Etsy adds an AI-image check to the listing workflow
Etsy has changed how sellers document handmade and designed goods. Most listings must now include at least one unedited photo of the actual product in a real setting. Functional guides such as patterns and recipes must also show a real result made from the instructions; the guide itself cannot be AI-generated.
The listing form now warns a seller when Etsy detects that an uploaded image may use AI. That makes image review part of catalog publishing, before a buyer sees the page. Bot-generated listings are prohibited, while listings that fail the new image rules may expose the seller to a refund claim if the delivered item differs materially from the page. Etsy says it plans additional buyer context and more proactive enforcement.
Etsy has changed how sellers document handmade and designed goods. Most listings must now include at least one unedited photo of the actual product in a real setting. Functional guides such as patterns and recipes must also show a real result made from the instructions; the guide itself cannot be AI-generated.
The listing form now warns a seller when Etsy detects that an uploaded image may use AI. That makes image review part of catalog publishing, before a buyer sees the page. Bot-generated listings are prohibited, while listings that fail the new image rules may expose the seller to a refund claim if the delivered item differs materially from the page. Etsy says it plans additional buyer context and more proactive enforcement.
🤓1
Walmart gives outside shopping agents a controlled account
Walmart plans to add Agent Profiles to customer accounts within a month. A shopper will link an external AI assistant to a separate profile, then choose whether every purchase needs approval or the agent may buy within a preset spending limit.
For developers, the retailer is preparing a beta Agent Gateway built on Model Context Protocol. It gives approved agents a route to search Walmart and Sam’s Club inventory and place orders without exposing the customer’s payment-card details or full purchase history. Orders made by an agent remain separate from the shopper’s own activity, and the profile can be deleted.
The commerce workflow is therefore permission first, shopping second: connect one agent, define its authority, review its orders, and revoke access without changing the main customer account.
Walmart plans to add Agent Profiles to customer accounts within a month. A shopper will link an external AI assistant to a separate profile, then choose whether every purchase needs approval or the agent may buy within a preset spending limit.
For developers, the retailer is preparing a beta Agent Gateway built on Model Context Protocol. It gives approved agents a route to search Walmart and Sam’s Club inventory and place orders without exposing the customer’s payment-card details or full purchase history. Orders made by an agent remain separate from the shopper’s own activity, and the profile can be deleted.
The commerce workflow is therefore permission first, shopping second: connect one agent, define its authority, review its orders, and revoke access without changing the main customer account.
A practical test for Shopify's semantic search
Shopify's semantic search can match intent even when a shopper's words do not appear in a product title. It uses product descriptions, categories and image signals such as visible text and colour. That makes catalog quality part of search tuning: vague descriptions or misleading imagery can send a query toward the wrong items.
A useful test is to collect ten natural-language queries that currently return weak results, record the baseline, then improve only the relevant product data and compare the result set. Product boosts can still guide important queries, but they do not apply when a query uses search syntax.
The feature is limited to Grow, Advanced and Plus stores with fewer than 200,000 products. It does not cover predictive search and is not supported for the Japanese locale.
Shopify's semantic search can match intent even when a shopper's words do not appear in a product title. It uses product descriptions, categories and image signals such as visible text and colour. That makes catalog quality part of search tuning: vague descriptions or misleading imagery can send a query toward the wrong items.
A useful test is to collect ten natural-language queries that currently return weak results, record the baseline, then improve only the relevant product data and compare the result set. Product boosts can still guide important queries, but they do not apply when a query uses search syntax.
The feature is limited to Grow, Advanced and Plus stores with fewer than 200,000 products. It does not cover predictive search and is not supported for the Japanese locale.
❤1
Btab turns a store idea into a routed catalog
Btab has launched Btab.app, a commerce platform that combines AI-assisted store creation with a supplier catalog and order routing. A merchant describes the kind of shop they want, adjusts the resulting brand and layout, then chooses products from Btab’s catalog, their own range, or both.
The workflow extends beyond page generation. When a customer orders a catalog item, Btab sends the order to the supplier; an existing retailer can also connect its own website to the catalog through an API. That links storefront setup, product supply and order handoff in one system instead of leaving the generated site as a visual prototype.
The launch is live, but the company has not reported conversion, fulfillment-time or merchant-retention results. Its claims about scale and opportunity remain vendor statements.
Btab has launched Btab.app, a commerce platform that combines AI-assisted store creation with a supplier catalog and order routing. A merchant describes the kind of shop they want, adjusts the resulting brand and layout, then chooses products from Btab’s catalog, their own range, or both.
