Digital Outlook 2027E-Commerce EditionPublished September 2026
The cart moved somewhere other than your store.
What the second half of 2026 taught online retailers about AI shopping, commerce technology, discovery, paid media, and retention, and the ten things we expect in 2027. Written for established retailers at roughly $10 to $15 million in revenue and up.
The edition in brief
In the first half of 2026, AI systems learned to read a retailer's catalog. In the second half, they started holding the shopper's cart. Google's Universal Cart gathers products from many retailers across Search and Gemini and watches prices and stock on the shopper's behalf, while the brand stays the merchant of record. What that arrangement does not protect are the on-site moments that used to build the commercial relationship: the cross-sell, the email capture, the fit guidance, and the gift note. The defining question for 2027 is how a retailer stays the store a customer remembers when the checkout happens somewhere else.
How we source
We retired an industry statistic that didn't hold up
Widely repeated · Retired
82%
No grade. Will not be cited again.
The claim that 82% of TikTok Shop sellers lose money circulated widely last year, and we repeated it in our June edition. For this one, we went looking for the primary source and couldn't find it, so we've retired the figure and won't cite it again.
That's the standard we want to be held to, and it's why the scorecard opens this edition instead of closing it. We attribute industry data by source wherever it informs a point, and where our own results sharpen one, we aggregate and anonymize the figures across the retailers we manage so they never identify a client. When the evidence is still mixed, we say so and show you where.
01 · The scorecard
How did the forecasts hold up?
A year of confident retail forecasting produced a mixed record, and the direction of the misses is instructive. The demand forecasts ran conservative, with Adobe predicting 5.3% growth for the 2025 holidays against actual growth of 6.8% reaching $257.8 billion. The agentic checkout forecasts ran hot. OpenAI and Shopify said in September 2025 that millions of merchants would soon offer Instant Checkout inside ChatGPT, and roughly a dozen to thirty went live before OpenAI moved checkout back to merchants in March 2026.
Our own June calls fared better, and we grade them in the same format. In-chat purchasing is not yet a real revenue channel and discovery is, which held. AI referrals are under 1% of sessions but the curve is the point, which held, with assistant sessions reaching 0.54% of total in Q3 from 0.25%, and monthly volume growing 2.3 times.
Interactive · Make your call
Each card shows only what was predicted. Guess before you reveal.
Ten forecasts: five from the industry, five from our own June edition, graded September 2026.
0 / 0
Called correctly
- Under-called
- Missed
- Behind schedule
- Holds for repeat purchases
- Held
- Half right
- Held, with a caveat
Figure 01 · Industry forecasts, graded
Adobe · Holiday 2025
US online holiday sales would grow 5.3% across November and December 2025.
Your read?
OpenAI and Shopify · Sep 2025
Millions of Shopify merchants would soon offer Instant Checkout inside ChatGPT.
Your read?
Google · Jan 2026
Checkout through the Universal Commerce Protocol would expand beyond the US during the year.
Your read?
Google · May 2026
Universal Cart would roll out across Search and the Gemini app in the US over the summer.
Your read?
Adobe · Holiday 2025
Shoppers arriving from AI tools convert 31% more often than other visitors.
Your read?
Graded by usFigure 02 · Our own June calls
Happy Cog · June
In-chat purchasing is not yet a real revenue channel for most retailers, and discovery is.
Your read?
Happy Cog · June
AI referrals are under 1% of sessions, but the curve is the point.
Your read?
Happy Cog · June
Amazon walled itself off from outside AI agents.
Your read?
Happy Cog · June
Protect AOV and judge conversion by channel and cohort.
Your read?
Happy Cog · June
The biggest efficiency gains show up where there's the most headroom and the cleanest first-party data.
Your read?
Industry estimate · Repeated in June
82% of TikTok Shop sellers lose money.
A widely repeated figure we couldn't trace to a primary source, so we retired it.
