, and it decides what your product report can contain.
| API | Published shopping endpoint or field | What arrives with the product |
|---|---|---|
| API for GEO | POST /v1/shopping/products, reading the blocks captured during normal runs | product, externalId, visibility, avgPrice, priceRange, avgRating, topMerchants[], shareOfVoice |
| LLM Pulse | GET /dimensions/shopping, "shopping results returned inside AI answers", Scale plan and above | the shopping dimension next to share of voice, position, sentiment, source categories, fan-out queries |
| AthenaHQ | shopping metrics endpoint | top_own_products and top_competitor_products with title, merchant, avg_position, avg_rating, product_url, plus merchant_breakdown |
| Scrunch | shopping_results[] on the list-responses endpoint | url, title, rating, reviews, purchasing_options with price, currency and merchant |
| Promptwatch | GET /shopping/items, plus Get Shopping Product, Position Analytics, Top Merchant Domains | title, price text, merchants, rating, review count, position in response, foreignId, price statistics |
| Peec AI | products[] in the chat endpoint, plus Products and shopping summary, performance and demand reports | product name and the queries that produced it |
| Profound | POST /v2/reports/shopping/products, plus brands, merchants and trigger-rate reports | visibility and position as fractions between 0 and 1, rating, review count, merchant offers behind include_merchants |
| GEOly | list_public_shopping_products and get_public_shopping_product_detail | a public shopping dataset; its AI Shelf is documented for ChatGPT and Google AI Mode |
| OtterlyAI | shopping[] on the AI-responses endpoint | title, image, link, tag, position, price, rating, with a separate ads[] array |
| Rankscale | Metrics API, exposing the in-app shopping and ads aggregates since August 13, 2026 | products, providers, buy box winners, price and rating signals |
| DataForSEO | chat_gpt_products in the ChatGPT LLM Scraper response | title, price, currency, merchants, rating, url, domain, images, product_ids, id_to_token_map; also chat_gpt_local_businesses and chat_gpt_ad |
| cloro | shoppingCards and inlineProducts behind include.shopping | product titles, URLs, prices, merchant offers, ratings, images; shopping cards also documented on Perplexity, Copilot, Google AI Mode, Gemini |
Field and endpoint names from each vendor's own API reference, checked on October 23, 2026.
Two readings of that table are worth having before a sales call. The split between your products and the competition is published by two vendors, AthenaHQ in top_competitor_products and API for GEO through a per-product share of voice; everywhere else that comparison is an aggregation you write. And a position metric needs its scale checked, since Profound documents its shopping position as a fraction between 0 and 1, which reads as a collapse in a dashboard that renders it as a percentage.
What a real catalogue schedule costs each month
Take 80 prompts, 40 category and 40 product, on 5 engines in 2 markets, refreshed twice a week: 6,400 answers a month.
| API | Monthly bill at 6,400 answers | Rate per 1,000 | What the plan covers |
|---|---|---|---|
| DataForSEO | $4.61 of usage, on the 3,840 answers it reaches | $1.20 | no plan; $50 minimum payment; 3 of these 5 engines |
| cloro | $30, the plan floor | $4 | Lite, $30, 37,500 credits; this job uses about 32,000 of them asynchronously |
| API for GEO | $83 | $13 | no plan, pay as you go; 6,400 sits inside the first 50,000-credit tier |
| Promptwatch | $101 of usage, above the Essential plan | $15.83 | Essential, $95, 50 prompts and 6,000 responses, no top-up |
| OtterlyAI | $189, the plan floor, for $101 of usage | $15.75 | Standard, $189, about 12,000 answers and 2,000 API reads |
| Rankscale | $229, the plan floor, for $122 of usage | $19.08 | Pro+, $229, about 12,000 answers |
| LLM Pulse | €299, the plan floor, for €196 of usage | €30.67 | Scale weekly, €299, about 9,750 answers |
| Scrunch | $500, the plan floor | not published | Agency Core, $500, no volume published |
| AthenaHQ | $524 of usage, plus the plan and its API add-on | $81.94 | Starter, $295, about 3,600 credits |
| GEOly | $999, the plan floor, for $85 of usage | $13.32 | Plus, $999, 500 prompts and 75,000 answers |
| Peec AI, Profound | not computable | not published | on quote |
Our arithmetic on each vendor's published rate and allowance, prices checked on October 23, 2026: 80 prompts × 5 engines × 2 markets × 8 runs = 6,400 answers. The right-hand column exists because the plan floor is higher than the usage on seven of these routes.