The workflow extends beyond page generation. When a customer orders a catalog item, Btab sends the order to the supplier; an existing retailer can also connect its own website to the catalog through an API. That links storefront setup, product supply and order handoff in one system instead of leaving the generated site as a visual prototype.
The launch is live, but the company has not reported conversion, fulfillment-time or merchant-retention results. Its claims about scale and opportunity remain vendor statements.
❤1
Amazon Business puts procurement value in one view
Amazon Business has launched Value Intelligence for US buyers. The reporting tool combines an organisation’s own purchasing data with Amazon’s benchmarks to quantify savings, operational efficiency and policy compliance. Procurement teams can use the output to show finance leaders where time was saved and off-policy spend was prevented, rather than reporting price reductions alone.
The same release expands Business Assistant from general account guidance to account-specific work: tracking orders, explaining approval status, suggesting savings, allocating payments and analysing spend. Both tools are live for US Amazon Business customers. Amazon has not published independent productivity or savings results, so the change is a workflow rollout, not a proven performance gain.
Amazon Business has launched Value Intelligence for US buyers. The reporting tool combines an organisation’s own purchasing data with Amazon’s benchmarks to quantify savings, operational efficiency and policy compliance. Procurement teams can use the output to show finance leaders where time was saved and off-policy spend was prevented, rather than reporting price reductions alone.
The same release expands Business Assistant from general account guidance to account-specific work: tracking orders, explaining approval status, suggesting savings, allocating payments and analysing spend. Both tools are live for US Amazon Business customers. Amazon has not published independent productivity or savings results, so the change is a workflow rollout, not a proven performance gain.
An AI order tool that pauses before changing the order
Salesforce documents a beta workflow in which an AI client can handle common Order Management tasks through conversation. For a cancellation, the operator provides the order goal; the system looks up the items, checks which ones are eligible, submits the permitted change and reports the result.
The useful design choice is the stop before execution. The agent can read and update order, inventory, delivery and routing data, but Salesforce says it confirms each change before acting. It should also explain constraints and return blockers with next steps. This makes a practical pilot measurable: compare handling time and error rates, while keeping the confirmation step and an audit trail. The Headless 360 MCP service is still beta, so it belongs in a controlled test rather than an unattended rollout.
Salesforce documents a beta workflow in which an AI client can handle common Order Management tasks through conversation. For a cancellation, the operator provides the order goal; the system looks up the items, checks which ones are eligible, submits the permitted change and reports the result.
The useful design choice is the stop before execution. The agent can read and update order, inventory, delivery and routing data, but Salesforce says it confirms each change before acting. It should also explain constraints and return blockers with next steps. This makes a practical pilot measurable: compare handling time and error rates, while keeping the confirmation step and an audit trail. The Headless 360 MCP service is still beta, so it belongs in a controlled test rather than an unattended rollout.
❤1
Amazon turns a few campaign inputs into a full ad plan
Amazon Ads Agent now supports Full-Funnel Campaigns across sponsored ads, display, video and streaming TV. An advertiser supplies a small set of inputs, including budget and duration; the system then plans, launches and continuously optimizes the campaign.
The workflow also creates display images and TV-style video from existing product pages, so the catalog page becomes an input to campaign production as well as a sales surface. Amazon describes what the agent can execute, but the announcement does not provide a measured sales lift or a comparison with campaigns built manually. Budget and duration remain advertiser inputs; the published summary does not say that the agent can raise them on its own.
Amazon Ads Agent now supports Full-Funnel Campaigns across sponsored ads, display, video and streaming TV. An advertiser supplies a small set of inputs, including budget and duration; the system then plans, launches and continuously optimizes the campaign.
The workflow also creates display images and TV-style video from existing product pages, so the catalog page becomes an input to campaign production as well as a sales surface. Amazon describes what the agent can execute, but the announcement does not provide a measured sales lift or a comparison with campaigns built manually. Budget and duration remain advertiser inputs; the published summary does not say that the agent can raise them on its own.
Akeneo puts an approval gate between supplier data and the catalog
Akeneo’s Autumn 2026 release adds AI-assisted product model suggestions and a supplier workflow for collecting and updating product information. A supplier can contribute missing specifications, while the retailer’s team reviews those changes before accepting them into the trusted catalog.
The release also adds Readiness Policies. Instead of treating a filled field as proof that an item is ready to sell, a team can define rules for a channel, market or business objective, then see which products fail those rules. Compatibility rules can maintain links between products, spare parts and accessories as the catalog changes.