Retired · No gradeSources: Adobe Digital Insights (2025 holiday season results), OpenAI (March 24, 2026 merchant checkout update), Google (Universal Commerce Protocol and Universal Cart announcements). June positions are drawn from the first e-commerce edition, shared with clients in June 2026.
02 · Executive summary
What changed in the second half of 2026
Checkout came back to the merchant, for now, with assistants handling discovery while stores handle the sale. The protocols narrowed toward Google's, and most retailers will end up supporting both, which a clean product feed already serves. AI traffic kept compounding and stayed small. Paid automation absorbed Shopping, with Performance Max becoming the largest single channel across our full paid book, which makes conversion data quality the main lever a retailer still controls. And first-party capture became the retention problem, because orders that start in an assistant skip the on-site moments that fed email and SMS.
-
01
Checkout came back to the merchant, for now.
OpenAI retreated from in-chat checkout in March, and the working model since has been that assistants handle discovery while stores handle the sale. Google is the one company still pushing native checkout, and carefully.
-
02
The protocols narrowed toward Google's.
OpenAI's Agentic Commerce Protocol now powers a smaller set of retailer apps, while Google's Universal Commerce Protocol keeps adding cart, catalog, and loyalty linking. A clean product feed serves both.
-
03
AI traffic kept compounding and stayed small.
Assistant referrals more than doubled in our retail book and still account for about half a percent of sessions. The conversion premium shows up first in repeat purchases.
-
04
Paid automation absorbed Shopping.
Performance Max became the largest single channel across our full paid book, and Shopping grew inside it. Conversion data quality is the main lever a retailer still controls.
-
05
First-party capture became the retention problem.
Orders that start in an assistant skip the on-site moments that fed email and SMS. The post-purchase flow is now the front door of the retention program.
03 · The macro picture and the holiday read
A steady season on a bigger base
$318.9B
Deloitte's upper forecast for holiday e-commerce, November 2026 through January 2027
6.6%
eMarketer's projected e-commerce growth for the 2026 season
$26.4B
US online spend across Prime Day, June 23 to 26, 2026 (Adobe)
+393%
Growth in AI-referred traffic to US retail sites, Q1 2026 versus a year earlier (Adobe)
The 2026 holiday forecasts point to a normal year. Deloitte expects holiday e-commerce to grow 7.5% to 8.4%, reaching as much as $318.9 billion across November through January, and eMarketer projects roughly 6.6% growth for the season. The more interesting number arrived in June: US retailers drove $26.4 billion in online spend across the four days of Prime Day, up 9.3% year over year (Adobe), against $32.4 billion across Thanksgiving, Black Friday, and Cyber Monday in 2025.
For a growth-stage retailer, the promotional calendar now has two peaks to plan around, and the summer one is where competitors test the offers they'll run in November.
Build 2027 targets off a trailing twelve months, split by channel and cohort, so an inflated 2026 half doesn't set the bar.
04 · Agentic commerce
What do you keep when an agent runs the checkout?
OpenAI moved away from in-chat checkout in March 2026, and Walmart said purchases completed inside ChatGPT converted at one-third the rate of those that clicked through to its website. Google took the opposite path, introducing the Universal Commerce Protocol in January with Shopify, Etsy, Wayfair, Target, and Walmart, then announcing Universal Cart in May.
Google's protocol offers two ways to sell. In the native version, the AI surface renders the checkout. In the embedded version, the merchant keeps its own checkout inside the agent's flow. Both keep you as merchant of record, and they leave you with different shares of the customer relationship.
| Native checkout | Embedded checkout | |
|---|---|---|
| Legal relationship and returns | You keep it | You keep it |
| Order and payment data | Passed to you after the sale | Captured in your own checkout |
| Email and SMS opt-in at purchase | Depends on the surface; often absent | Your own consent flow |
| Cross-sell, bundles, and free-shipping thresholds | Largely lost | Available in your checkout |
| Loyalty and member pricing | Through identity linking, where supported | Your program as it runs today |
| Conversion friction for the shopper | Lowest | Slightly higher |
Happy Cog's take
Sell where the agents are, and decide deliberately how much of the checkout you hand over. Native checkout makes sense for replenishment and low-consideration items, where the shopper wants speed and the cross-sell is small. It's a harder call for gifting, high-AOV, and fit-sensitive products, where the on-site moments carry the margin. Whichever you choose, keep a first-party capture step after every agent-originated order.