Below about 10,000 answers a month, the plan floor sets the bill more often than the rate does, so the published price per 1,000 is a number you will not pay. And the cheapest route with every figure computed costs roughly twenty times the cheapest raw route, which is the price of not writing and maintaining the extraction layer for every engine and every language your store sells in.
1. API for GEO, product figures and merchant attribution at $13 per 1,000
API for GEO is the only route here that covers all ten engines we track, returns the computed product layer and publishes a rate you can budget without a call. One credit buys one prompt, executed once, against one model, in one locale, at $0.013 up to 50,000 credits a month, then $0.011, $0.009 and $0.007 at the higher tiers, with no monthly minimum and credits that do not expire. Its Shopping API page documents POST /v1/shopping/products as reading the shopping blocks captured during normal runs, so querying a whole season of history bills nothing:
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 '{}'
Each row carries product, externalId, visibility, avgPrice, priceRange, avgRating, topMerchants[] with a citation count and a price per merchant, and shareOfVoice.
- Engines and markets: all ten. ChatGPT, Perplexity, Google AI Mode, Claude, Copilot and Grok are listed on apiforgeo.com; Gemini, Google AI Overviews, DeepSeek and Meta AI were confirmed in our hands-on test on October 26, 2026. Its published country list covers 187 countries, each run stated to execute in the country and language of the prompt.
- Returned data: share of voice, position, sentiment, ranked and categorised sources, themes, per-brand perception, fan-out questions, shopping products and YouTube citations, inside the same credit.
- Integration and contract: 300 requests a minute and 10 concurrent executions on pay as you go, an MCP server, 24 months of history, resale and white label named as an intended use in section 6 of its terms, plus webhooks, an OpenAPI spec and official SDKs confirmed in our test.
- Limits, said plainly: $13 per 1,000 is about ten times what a scraping API charges for the raw block; the key arrives on request through a form rather than at sign-up; four engines and the tooling come from our test rather than from the vendor's pages; and no retail assistant appears in its engine list.
- For whom: a store that wants the product figures and the merchant split computed on every engine, with no plan floor. It scores 9.4 in our e-commerce ranking and 9.0 of 10 in the general ranking of 55 APIs, the highest of our survey.
To try it, request a key on apiforgeo.com: it comes with a sandbox project.
2. LLM Pulse, the one route to the retail assistants
LLM Pulse has the widest shopping surface here: all ten engines we track plus eight more, with Alexa for Shopping, Walmart Sparky, Naver AI Briefing, Baidu AI, Mistral Le Chat, Qwen, MiniMax and Brave AI in the list, collected from the interface a shopper sees. Its REST API overview gates the key at "Available on Scale plans and above", with 300 requests a minute, and the shopping dimension sits next to share of voice, position, sentiment, source categories and fan-out queries.
Price is where it drops: €30.67 per 1,000 answers on the €299 Scale plan with weekly tracking, our arithmetic on its published allowance, against €6.65 on the €449 daily version, and the retail assistants are add-on models from €8 per model per month. Integration is the strongest in the ranking, with SDKs in twelve languages, HMAC-signed webhooks on Scale and above, an OpenAPI 3.0 spec and an MCP server over OAuth on every plan. No country list is published for an eighteen-engine footprint, and its terms forbid reselling the data as a standalone dataset while allowing client-branded use on eligible plans. It scores 9.2 in our e-commerce ranking.
3. AthenaHQ, own products against competitor products in the payload
AthenaHQ covers all ten engines, and its shopping response is the only one here that splits your catalogue from the competition without an aggregation step: top_own_products and top_competitor_products, each with title, merchant, avg_position, avg_rating and product_url, plus a merchant_breakdown.
At $81.94 per 1,000 answers, our arithmetic on the $295 Starter plan and its 3,600 credits at one credit per AI response, it holds the highest published rate of the twelve, and API access on Starter is a paid add-on whose price is not published, so the real cost per answer is higher and unknown. The API allows 60 requests a minute per organisation, with synchronous reads plus a response-streaming job you start and poll, an OpenAPI spec, an MCP server with read and write tools, and dashboard embedding through iframe tokens; no webhooks and no SDK are published. Neither a country list nor a collection method is published, so a product figure for a German SKU arrives without a provenance you can describe to your own team. It scores 8.2 in our e-commerce ranking.