These functions are available from October 28, though availability varies by Product Cloud edition. Akeneo reports capabilities, not measured reductions in onboarding time or catalog errors.
Akeneo’s Autumn 2026 release adds AI-assisted product model suggestions and a supplier workflow for collecting and updating product information. A supplier can contribute missing specifications, while the retailer’s team reviews those changes before accepting them into the trusted catalog.
The release also adds Readiness Policies. Instead of treating a filled field as proof that an item is ready to sell, a team can define rules for a channel, market or business objective, then see which products fail those rules. Compatibility rules can maintain links between products, spare parts and accessories as the catalog changes.
These functions are available from October 28, though availability varies by Product Cloud edition. Akeneo reports capabilities, not measured reductions in onboarding time or catalog errors.
Virid puts a confirmation gate in front of AI store edits
Virid has released a connector that lets marketAgility users work with store data through compatible AI assistants. Staff can search products, orders and customers, then create or update catalog records, inventory and promotion codes from a natural-language request.
The permission map is the useful detail. Orders, carts and per-site inventory remain read-only, while every change or deletion requires confirmation. Access is limited to active marketAgility stores with API access and uses the employee's existing store account. The launch describes availability and controls, but provides no measured time or revenue result.
Virid has released a connector that lets marketAgility users work with store data through compatible AI assistants. Staff can search products, orders and customers, then create or update catalog records, inventory and promotion codes from a natural-language request.
The permission map is the useful detail. Orders, carts and per-site inventory remain read-only, while every change or deletion requires confirmation. Access is limited to active marketAgility stores with API access and uses the employee's existing store account. The launch describes availability and controls, but provides no measured time or revenue result.
NIQ and Google add store-level context to AI shopping
NIQ and Google have announced a data partnership aimed at a common weakness in shopping agents: knowing whether the suggested product is actually available nearby.
The plan is to combine Google’s Shopping Graph with NIQ’s product records and store intelligence. That could let an agent match a request to a specific item, check local availability and connect the recommendation with later purchase data. A separate integration will bring Google’s Meridian measurement framework into NIQ Cadence, so brands can compare marketing activity with real sales.
This is a collaboration, not a finished consumer rollout. The companies have not published launch dates for the shopping features and say specific milestones will be announced later.
NIQ and Google have announced a data partnership aimed at a common weakness in shopping agents: knowing whether the suggested product is actually available nearby.
The plan is to combine Google’s Shopping Graph with NIQ’s product records and store intelligence. That could let an agent match a request to a specific item, check local availability and connect the recommendation with later purchase data. A separate integration will bring Google’s Meridian measurement framework into NIQ Cadence, so brands can compare marketing activity with real sales.
This is a collaboration, not a finished consumer rollout. The companies have not published launch dates for the shopping features and say specific milestones will be announced later.
❤1
Rep AI lets shoppers change a variant before fulfilment
Rep AI has added an order-change workflow for Shopify support. A shopper can ask in chat to swap the size or colour of an item that has not yet been fulfilled. The agent checks the order state and can complete a same-price variant change without sending the customer through a manual ticket.
The boundary is explicit: a more expensive replacement, a discounted order or a subscription is handed to the merchant’s team. Rep also added a daily review flow that proposes new knowledge from human agents’ answers, but requires approval before using it with shoppers.
The October update describes released functions, not measured changes in resolution time or error rates.
Rep AI has added an order-change workflow for Shopify support. A shopper can ask in chat to swap the size or colour of an item that has not yet been fulfilled. The agent checks the order state and can complete a same-price variant change without sending the customer through a manual ticket.
The boundary is explicit: a more expensive replacement, a discounted order or a subscription is handed to the merchant’s team. Rep also added a daily review flow that proposes new knowledge from human agents’ answers, but requires approval before using it with shoppers.
The October update describes released functions, not measured changes in resolution time or error rates.
Scene turns generated images into a shopping feed
Dalet Labs has launched early access to Scene, an invite-only iOS app built for shoppers who do not begin with a precise query. The app generates lifelike scenes containing real products across categories such as fashion, home and sport. As a shopper reacts to the feed, later scenes adapt to those signals.
The commerce workflow stays inside the app: tap an item shown in a generated scene, add products from different brands to one basket, then use a unified checkout. That makes the generated image the discovery layer rather than the product itself.
Scene is currently an early-access product, not a proven sales channel. Dalet Labs has not published conversion or repeat-use results, and the company site is still asking users to request an invitation.