05 · Discovery and AI visibility
Why is AI traffic so small and so valuable at the same time?
Across our retail book, assistant referrals more than doubled and still account for about half a percent of sessions. Adobe measured traffic from AI sources to US retail sites up 393% year over year in Q1 2026, following a 693% jump during the 2025 holidays, and is careful to note the base remains modest. The growth rate makes headlines. Your budget should follow the share of sessions.
The conversion premium is real and it is uneven. It shows up first where the purchase is a repeat: at our replenishment brand, assistant visitors convert at roughly twice the organic rate. In considered purchases the premium is still building, which is why we grade Adobe's 31% conversion-premium finding as holding for repeat purchases rather than holding outright.
Salesforce predicts 20% of holiday e-commerce traffic in 2026 will come from AI agents, a figure that includes shopping assistants, background agents, and competitors' price-scraping bots. We expect that to look much smaller in most retailers' own analytics, and the gap between industry estimates and first-party data is one of our ten calls.
From our portfolio · Aggregated and anonymized · Five managed retail GA4 properties · Jul 2025 – Sep 21, 2026
2.3x
Monthly AI-assistant sessions, second half of 2025 to Q3 2026
0.54%
Share of total sessions from AI assistants in Q3, up from 0.25%
2x
Conversion rate of AI-referred sessions versus organic search at our replenishment brand, Q3 2026
| H2 2025 | 1,003 | |
|---|---|---|
| H1 2026 | 1,495 | |
| Q3 2026 | 2,335 |
| AI-referred sessions | 10.1% | |
|---|---|---|
| Organic search | 5.1% |
Interactive · The AI traffic reality check
What share of your traffic comes from assistants?
Enter a month of sessions from GA4. The share is compared with our portfolio benchmark of 0.54% and Salesforce's 20% holiday forecast. Everything is calculated in your browser, and nothing you type is sent anywhere.
Your assistant share
—
Enter both numbers to see your share.
Bars share one scale, from 0% to 20% of sessions.
Where do assistants send shoppers?
At our largest multi-brand retailer, ChatGPT sent visitors to 1,442 distinct pages this year, and the homepage took 37% of them. In our higher education work, the homepage draws only about 6% of assistant traffic.
Assistants recommend a multi-brand retailer as a place to shop as often as they recommend a specific product from it. That puts store-level facts (who you sell to, whether you require a license, how shipping and returns work, and whether products are authentic) on the same footing as product detail.
| Homepage | 37% | |
|---|---|---|
| Other pages | 31% | |
| Category and brand pages | 14% | |
| Product pages | 10% | |
| How-to articles | 5% | |
| Not set | 3% |
Happy Cog's take
Make the store legible, and then make the catalog legible. We'd start with the pages assistants already cite (the homepage, the top category pages, and the policy pages) and rewrite them to state plainly who you serve, what you carry, and how buying from you works. Then complete product structured data, return and shipping policy markup, and feed attributes for the top sellers, and confirm that GA4's AI Assistant channel captures every assistant source label.
Read the storefront, commerce technology, and retention chapters.
The full edition runs 26 pages, with the portfolio data and the methodology behind every figure. No form required.
06 – 09 · The rest of the store
Storefront, technology, paid media, and retention
06 · Storefront design
Design for the shopper who arrives mid-decision
Fit guidance, size charts, comparison tables, and gift recommendations were designed as on-site moments. When an assistant answers the shopper's question before they arrive, those moments only count if the assistant could read them. Speed still decides the sale: each added second of load time is associated with up to a 7% drop in conversions, and average order value held at all three brands we track for a third consecutive period.