4. Scrunch, purchasing options on nine engines behind a $500 plan
Scrunch returns shopping_results[] with url, title, rating, reviews and purchasing_options carrying price, currency and merchant, on nine of our ten engines with DeepSeek the only gap, which earns it a data score of 10 out of 10. Its authentication page states that "API key creation is available on the Agency and Enterprise plans", and Agency Core is published at $500 a month with no volume attached, so no rate per answer can be computed.
It is one of the few routes here that will run a prompt on demand, through POST /{brand_id}/collections, alongside an OpenAPI spec and an MCP server on any plan capped at 1,000 records per request. Webhooks are marked "not generally available", no SDK is published, the collection method is not published, geography is stated at country level without a count, and the 7-day trial needs no credit card. It scores 8.1 in our e-commerce ranking.
5. Promptwatch, product appearances and merchant domains from $95
Promptwatch is the cheapest plan here with computed metrics, and its API comes with every paid plan rather than the top one. Its shopping group publishes four endpoints, a paginated list of product appearances, a single product lookup, position analytics and top merchant domains, and each appearance carries title, price text, merchants, rating, review count, position in the response, price statistics and foreignId, the product id in the source engine.
The rate is $15.83 per 1,000, from the $95 Essential plan and its 50 prompts and 6,000 responses, which cannot be topped up: more volume means a higher tier. It covers 8 of our 10 engines, Grok and Meta AI absent, with Amazon Alexa in its own list, and names its collection method per engine. Share of voice is the missing headline figure, so the category split stays your aggregation; there are no webhooks and no SDK, only a Postman collection; and rate limits are set per organisation and shared between REST, MCP and every key. White-label reports are allowed. It scores 7.9 in our e-commerce ranking.
6. Peec AI, product demand reports and an Amazon channel
Peec AI computes the full metric set over nine of our ten engines, Meta AI missing, and goes wider than our grid with Amazon Rufus, Naver AI, Qwen, Mistral and the OpenAI Search API. On the product side it returns products[] with the queries that produced each one, plus shopping summary, performance and demand reports. Its model channel concept is worth copying: a channel is a stable identifier for one AI surface that survives a model upgrade, and its Amazon channel ships with an unsupported_country_codes list.
Access moved recently. Its API introduction limited the Customer API to Enterprise customers on October 23, 2026, and its changelog records under v0.39.0, dated October 27, 2026, that availability moved to Pro, Advanced, Enterprise and agency plans. No price per answer is published on any tier. It allows 200 requests a minute per project and batch writes up to 1,000 items, with an OpenAPI spec and an MCP server on every paid plan, no SDK and no webhooks documented. No endpoint executes a prompt on demand, and resale is forbidden at 1.3 and 1.6 of its terms. It scores 7.8 in our e-commerce ranking.
7. Profound, the deepest shopping report set, on quote
Profound publishes the most shopping endpoints of any vendor here, brands, products, merchants and a trigger-rate report, over nine engines captured in the browser across 150+ regions and 30+ languages. Its shopping products endpoint takes a category id, a date window and a metric list, attaches { name, price } merchant offers behind include_merchants, and documents its position metrics as fractions between 0 and 1. It is also the only vendor in this ranking with a published study on when shopping fires at all.
The API sits on the Enterprise plan at a price on quote, and the free trial of 50 prompts a day for 7 days on three engines excludes API access, so the endpoint cannot be tested on your own SKUs before negotiating. Delivery is synchronous only at 600 requests an hour per key by default, and the Master Subscription Agreement bars distributing, selling or sublicensing the services. Python and TypeScript SDKs, an OpenAPI spec and a hosted MCP server come with it. Our guide to Profound alternatives with a self-serve API covers the field at published prices. It scores 7.8 in our e-commerce ranking.
8. GEOly, a complete payload with a $999 floor
GEOly leaves nothing to build in its response, share of voice, position, sentiment, source categories, fan-out queries and shopping products, at $13.32 per 1,000 answers, the second cheapest rate among the routes that compute metrics. The arithmetic holds it back: the $999 Plus plan is the first tier where the pricing page shows the API, it covers 500 prompts and 75,000 answers a month, and a 6,400-answer catalogue pays $999 for $85 of usage. Its shopping surface, published as AI Shelf, covers ChatGPT and Google AI Mode, and its seven engines leave out Claude, DeepSeek and Meta AI. Delivery is synchronous only, the collection method is not published, and the trial lasts 3 days. Above 75,000 answers a month the plan turns cheap. It scores 7.5 in our e-commerce ranking.
