Ecommerce marketing in 2026: marketplaces, retail media, social commerce and AI shopping agents

Ecommerce marketing in 2026 is a product data problem as much as an advertising problem. Worldwide retail ecommerce is on track for about $6.88 trillion this year, AI assistants now send retail traffic that converts roughly 60% better than everything else, and the in-chat checkout that everyone predicted has quietly moved back onto merchants’ own sites.

The numbers that set the 2026 baseline

Start with the size of the pool. eMarketer’s most recent published figures put worldwide retail ecommerce at $6.419 trillion in 2025, or 20.5% of all retail, growing 6.8% year on year, with the 2026 total forecast at around $6.88 trillion. That is roughly one retail pound in five, and the share is still climbing about a point a year.

What that means in practice: the category is no longer growing fast enough to hide bad unit economics. When online retail was compounding at 20% a year you could buy your way to a revenue number. At 7% or 8% you cannot. Most of the brands I see struggling in 2026 are not struggling because a channel died. They are struggling because acquisition cost drifted up while average order value stayed flat, and nobody rebuilt the model.

Where the budget should go, by stage

There is no universal split, but there is a defensible starting point that changes with revenue. A store doing $50,000 a month cannot afford the same portfolio as one doing $5 million, and the mistake at both ends is symmetrical: small brands spread themselves across nine channels, large brands keep pouring money into the two that are already saturated.

The table below is a starting allocation, not a rule. Treat it as the shape of a first draft that your own contribution-margin data then bends.

Two notes on that table. First, the owned percentage looks small because email and SMS are cheap to run, not because they matter less. Measured on revenue rather than spend they usually punch four or five times above their budget line. Second, the organic and content slice is where most brands under-invest, then complain that paid costs keep rising. Those two facts are related. Our complete guide to digital marketing channels covers the wider budget logic if you are building the plan from scratch.

Your product feed is the new landing page

For most of the last decade the product feed was plumbing. Get the GTIN right, keep the price in sync, move on. That changed in 2026 because three different systems now read the same file and make consequential decisions from it.

Google made this explicit with AI Max for Shopping campaigns. Google’s own documentation says the system pulls attributes from your Merchant Center feed, things like waterproof or available for click and collect, and uses them to match conversational, long-tail queries that keyword-matched Shopping campaigns never saw. Google states that advertisers who turn it on typically see about 5% more conversions at a similar CPA. The controls are real too: up to 25 term exclusions and 40 messaging restrictions per campaign, plus brand and URL exclusions.

The same fields feed marketplace search relevance and, increasingly, what an AI assistant says about your product when a shopper asks. A missing material attribute is no longer a data-quality warning. It is a query you cannot appear for.

Practical priorities, in the order I would work them:

• Titles that lead with the attribute people search, not the internal SKU name. Brand plus product type plus the two attributes that differentiate it.
• Fill the optional attributes: material, pattern, fit, size system, age group, product highlights. These are exactly what AI Max and assistant tools read.
• Structured data on the product page that matches the feed. Price, availability and review counts that disagree between feed and page cause disapprovals and, worse, wrong AI answers.
• Real images at multiple angles, with lifestyle shots as secondary images rather than replacements for the clean product shot.
• Availability accuracy over everything. Nothing destroys assistant recommendations faster than out-of-stock items being surfaced.

If you are also running search campaigns, the feed work overlaps heavily with the account structure decisions in our guide to PPC and paid search.

Retail media and the 89% problem

Retail media is the biggest budget shift of the decade and the most oversold. eMarketer forecasts $71.09 billion of US retail media spend in 2026, and in the same analysis notes that Amazon and Walmart will take more than 89% of the incremental spend. Amazon’s own advertising services line reached $19.8 billion in Q2 2026, up 26% year on year, with Sponsored Products still doing most of the work. eMarketer expects Amazon’s retail media revenue to pass $75 billion by 2028, more than $65 billion clear of the next network.

So the honest read is this. Two networks are genuine channels. The rest are trade-marketing line items you negotiate with a buyer, and you should budget them that way. Instacart, for context, generated about $960 million in US ad revenue in 2025, which is a real business and still an order of magnitude below Amazon.

The more interesting movement is off-site. Walmart Connect opened self-service access to full-funnel campaigns on Meta in April 2026, with TikTok due later in the year, and began beta testing an Add to Cart control inside social ads that drops up to 10 items straight into a Walmart basket. In a Burt’s Bees beta, that combination drove 42% of ad-attributed sales, with 95% of purchases inside a week. eMarketer’s March 2026 forecast has US retail media social spend growing 23.6% to $1.26 billion.