Happy Cog's take
Build every product and category page for two readers at once. We'd audit your top 50 landing pages from assistant traffic first, since those are the pages already being recommended.
07 · Commerce technology
The feed became infrastructure
Google's protocol requires a live product feed and an eligibility attribute in Merchant Center before a product can show a buy button in AI Mode, and a feed that's stale by Thursday shows an agent the wrong price. Orders that start in an assistant also break session-based measurement, so server-side conversion tracking is now the only way to count them at all.
Happy Cog's take
Build one source of truth for product data and send it everywhere. As a Stripe Premier Partner and Craft Verified Partner, we wire the payment and checkout side of those integrations as well as the catalog side.
08 · Paid and commerce media
Automation took Shopping, and ads moved into the assistants
Across 56 managed Google Ads accounts and about $23.8 million in measured spend, Performance Max grew from 38.7% to 45.3% of spend and Shopping from 3.7% to 5.9%. At one DTC eyewear brand, Q3 Google ROAS rose about 42% year over year on roughly 37% less daily spend, while two other retailers gave about a quarter back. ChatGPT opened self-serve ad buying in May.
Happy Cog's take
Feed the automation clean data, then check its work. Give assistant ad placements a small, dated test budget in Q1 and a clear rule for when to stop.
09 · Retention and first-party data
The post-purchase flow is now the front door
Roughly 60% of DTC revenue comes from returning customers (Swell). An order placed through an assistant's native checkout can arrive without the email opt-in, the account, or the browsing history that fed every lifecycle program built over the past decade. Loyalty that can be linked inside an agent's checkout is one of the few reasons an agent will choose your store.
Happy Cog's take
Build the capture before you need it. As a Klaviyo Partner, we set up post-purchase opt-in flows, tag orders by origin, and report repeat rate for assistant-originated customers separately.
11 · The agenda · Interactive
What should a retailer prioritize in Q4 and Q1?
Five things before the holiday code freeze, and five in January. Your progress is saved in this browser only.
0 / 10
Priorities underway
Before the holiday code freeze
In January
12 · Scored in next year's edition
Ten calls for 2027, specific enough to grade
A prediction nobody could disagree with is not worth printing, so some of these are uncomfortable.
-
01 Moderate confidence
Google makes native checkout self-serve.
US merchants will be able to enable checkout through Google's protocol from Merchant Center, without an interest form or approval queue, by June 30, 2027.
-
02 Moderate confidence
The protocols settle into two.
OpenAI's protocol either formally interoperates with Google's or remains limited to a small set of large retailers through the end of 2027, and no third standard gains broad merchant adoption.
-
03 Moderate confidence
AI traffic stays small and converts well.
AI-referred sessions remain below 5% of total sessions for most mid-market brands through 2027, while converting at a materially higher rate than average organic traffic. Retail is where this call faces its hardest test, since today the premium is clear in repeat purchases and still building in considered ones.
-
04 High confidence
The one-in-five holiday forecast won't appear in retailers' dashboards.
Measured assistant sessions will account for well under 5% of holiday 2026 visits in most mid-market retailers' own analytics, and the gap between industry estimates and first-party data becomes the story of January.
-
05 Moderate confidence
A well-known DTC brand says no to native checkout.
At least one recognized DTC brand publicly declines native agentic checkout in favor of its own, citing the customer relationship, before the end of 2027.
-
06 Moderate confidence
Assistant ads become a standard line.
Sponsored placements inside AI shopping assistants appear as a standing line item in most mid-market retail media plans by Q4 2027, with measurement still the main complaint.
-
07 Lower confidence, high consequence
Agent orders come back more often.
Return rates on orders placed through native agentic checkout run measurably higher than on-site orders for apparel and fit-sensitive categories, and the difference becomes a margin conversation in board reporting.
-
08 Lower confidence
Amazon opens a door.
Amazon offers a sanctioned path for at least one outside AI agent to transact on its marketplace by the end of 2027.