9. OtterlyAI, shopping and ads arrays at $189 a month
OtterlyAI returns a shopping[] array with title, image, link, tag, position, price and rating on its AI-responses endpoint, with a separate ads[] array, which makes it one of the few routes that keeps a paid placement apart from an organic recommendation. Its $189 Standard plan buys about 12,000 answers at $15.75 per 1,000 plus 2,000 API requests a month, over 7 engines with Grok, DeepSeek and Meta AI absent. Source categories are the single missing field, so you see where a product sits in an answer without knowing what kind of site the engine cited for it. There is no SDK, no webhook and no native white-label option, and countries are advertised as unlimited without a number, the real list readable per engine through its engines endpoint. It scores 7.3 in our e-commerce ranking.
10. Rankscale, buy box signals on nine engines
Rankscale exposes the same shopping and ads aggregates through its Metrics API as its in-app shopping tab, a change its changelog dates to August 13, 2026, with products, providers, buy box winners, and price and rating signals, over nine engines and 240+ countries at $19.08 per 1,000 answers on the $229 Pro+ plan. The buy box angle is the distinctive one for a marketplace seller. Integration drops it: delivery is batch only, so there is no live lookup on a single SKU during a merchandising review, the REST documentation sits behind a login, and no webhook or SDK is published. Share of voice and source categories are missing, resale is allowed, and white-label dashboards come from the Growth plan up. It scores 7.2 in our e-commerce ranking.
11. DataForSEO, the cheapest raw product block
DataForSEO is the cheapest route by a factor of more than three, at $1.20 per 1,000 answers on the standard queue, $2.40 on the priority queue and $4 in live mode, with a $50 minimum payment as the only floor. Its chat_gpt_products element returns title, price, currency, merchants, a rating object, url, domain, images and the Google Shopping identifiers, the cleanest join key in this comparison, alongside chat_gpt_local_businesses and chat_gpt_ad items worth filtering out before counting.
Nothing is computed: no share of voice, no position, no sentiment. Coverage is the narrowest here, five engines without Google AI Mode or Copilot, and its LLM Mentions data for ChatGPT is published for the United States and English only. Integration is the strongest of the group, with webhooks through postback and pingback URLs, up to 100 tasks per call and SDKs in Python, C#, TypeScript and Java. Our guide to DataForSEO alternatives with share of voice and sentiment covers the step up, and API for GEO vs DataForSEO compares the two payloads. It scores 6.7 in our e-commerce ranking.
12. cloro, documented browser sessions and state-level targeting
cloro documents its collection method outright, real browser sessions against each engine's web interface, and it is the only route here with US state targeting, available on ChatGPT, Copilot, Perplexity and Gemini for 2 extra credits. Its shoppingCards and inlineProducts fields sit behind an include.shopping flag on ChatGPT, with shopping cards also documented on Perplexity, Copilot, Google AI Mode and Gemini.
The $30 Lite plan buys 37,500 credits at $0.80 per 1,000 credits and the free plan gives 500 credits a month, enough to prototype. A ChatGPT answer costs 5 credits asynchronously, 7 with the shopping flag, plus 2 more for a synchronous call, and credits do not roll over, which charges a seasonal catalogue for the months it skips. Six engines, no Claude and no Grok since October 15, 2026, no computed metrics, and resale of the services is forbidden at 2.3(a) of its terms. Batches hold 500 tasks and webhooks are HMAC-signed. It scores 6.7 in our e-commerce ranking, and API for GEO vs cloro sets the two side by side.
Which engines a store should pay for
Engine coverage is the second heaviest criterion in our ranking, and for a catalogue the question is narrower: which engines show product cards, and which of them the vendor actually collects. ChatGPT, Perplexity, Gemini and Google AI Overviews are covered by all twelve, Google AI Mode and Copilot by eleven, DataForSEO the exception on both. A vendor can also cover an engine for text and not for products, as GEOly's AI Shelf on two engines shows.
Three published engine lists go inside a store: LLM Pulse names Alexa for Shopping and Walmart Sparky, Peec AI names Amazon Rufus, Promptwatch names Amazon Alexa. No Amazon API returns what that assistant answered a shopper, so each of the three collects from the live product, and our page on what brands can measure on Alexa for Shopping and Rufus sets out what Amazon publishes.
