eCommerceSEO

AI in eCommerce in 2026: search, recommendations, content and shopping agents

Updated 10 min readBy Jaskirat Singh Talwar

A storefront with a striped awning, plants and a small document on an orange background

For an online store in 2026, AI is four practical things: better on-site search and recommendations, product content produced at scale and reviewed by a person, a support assistant that answers from your own policies, and readiness for the AI shopping agents that now send measurable traffic. Adobe measured a 693.4% year-on-year increase in traffic to US retail sites from generative AI tools during the 2025 holiday season, from a small base. The 2019 version of this article predicted chatbots, personalised ads and voice commerce; two of those arrived in a different form, and one did not.

Key takeaways

  • AI referrals are real but small: Adobe Analytics reported that "generative AI tools drove a 693.4% increase in traffic to retail sites" between 1 November and 31 December 2025, during a season in which US consumers spent $257.8 billion online.
  • AI answers now sit on research queries more than purchase queries: Semrush's July 2026 study of 600,000+ keywords found AI Overviews grew 71% on commercial-intent results while falling 5% on transactional ones.
  • Native tooling is cheaper than it was. Shopify's Search & Discovery app (filters, synonyms, boosts, recommendations) is free; Adobe Commerce Live Search is "included in your license"; WooCommerce has added MCP for agent access, which its documentation describes as "currently in developer preview" (October 2026).
  • Shopify now exposes eligible products to AI channels through Shopify Catalog automatically; you can block individual agentic storefronts but "can't opt out of Shopify Catalog itself".

What the 2019 version got right, and wrong

The original article listed five ways AI would change eCommerce: chatbots, personalised advertising, behavioural retargeting, voice search and personalised recommendations. Recommendations and ad targeting were already mainstream then and still are; the platforms do most of that work now. Chatbots arrived, but as grounded assistants that read your policies rather than scripted menus (see how to build a support chatbot). Voice commerce did not become a channel worth building for. What nobody on the 2019 list anticipated is the fourth area below: shoppers asking an AI assistant to find and compare products, and the assistant fetching your catalogue.

The four areas, and what each needs

Area What it does for the store Native options (as of October 2026) What to measure
Search and discovery Understands "waterproof jacket under 200" rather than matching the string; re-ranks by behaviour; fixes synonyms and zero-result queries Shopify Search & Discovery (free; filters, synonym groups, product boosts, related and complementary products). Adobe Commerce Live Search (replaces default search; "AI-powered dynamic faceting and re-ranking"; included in the licence). WooCommerce: third-party search plugins or an external engine Zero-result rate, search-to-cart rate, revenue per search session
Recommendations and merchandising Related, complementary and "customers also bought" placements; rule-based boosts for promotions Shopify Search & Discovery recommendations; Adobe Live Search merchandising rules (if-then boost or bury); third-party engines Click-through on recommendation slots, attach rate, average order value
Product content at scale Drafts descriptions, attributes, alt text and translations from your specifications, for a person to review and publish Platform assistants (Shopify Magic and Sidekick; see Shopify's help centre for current scope) and any language model with a review step Attribute completeness, time per SKU, return rate on AI-described products
Support and post-purchase Answers order, shipping and returns questions from your policies; hands off to a person Helpdesk AI agents or a knowledge-base assistant (see our RAG buyer's guide) Deflection rate, hand-off rate, answer accuracy on a fixed test set
Agent readiness Makes the catalogue, prices, stock and policies readable by AI shopping assistants and search engines Shopify Catalog and agentic storefronts; Merchant Center feeds plus Product structured data on every platform; WooCommerce MCP (announced for 10.3) Sessions in GA4's "AI Assistant" channel, orders from AI referrers, crawler hits in logs

Two notes on the table. First, the native options have closed much of the gap with third-party engines for small and mid-size catalogues; pay for an external search engine when you have a large catalogue, multiple languages or merchandising needs the native tool cannot express. Second, "AI-generated product content" only pays off with a review step. Google's spam policies treat scaled content produced primarily to manipulate rankings as abuse, and unreviewed descriptions are also where product-fact errors creep in.

Where the AI traffic comes from

Three sources, measured differently:

  • AI assistants (ChatGPT, Gemini, Copilot, Perplexity, Claude) that answer a shopping question and link to you. GA4 now has a default channel called "AI Assistant" for "users [who] arrive at your site from sources like ChatGPT", Gemini, Copilot and others; Google notes that its own AI Overviews and AI Mode are counted under Organic Search instead.
  • Google AI Overviews and AI Mode, which cite product and category pages and are reported inside Search Console's normal Performance report.
  • Agentic storefronts, where the assistant presents your products directly and may complete checkout. Shopify's help centre describes Shopify Catalog as "a structured source of product information gathered from eligible products that are sold by stores on Shopify"; qualifying products are "automatically included", and merchants "can choose to stop individual agentic storefronts from having access" but cannot opt out of Catalog. Checkout behaviour differs by channel and changes often; our agentic commerce readiness guide tracks the protocols (ACP and UCP) and the current state.