-
09 Moderate confidence
Email and SMS platforms build for the agent-originated customer.
At least two major email and SMS platforms ship flows or triggers designed specifically for orders that originated in an AI agent.
-
10 Moderate confidence
Blended conversion rate loses its place on the scorecard.
Revenue per visitor and contribution margin replace blended conversion rate as the headline efficiency measure in most growth-stage retail board reporting, as assistant and upper-funnel traffic keep diluting the blended rate.
Want to see how these hold up?
Tick "Tell me when these are graded" on the form at the foot of the page and we will send you next year's scorecard when it publishes.
Questions
Frequently asked
What is Google's Universal Cart?
Universal Cart, announced by Google in May 2026, gathers products from many retailers across Search and the Gemini app and watches prices and stock on the shopper's behalf. It checks out through Google Pay or hands the cart to the merchant's own site. The retailer stays the merchant of record. As of late September 2026, Google had not confirmed a broad US launch.
Did ChatGPT Instant Checkout work for retailers?
Not at the scale promised. OpenAI and Shopify said in September 2025 that millions of merchants would soon offer Instant Checkout inside ChatGPT, and roughly a dozen to 30 went live. On March 24, 2026, OpenAI said merchants would use their own checkout while ChatGPT focused on discovery. Walmart reported that purchases completed inside ChatGPT converted at one-third the rate of those that clicked through to its site.
How much traffic do AI assistants send to online retailers?
Less than the headlines suggest. Across five retail GA4 properties Happy Cog manages, AI-assistant sessions reached 0.54% of total sessions in Q3 2026, up from 0.25%, with monthly volume growing 2.3 times. Adobe measured AI traffic to US retail sites up 393% year over year in Q1 2026 and notes the base remains modest.
Do shoppers from AI assistants convert better?
In repeat purchases, clearly. Adobe reported a 31% conversion premium for AI-referred shoppers during the 2025 holidays. At a replenishment retailer Happy Cog manages, AI-referred sessions converted at 10.1% in Q3 2026 against 5.1% for organic search. In considered purchases such as gifts and eyewear frames, assistant visitors have not yet matched the organic rate on a last-click basis.
Will 20% of holiday traffic really come from AI agents?
Salesforce predicts 20% of 2026 holiday e-commerce traffic will come from AI agents, a figure that includes shopping assistants, background agents, and competitors' price-scraping bots. Happy Cog expects measured assistant sessions to account for well under 5% of holiday 2026 visits in most mid-market retailers' own analytics. We rate that call high confidence.
Should a retailer enable native agentic checkout?
Decide category by category. Native checkout suits replenishment and low-consideration items, where the shopper wants speed and the cross-sell is small. It is a harder call for gifting, high-AOV, and fit-sensitive products, where on-site moments carry the margin. Whichever you choose, keep a first-party capture step after every agent-originated order.
Which pages do AI assistants send shoppers to?
At the largest multi-brand retailer Happy Cog manages, ChatGPT sent visitors to 1,442 distinct pages between January 1 and September 21, 2026. The homepage took 37% of those visits, category and brand pages about 14%, product pages about 10%, and how-to articles about 5%. Store-level facts such as shipping, returns, and authenticity therefore matter as much as product detail.
What should an online retailer prioritize before the 2026 holiday code freeze?
Clean the product feed so prices, stock, shipping cutoffs, and return windows are current everywhere it syndicates. Check eligibility for Google's native checkout, move purchase tracking server-side, confirm GA4's AI Assistant channel captures every assistant source, and put budget caps and a daily review on Performance Max through the promotional window.
Start a conversation
The best stores start with a good conversation.
If you're rethinking how your store gets found by people and by the agents they send, we'd like to help you plan 2027. We run the storefront, the product data, and the customer relationship as one team.
The full edition runs 26 pages, with the storefront, commerce technology, and retention chapters, the portfolio data, and the methodology behind every figure. No form required.