Each engine you add has a price, since the engine is part of the billing unit. On 80 prompts in 2 markets refreshed twice a week, one more engine is 1,280 answers a month: $16.64 with API for GEO, $24.42 with Rankscale, $1.54 with DataForSEO.
Tracking the same products in several countries
Five of the twelve publish a geographic footprint. cloro states 250 countries with state-level targeting on four engines, Rankscale 240+, API for GEO 187 with each run stated to execute in the country and language of the prompt, Profound 150+ regions and 30+ languages, Peec AI dedicated infrastructure in 80+ countries. LLM Pulse, AthenaHQ, Scrunch, Promptwatch, GEOly, OtterlyAI and DataForSEO publish no country count, even where the engine list is the widest.
Markets hurt a shopping job more than a brand job, because stock, price, currency and merchant list all change at the border. A product recommended in France by a retailer that does not ship to Belgium is a different problem from a product that is simply absent there. Three rules follow.
- Price the market before adding it. One extra market on the schedule above is 3,200 answers: $41.60 with API for GEO, $50.40 with OtterlyAI, $262 with AthenaHQ, about $12.80 of cloro credits.
- Check how the answer is collected there. API for GEO states native execution in the country and language of the prompt, while LLM Pulse, Profound and cloro read the real interface. Promptwatch, Peec AI, Rankscale, OtterlyAI and DataForSEO mix the official model API, the official API with web search and the real interface, and an official model API has no country of its own, so a localised figure from that route describes the prompt rather than the market.
- Keep the currency in the row. Two payloads above return a currency field explicitly, Scrunch in
purchasing_optionsand DataForSEO inchat_gpt_products.
The official merchant APIs, and what they leave out
Every engine with a shopping surface publishes a way in, and none publishes a way to read back what it showed.
| Surface | Official merchant interface | Official API reporting what the assistant showed |
|---|---|---|
| ChatGPT Shopping | product feed spec with nine required fields per row, plus the Agentic Checkout and Delegated Payment specs | none published |
| Microsoft Copilot | Microsoft Merchant Center feed plus the Universal Commerce Protocol, with Copilot Checkout | none published |
| Google AI Mode and Gemini | Universal Commerce Protocol, described as enabling "transactions directly on Google's AI surfaces (Search, Gemini)" from a Merchant Center account configured with shipping, returns and a product feed | none published |
| Alexa for Shopping | Alexa Shopping Actions for Alexa Skills, for a skill to present products | none published |
From each provider's own developer documentation, read on November 7, 2026. Google's UCP guide states that "UCP is an evolving standard, not all features defined in the spec are available on Google's surfaces".
The feed side still sets the ceiling, because a product missing from the feed cannot appear in a card and will not appear in any API response either. Our guide on tracking products in ChatGPT Shopping and Copilot goes through those requirements field by field. The read side is where the twelve APIs above exist, and it is why their prices compare per 1,000 collected answers rather than per seat.
Five mistakes that cost a store a quarter of product data
- Comparing a plan floor to a rate per answer. At 6,400 answers a month, seven of the twelve bill their plan floor rather than their rate, so $85 of GEOly usage costs $999 and $196 of LLM Pulse usage costs €299.
- Buying on engine count instead of shopping coverage. A seven-engine vendor whose shopping surface is published for two of them reports products on two engines. Ask for the shopping endpoint per engine, in writing.
- Sizing the job on prompts. At a shopping trigger rate around 9%, an 80-prompt list produces a few dozen product blocks per run. Report the trigger rate next to every product figure, or the first flat month will be read as a collapse.
- Letting the engine's product title be the join key. Titles are composed by the engine and change without notice. Join on an identifier,
externalId,foreignIdor a Google Shopping id, and keep the title next to it as evidence. - Reporting product visibility without the merchant. A SKU recommended in 40% of shopping answers whose citations go mostly to a marketplace is a distribution problem, and five of the twelve publish the merchant field that shows it.
Our view
Four defensible buys, depending on what binds hardest.
- Product figures and merchant attribution in one response. API for GEO is our pick: all ten engines, the product with your own SKU id, mean price, price range, rating, merchants and share of voice, at $13 per 1,000 answers down to $7, no plan floor, reads over stored runs unmetered, resale allowed in writing, and 9.0 of 10 in our general ranking. The trade-off to state plainly is the price per answer, about ten times a scraping API's, plus the documentation gap on four engines and the tooling, which come from our test of October 26, 2026.