The Semrush finding matters for expectations: AI answers are taking over "which should I buy" questions faster than "buy this now" queries. Category pages, buying guides and comparison content are where AI citations are won; the product page still closes the sale.

A 90-day plan for a store

  1. Baseline the numbers. Zero-result searches, search-to-cart rate, support tickets by intent, GA4 sessions in the "AI Assistant" channel, and Search Console impressions for your top 20 category pages. Without this, nothing below can be shown to have worked.
  2. Turn on the native search tool and fix the top 50 zero-result queries with synonyms and filters. On Shopify this is the free Search & Discovery app; on Adobe Commerce, Live Search; on WooCommerce, pick a search plugin or engine. One week.
  3. Clean the catalogue data that everything else depends on: titles, GTINs, attributes, availability, images. AI search engines, Merchant Center and shopping agents all read the same fields, and the agent protocols cannot repair bad data.
  4. Publish Product structured data on every product page and a Merchant Center feed. Google's documentation says "providing both structured data on web pages and a Merchant Center feed maximizes your eligibility" for product experiences. Keep the markup identical to the visible price and stock.
  5. Add recommendations in two slots (product page and cart) using the native tool, and measure attach rate for four weeks before adding a paid engine.
  6. Run a product-content pilot on one category: generate descriptions and attributes from the specification sheet, have a merchandiser review every one, and compare completeness and time per SKU with the manual process.
  7. Pilot a grounded support assistant on order-status, shipping and returns questions only, with a hand-off to your helpdesk, using the procedure in our chatbot guide.
  8. Check agent and crawler access: robots.txt and CDN rules allow the search crawlers (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot); on Shopify, review Catalog and agentic storefront settings; on WooCommerce, decide whether to enable the MCP server and for whom.
  9. Report monthly on the baseline metrics from step 1 plus AI-referred sessions and orders.

Platform notes

Shopify. The most "on by default" platform for AI distribution: eligible products flow into Shopify Catalog, and the controls to block individual agentic storefronts live in the admin. Search & Discovery and the platform's own content assistants cover the first three areas without an app bill. The work is data quality and deciding which agent channels you want.

Adobe Commerce. Live Search replaces the default search, adds a product listing widget with filters and uses behaviour-based re-ranking; it is included in the licence, so there is rarely a reason to leave default search running. Recommendations and content assistance are separate services; agent readiness is feeds plus structured data, and checkout by agents depends on your payment provider.

WooCommerce. The most assembly required. Search, recommendations and content are plugin or service choices. The platform's own direction is agent access through MCP, with, in the words of its October 2025 developer post, "the goal of expanding beyond store management and potentially powering consumer shopping and checkout". As of October 2026 WooCommerce's developer documentation states that "the MCP implementation in WooCommerce is currently in developer preview"; check the official WooCommerce MCP documentation for current MCP support before planning around it.

What not to do

  • Do not publish AI-written product copy without a reviewer who knows the product. The error rate is low per description and unacceptable across a catalogue.
  • Do not block "AI bots" wholesale at the CDN; you remove yourself from AI search answers while training crawlers can be blocked separately. Our GEO guide lists the crawlers.
  • Do not buy a search engine before fixing synonyms and attributes in the native tool; most zero-result problems are data problems.
  • Do not let an assistant promise delivery dates or refunds it cannot see in your systems; ground it in policy pages and hand off the rest.

FAQ

Does AI actually send traffic to online stores yet?

Yes, from a small base. Adobe Analytics recorded a 693.4% year-on-year rise in traffic to US retail sites from generative AI tools over the 2025 holiday season (1 Nov to 31 Dec). GA4's "AI Assistant" channel lets you see your own number; Google's AI Overviews and AI Mode are counted under Organic Search.

Which AI feature should a small store add first?

Site search. The native tools (Shopify Search & Discovery, Adobe Commerce Live Search) are free or included, the zero-result rate is easy to measure, and fixing synonyms and filters improves conversion for every visitor, not just those arriving from AI.

Is it safe to generate product descriptions with AI?

With review, yes. Generate from the specification sheet, have a person check facts and tone, and measure return rates on the affected products. Unreviewed descriptions at scale risk product-fact errors and fall under Google's scaled-content spam policy.

Do I need to do anything for AI shopping agents?

On Shopify, eligible products are already in Shopify Catalog; review which agentic storefronts you allow. On every platform, publish Product structured data that matches the visible page, keep a Merchant Center feed current, and allow the search crawlers. Checkout inside assistants is still changing; keep your own checkout reliable.

What does AI in eCommerce cost?

Native search, recommendations and content assistants are free or bundled on Shopify and Adobe Commerce. Third-party search engines, helpdesk AI agents and knowledge-base assistants are priced per volume (queries, resolved conversations or tokens); vendor pricing pages are linked below. The largest cost is usually catalogue data clean-up, which every other item depends on.

Impinge Solutions builds and maintains Shopify, Adobe Commerce and WooCommerce stores and offers an AI for eCommerce readiness check. Ask for the store check and we will start with your zero-result searches and catalogue data.

Sources (11 links)
Share this article