“With these social offerings, we’re giving more advertisers more opportunities to access our full-funnel capabilities via a self-service model.” Diana Finster, Walmart Connect

The mechanics of buying that inventory sit closer to display than to search, so the auction and supply-path thinking in our programmatic advertising guide transfers directly.

AI shopping agents and what agentic checkout became

This is the section most 2026 ecommerce advice gets wrong, usually because it was written in late 2025 and never updated.

OpenAI launched Instant Checkout in September 2025 with Etsy and Shopify merchants, built on the Agentic Commerce Protocol it developed with Stripe. By March 2026 it had changed course. Digital Commerce 360 reported on 6 March 2026 that OpenAI was de-prioritising in-chat checkout for product listings in favour of merchant-branded apps inside ChatGPT, with Target, DoorDash, Instacart and The Knot among the first. Merchants had pushed back on how much a real transaction involves.

“That is the checkout itself, that is subscriptions, that is the inventory, that is shipping taxes.” Shopify’s president, on transaction complexity

Google moved in the opposite direction. At I/O on 19 May 2026 it introduced Universal Cart, a cart that works across merchants and across Search, Gemini, YouTube and Gmail, sitting on a Shopping Graph of more than 60 billion listings. Launch checkout partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair and Shopify merchants such as Fenty and Steve Madden, with the Universal Commerce Protocol expanding to Canada, Australia and later the UK. Payments are handled by the Agent Payments Protocol, which uses signed intent, cart and payment mandates so a merchant can verify that a human actually authorised the purchase.

And the traffic is real, whoever owns the checkout. Adobe’s analysis of more than a trillion visits to US retail sites found AI-referred traffic up 62% year on year in July 2026, converting 60% better than non-AI traffic, generating 53% more revenue per visit, bouncing 33% less and adding to cart 28% more often.

There is also a legal answer forming to the question of whether you can keep agents out. Amazon sued Perplexity in November 2025 over its Comet browser agent and won a preliminary injunction in March 2026. On 4 August 2026 the Ninth Circuit vacated it, reasoning that when a user directs an assistant to shop, it is the user reaching Amazon’s servers, not Perplexity. The underlying case continues. But the direction of travel is that terms of service alone will not keep agents away, so plan for them to arrive.

Social commerce, TikTok Shop and paid social

TikTok Shop is now a serious US retailer. eMarketer forecast on 27 April 2026 that its US ecommerce sales would rise another 48% in 2026 after 108% growth the year before, with roughly 24.9% of US digital buyers and more than half of social buyers making a purchase there, at an average of $118 a year across three or four orders.

That $118 figure is the one to sit with. TikTok Shop is a high-frequency, low-basket channel. It suits consumables, beauty, accessories and impulse price points. It is a poor first channel for a considered $600 product, whatever the case studies say.

Salesforce’s 2026 holiday forecast, published 20 July 2026, expects social commerce to grow at nine times the rate of traditional ecommerce, with social shops accounting for 7% to 9% of orders and 28% of Gen Z planning to shop in social apps. Salesforce also expects 20% of holiday ecommerce traffic to originate from AI chat agents and reports that 50% of shoppers now use AI assistants somewhere in the buying journey, up 67% year on year.

On paid social, the automation products (Meta Advantage+ and its equivalents) work well when your account already sends dense, clean purchase signals and struggle when it does not. That is a data quality question, not a creative one. Our social media marketing guide covers platform-by-platform mechanics, and the creator side of the same budget is in the influencer and creator marketing guide.

Marketplaces versus your own store

The old framing was a choice. The 2026 framing is a portfolio with a deliberate margin trade.

Marketplaces give you demand that already exists and a checkout that already converts. They take a commission, they own the customer relationship, and they sell advertising back to you against your own listings. Your own store gives you margin, first-party data, merchandising control and the ability to build a repeat-purchase business, but you pay for every visit.

Shopify’s numbers show the DTC route is far from dead. The quarter ended 30 June 2026 produced more than 30% growth in GMV, revenue, gross profit and free cash flow, with GMV reported at $115.6 billion, up 32%, a fifth consecutive quarter above 30%.

A workable rule: use marketplaces to buy volume and category data, use your own store to buy customers. If more than about 70% of revenue comes from one marketplace, you have a distribution risk rather than a business, and the fix is usually email capture and a reason to reorder direct rather than a new channel.

Store CRO: the checkout is where money leaks

Baymard Institute’s rolling figure across 50 studies is a 70.22% average cart abandonment rate. More useful than the headline are the stated reasons, because four of the top six are things you control in an afternoon.

Baymard estimates the average large ecommerce site can gain a 35.26% conversion improvement from checkout design alone, and puts the recoverable lost orders across US and EU ecommerce at about $260 billion. Those are not typical results for a single test. They are the size of the gap between an average checkout and a well-built one, accumulated over many fixes. The testing method for getting there is in our conversion rate optimisation guide.