- The retail assistants in the same series. LLM Pulse at €30.67 per 1,000 weekly or €6.65 daily, the only route in our ranking that names Alexa for Shopping and Walmart Sparky, on a €299 floor.
- The cheapest raw blocks, with a data team on hand. DataForSEO at $1.20 per 1,000 with Google Shopping identifiers to join on, or cloro at $4 per 1,000 inside a $30 plan, with its collection method published. Both compute nothing, so budget the engineering next to the savings.
- Computed metrics under $200 a month. Promptwatch at $15.83 per 1,000 inside a $95 plan for product appearances and merchant domains, or OtterlyAI at $189 for about 12,000 answers with shopping and ads separated.
For a store, in our view the billing model counts for as much as the payload, because a catalogue job is seasonal: the six weeks before a launch or a holiday peak need four times the volume of February, and a $295 or €299 floor charges the same in both months. That argues for paying per answer wherever the payload allows it.
Frequently asked questions
Which AI visibility API is best for an e-commerce brand?
It depends on whether you want the product block raw or the product figures computed. Our pick after hands-on testing by Mathias C. is API for GEO, which returns the product, your own SKU id, the mean price quoted, the rating, the merchants credited and a share of voice in one response at $13 per 1,000 answers, down to $7 at its largest tier with no monthly minimum. For a store with a data team, cloro at $4 per 1,000 and DataForSEO at $1.20 per 1,000 hand back the shopping cards and compute nothing, while LLM Pulse is the only route in our ranking that names Alexa for Shopping and Walmart Sparky among its engines, at €30.67 per 1,000 on a €299 plan.
How much does it cost to track 80 product prompts a month?
Eighty prompts on five engines in two markets, refreshed twice a week, is 6,400 collected answers a month. On the rates published on October 23, 2026 that month costs $30 inside cloro's Lite plan, $83 with API for GEO at $0.013 a credit, $101 with OtterlyAI inside its $189 Standard plan, $101 of usage with Promptwatch whose $95 Essential plan stops at 50 prompts and 6,000 responses, $122 with Rankscale inside its $229 Pro+ plan, $85 of usage with GEOly behind a $999 plan and $524 with AthenaHQ. Peec AI, Profound and Scrunch publish no rate per answer.
Do any of these APIs cover Amazon's shopping assistant?
Three name it, under three labels, and none holds an Amazon endpoint: LLM Pulse lists Alexa for Shopping and Walmart Sparky, Peec AI documents an `amazon-rufus-scraper` channel with an `unsupported_country_codes` list, and Promptwatch lists Amazon Alexa. Amazon publishes no API that returns what the assistant answered a shopper, so every organic figure on that surface comes from a vendor collecting it from the live product, starting at $15.83 per 1,000 answers inside Promptwatch's $95 Essential plan.
Can one API track the same products in several countries?
Five of the twelve publish a geographic footprint: cloro states 250 countries with US state targeting on four engines, Rankscale 240+, API for GEO 187, Profound 150+ regions and 30+ languages, and Peec AI dedicated infrastructure in 80+ countries. The other seven publish no country count, including vendors with the widest engine lists. Every market multiplies the bill, since the country is part of the billing unit: one extra market on an 80-prompt, 5-engine, twice-weekly job adds 3,200 answers, which is about $12.80 of cloro credits, $41.60 with API for GEO or $262 with AthenaHQ.
Which API tells you which merchant got credited for your product?
Five publish a merchant field in their reference. API for GEO returns `topMerchants[]` with a citation count and a price per merchant, AthenaHQ a `merchant_breakdown` next to `top_own_products` and `top_competitor_products`, Scrunch `purchasing_options` with price, currency and merchant, Profound `{ name, price }` offers behind an `include_merchants` flag, and Promptwatch a Shopping Top Merchant Domains endpoint. That field is the one an online store cannot reconstruct from its own analytics, because it shows the assistant sending a buyer to a reseller instead.
Is there an official API from OpenAI or Google that reports AI product visibility?
No. The official interfaces run the other way: OpenAI's product feed spec requires nine fields per row to get a catalogue into ChatGPT, Microsoft pairs Merchant Center feeds with its Universal Commerce Protocol, and Google's Universal Commerce Protocol enables transactions on Search and Gemini from an existing Merchant Center feed. None of the three returns the answers an assistant gave other shoppers, which is why visibility collection is bought from third parties at $1.20 to $81.94 per 1,000 answers.