Dalet Labs has launched early access to Scene, an invite-only iOS app built for shoppers who do not begin with a precise query. The app generates lifelike scenes containing real products across categories such as fashion, home and sport. As a shopper reacts to the feed, later scenes adapt to those signals.
The commerce workflow stays inside the app: tap an item shown in a generated scene, add products from different brands to one basket, then use a unified checkout. That makes the generated image the discovery layer rather than the product itself.
Scene is currently an early-access product, not a proven sales channel. Dalet Labs has not published conversion or repeat-use results, and the company site is still asking users to request an invitation.
Filtered recommendations need a fallback check
Amazon Personalize can return real-time product recommendations after a store applies catalog filters. There is an operational catch: when a filter removes too many candidates, the service may fill the requested list with popular items that satisfy the filter but have no relevance score for that shopper.
That changes the QA workflow. A commerce team should log each item’s recommendation reason, separate “popular item” placeholders from scored results, and test whether fallback-heavy lists still fit the page context. The check matters most for narrow availability, price or category filters, where a full-looking carousel can hide a thin personalized result set. This is documented behavior, not a measured sales uplift.
Amazon Personalize can return real-time product recommendations after a store applies catalog filters. There is an operational catch: when a filter removes too many candidates, the service may fill the requested list with popular items that satisfy the filter but have no relevance score for that shopper.
That changes the QA workflow. A commerce team should log each item’s recommendation reason, separate “popular item” placeholders from scored results, and test whether fallback-heavy lists still fit the page context. The check matters most for narrow availability, price or category filters, where a full-looking carousel can hide a thin personalized result set. This is documented behavior, not a measured sales uplift.
❤1
Ozon adds prompt-based backgrounds to product-photo editing
Ozon has updated the AI photo editor in its seller account. When uploading product media, a merchant can now describe a setting in text or provide a reference image whose visual style becomes the basis for a new background.
The workflow sits inside the product-media upload rather than a separate creative tool: open the Media section, choose the AI editor, then supply a prompt or reference. The earlier automatic mode also remains available and returns two background options without instructions.
The release changes background production, not the underlying product shot. Sellers still need to check that the generated scene preserves the item’s shape, colour and scale before adding it to a live listing. Ozon has not published conversion results for the feature.
Ozon has updated the AI photo editor in its seller account. When uploading product media, a merchant can now describe a setting in text or provide a reference image whose visual style becomes the basis for a new background.
The workflow sits inside the product-media upload rather than a separate creative tool: open the Media section, choose the AI editor, then supply a prompt or reference. The earlier automatic mode also remains available and returns two background options without instructions.
The release changes background production, not the underlying product shot. Sellers still need to check that the generated scene preserves the item’s shape, colour and scale before adding it to a live listing. Ozon has not published conversion results for the feature.
AI referral tracking still breaks before the order
PYMNTS surveyed 60 senior ecommerce decision-makers at established US merchants. Among businesses selling through their own site or app, 37% could identify visits from AI platforms but lost the trail before purchase; another 51% had only inconsistent attribution.
To close that gap, merchants need to preserve the AI referral through product viewing and checkout, then connect it to order value, fees and returns. Teams can then compare listing changes or agent partnerships using completed orders rather than visits alone.
The study also qualifies what attribution can show. An AI touchpoint may influence a shopper who later returns directly, and attribution does not prove that the order was incremental.
PYMNTS surveyed 60 senior ecommerce decision-makers at established US merchants. Among businesses selling through their own site or app, 37% could identify visits from AI platforms but lost the trail before purchase; another 51% had only inconsistent attribution.
To close that gap, merchants need to preserve the AI referral through product viewing and checkout, then connect it to order value, fees and returns. Teams can then compare listing changes or agent partnerships using completed orders rather than visits alone.
The study also qualifies what attribution can show. An AI touchpoint may influence a shopper who later returns directly, and attribution does not prove that the order was incremental.
Guide: filter the catalogue before recommendations go live
Adobe Commerce lets a team apply inclusion and exclusion rules to its AI recommendations. A unit can include only products from a chosen category or price range, then remove items that are out of stock, low in stock or already shown elsewhere on the page.
Start by defining which products may appear. Preview the returned products before activating the unit. Adobe applies inclusion rules first and exclusions second. Dynamic rules can use the category or product currently being viewed; the home page supports only static conditions.
Each additional rule narrows the result set, so a heavily filtered unit may return fewer products than planned. Check the preview against real catalogue states, including low stock and discounted prices.
Adobe Commerce lets a team apply inclusion and exclusion rules to its AI recommendations. A unit can include only products from a chosen category or price range, then remove items that are out of stock, low in stock or already shown elsewhere on the page.