Retention, email and SMS

Retention is where ecommerce economics are decided, and email plus SMS is where retention is executed. Klaviyo’s 2026 benchmarks, drawn from more than 183,000 customers, show automated flows producing close to 41% of email revenue from 5.3% of sends, with revenue per recipient roughly 18 times higher than campaigns and placed-order rates around 13 times higher. Flow click rates average 5.58% against 1.69% for campaigns.

The finding I would act on first: nearly 48% of flow-driven email revenue comes from new buyers, against 16% for campaigns. Welcome, browse abandonment and cart abandonment flows are acquisition infrastructure, not retention infrastructure, and most brands staff them accordingly badly. Klaviyo also reports AI product recommendations lifting flow click rates to 3.75% on average and 8.79% for top performers.

Sequencing that actually matters, roughly in order of payback: welcome series, cart and checkout abandonment, browse abandonment, post-purchase and replenishment, winback, then VIP or loyalty. Deliverability sits underneath all of it, and the sender authentication rules that bite hardest are covered in the email and SMS marketing guide.

Affiliate and cashback partners also belong in the retention conversation, because they intercept a lot of returning-customer demand at the last click. Our affiliate marketing guide explains how to structure commissions so you are not paying full rate for traffic you already owned.

Measurement and the KPIs worth reporting

Three numbers run an ecommerce business: contribution margin after all variable costs, blended customer acquisition cost against contribution, and repeat purchase rate at 90 and 365 days. Channel ROAS is a diagnostic, not a target.

The measurement problem in 2026 is that AI assistants sit between the demand and your analytics, and they attribute poorly. A shopper who researches in ChatGPT or Gemini and then types your brand name into a browser shows up as direct or brand search. Adobe’s data suggesting AI visits convert 60% better is partly a selection effect: those visitors arrive further along. Both things can be true, and neither is a reason to under-count the channel.

What to do about it: track referral hosts from the major assistants separately, add a post-purchase survey question asking how people first heard about you, and compare that self-reported mix against your platform-reported mix every month. When they diverge, believe the survey for budget decisions and the platform for optimisation decisions. The full setup, including server-side tagging and incrementality testing, is in the marketing analytics guide, and the AI visibility side is covered in the AI in digital marketing guide.

Tariffs, parcel fees and margin

Cross-border economics changed twice in two years and a lot of pricing models have not caught up. The United States ended de minimis treatment for shipments from China and Hong Kong in May 2025 and for all other origins in August 2025, so low-value parcels now face formal entry and full duties.

Europe followed. On 12 December 2025 the Council of the EU agreed to levy customs duty on small parcels from 1 July 2026, with a temporary flat charge on consignments below the 150 euro threshold until a permanent system is built. The Commission’s own figure is roughly 4.6 billion sub-150 euro parcels entering the EU in 2024, more than 90% from China.

For a brand shipping cross-border, the practical consequence is that landed cost now needs to appear in your acquisition model, not only your finance model. Salesforce expects retailers to spend an extra $3 billion globally subsidising free shipping this holiday season, a 7% rise in logistics costs, while 35% of holiday shoppers trade down to cheaper alternatives. Squeezing both ends at once is how brands end up growing revenue and losing money.

A 90-day ecommerce marketing plan

If you are inheriting an account or resetting one, this is the order I would work in.

Weeks 1 to 3, fix the data. Audit the product feed against the optional attributes AI Max and marketplace search read. Correct availability and price mismatches between feed and page. Confirm product schema matches the feed. Get conversion tracking and server-side tagging verified before you judge any channel.

Weeks 4 to 6, fix the checkout. Work Baymard’s abandonment reasons top down. Surface shipping cost on the product page, give a delivery date, enable guest checkout, cut form fields, add monitoring specifically on the checkout path. Do not test yet, do repair.

Weeks 7 to 9, rebuild owned. Welcome, cart and browse abandonment flows first, because that is where Klaviyo’s new-buyer revenue concentrates. Then post-purchase and replenishment. Add SMS only where consent is properly captured.

Weeks 10 to 12, then buy growth. With clean data, a working checkout and functioning flows, turn attention to paid: Shopping and AI Max on Google, one or two retail media networks that actually matter to your category, and a creator programme with proper usage rights. Set contribution-margin targets, not ROAS targets.

The sequence is deliberate. Every one of the first three blocks makes the fourth cheaper. Reverse the order and you pay retail price for traffic that leaks out of a broken funnel. For what has changed most recently across these channels, see our ecommerce and retail media news for September 2026, and for the organic side, the SEO guide for 2026.

Last researched and updated: 7 September 2026.