Start by defining which products may appear. Preview the returned products before activating the unit. Adobe applies inclusion rules first and exclusions second. Dynamic rules can use the category or product currently being viewed; the home page supports only static conditions.
Each additional rule narrows the result set, so a heavily filtered unit may return fewer products than planned. Check the preview against real catalogue states, including low stock and discounted prices.
AI recommendations expose weak product descriptions at the return desk
Blue Yonder surveyed 2,000 U.S. shoppers about purchases influenced by AI recommendations. Nearly a quarter had bought this way. Of that group, 46% had returned at least one recommended item in the past year. A mismatch with the description was the leading reason, cited by 36% of returners, followed by incorrect size or fit at 23%.
Before recommendation feeds go live, catalog teams should compare titles, attributes, measurements, materials and variant images with the item the warehouse will actually ship. AI can select from the data it receives, but cannot fix a vague or contradictory product record when making a recommendation. These figures are survey responses, not a measured return rate across all AI-assisted orders.
Blue Yonder surveyed 2,000 U.S. shoppers about purchases influenced by AI recommendations. Nearly a quarter had bought this way. Of that group, 46% had returned at least one recommended item in the past year. A mismatch with the description was the leading reason, cited by 36% of returners, followed by incorrect size or fit at 23%.
Before recommendation feeds go live, catalog teams should compare titles, attributes, measurements, materials and variant images with the item the warehouse will actually ship. AI can select from the data it receives, but cannot fix a vague or contradictory product record when making a recommendation. These figures are survey responses, not a measured return rate across all AI-assisted orders.
A Russian-language feed tracking AI in retail operations
Ecom: The Future covers specific uses of AI in marketplaces and online retail. A recent post looks at Walmart’s Sparky workflow, which turns a handwritten or digital shopping list into a cart. It also states the limitation: the buyer still has to confirm the exact product, size and quantity. Other posts examine catalog work, search and store operations, and distinguish features already in use from vendor promises.
Ecom: The Future covers specific uses of AI in marketplaces and online retail. A recent post looks at Walmart’s Sparky workflow, which turns a handwritten or digital shopping list into a cart. It also states the limitation: the buyer still has to confirm the exact product, size and quantity. Other posts examine catalog work, search and store operations, and distinguish features already in use from vendor promises.
❤1
REP AI brings verified reviews into product conversations
REP AI has added a Loox integration for Shopify stores. After a merchant connects the reviews app using its Store ID and API key, the shopping agent can quote current star ratings and selected top reviews in chat. Product cards also show the Loox rating beside each item.
This gives the agent review evidence when a shopper asks how well a product has been received, instead of relying only on catalog copy. The evidence appears in the same conversation as other product questions. Each store can connect only one reviews app. Disconnecting Loox, Yotpo, Judge.me or Okendo immediately stops the AI from quoting its reviews. REP AI has not reported a conversion result for the release.
REP AI has added a Loox integration for Shopify stores. After a merchant connects the reviews app using its Store ID and API key, the shopping agent can quote current star ratings and selected top reviews in chat. Product cards also show the Loox rating beside each item.
This gives the agent review evidence when a shopper asks how well a product has been received, instead of relying only on catalog copy. The evidence appears in the same conversation as other product questions. Each store can connect only one reviews app. Disconnecting Loox, Yotpo, Judge.me or Okendo immediately stops the AI from quoting its reviews. REP AI has not reported a conversion result for the release.
❤1
Reformation puts AI into merchandise planning
Fashion retailer Reformation has implemented an AI-backed planning tool that evaluates product performance and forecasts demand. Planners use its signals to identify items gaining or losing momentum, adjust assortments by location and direct stock toward expected demand.
The company says the tool is improving in-stock rates and supporting more full-price selling. That is management's account from an earnings call, not an independently measured result. Reformation has not disclosed the vendor, model or forecast accuracy.
The tool does not automate buying. It adds a demand signal to the retailer's existing assortment, allocation and replenishment process, where planners still make the documented decisions.
Fashion retailer Reformation has implemented an AI-backed planning tool that evaluates product performance and forecasts demand. Planners use its signals to identify items gaining or losing momentum, adjust assortments by location and direct stock toward expected demand.
The company says the tool is improving in-stock rates and supporting more full-price selling. That is management's account from an earnings call, not an independently measured result. Reformation has not disclosed the vendor, model or forecast accuracy.
The tool does not automate buying. It adds a demand signal to the retailer's existing assortment, allocation and replenishment process, where planners still make the documented decisions.
❤1