ChatGPT Has 900M Weekly Users — Can AI Shoppers Find Your WooCommerce Store?
AI-referred traffic to US retail sites grew 393 percent year over year in Q1 2026 and, in March, converted 42 percent better than non-AI traffic — a complete reversal from twelve months earlier. Getting found by AI shoppers is a machine-readability problem: Adobe found 34 percent of retailer homepage content is invisible to AI models. Seeing the resulting sales is a separate problem, because AI-referred sessions frequently arrive with no usable referrer, which only server-side capture resolves.
- What Actually Changed in Twelve Months
- The Reversal Nobody Predicted: Discover in AI, Buy on Your Site
- Part One — Getting Found: What AI Can and Can’t Read
- Part Two — Seeing the Sale: Why AI Sessions Arrive Half-Anonymous
- The Protocol Question: ACP, UCP and Small Stores
- A Ninety-Day Order of Operations
- Key Takeaways
- Frequently Asked Questions
What Actually Changed in Twelve Months
AI referral traffic went from the worst-converting channel in US retail to the best, and the volume grew almost fivefold.
In March 2025, traffic arriving at US retail sites from AI assistants converted 38 percent worse than everything else. The consensus read was comfortable: AI sends browsers, not buyers.
Twelve months later Adobe Analytics reported the opposite. In March 2026, AI-referred visitors converted 42 percent better than non-AI traffic — a record, drawn from more than one trillion visits to US retail sites. Same channel, same stores, inverted sign.
The volume moved too. AI-referred traffic to US retailers grew 393 percent year over year in Q1 2026, peaking at 1,151 percent year over year in December. By May 2026 it was up 138 percent against the prior May, and had grown 1,324 percent since Adobe began tracking the category in October 2024.
In March 2026 AI-referred retail visitors converted 42 percent better than non-AI traffic, reversing a 38 percent deficit recorded twelve months earlier.
Behind it sits an audience that got large fast. ChatGPT reached 900 million weekly active users in February 2026, roughly double the 700 million OpenAI reported in September 2025. Adobe’s consumer survey found 39 percent of US respondents had used AI for online shopping, and 85 percent of those said it improved the experience.
The engagement numbers explain the conversion lift. AI-referred visitors spent 48 percent more time on page, viewed 13 percent more pages, and generated 37 percent more revenue per visit. They arrive further along, having already had the comparison conversation somewhere else.
The Reversal Nobody Predicted: Discover in AI, Buy on Your Site
In-chat checkout stalled in early 2026, which pushed the transaction back onto merchant sites — and made your own measurement the deciding factor.
The 2025 story was that AI would swallow the checkout. OpenAI launched Instant Checkout inside ChatGPT in September 2025 on the Agentic Commerce Protocol, co-developed with Stripe. Walmart signed up in October. Target followed.
Then it didn’t hold. Industry reporting in March 2026 described OpenAI scaling back its shopping ambitions, and the reason given was arithmetic rather than philosophy: checkout inside the chat reportedly converted around three times worse than sending the shopper through to the retailer’s own site, even though AI drove a materially higher new-customer rate.
The question isn’t whether AI will replace your checkout. The question is whether your store is readable enough to get recommended, and instrumented enough to prove what happened when the visitor arrives.
You may be interested in: Instant Checkout got scaled back — feed readiness is the move
That reversal is good news for independent WooCommerce stores. A marketplace-mediated checkout favours whoever has scale and a commission relationship. Discovery in AI followed by purchase on your own domain favours whoever has the clearest product data and the best measurement — both of which a small store can actually fix.
Part One — Getting Found: What AI Can and Can’t Read
AI models recommend what they can parse, and Adobe found roughly a third of retailer homepage content is not machine-readable.
Language models don’t browse the way people do. They ingest text and structured markup. Anything that only exists after JavaScript runs, or only exists inside an image, is a coin toss.
Adobe measured this directly: 34 percent of retailer homepage content is not machine-readable to AI models, and product pages average only 66 percent visibility. A third of what you wrote about your own products isn’t reaching the systems now driving your best-converting channel.
For a WooCommerce store, the practical checklist is short and unglamorous.
- Server-render the facts that matter. Price, stock status, variant options, dimensions, materials, shipping and returns should be in the HTML source, not assembled client-side in a tab.
- Keep structured data honest. Product, Offer and AggregateRating markup that contradicts the visible page is worse than none — it teaches the model your data is unreliable.
- Answer the comparison question on the page. AI assistants are asked to compare. Specifications, sizing guidance and use-case detail give a model something to cite instead of guessing.
- Don’t hide policy in a PDF. Shipping timelines and return windows are decision-critical and frequently unreadable.
None of this is new SEO work rebranded. It’s the same instinct — make the important facts legible — applied to a reader that can’t click.
Part Two — Seeing the Sale: Why AI Sessions Arrive Half-Anonymous
Many AI-referred visits land with no usable referrer, so the revenue gets credited to direct traffic and the channel looks like it isn’t working.
Here’s where most stores lose the thread. Even when an AI assistant sends a high-intent buyer, the session often arrives without the evidence needed to attribute it.
Several things cause it. Some assistants omit or strip the referrer. Some route through a redirect that drops it. App-based assistants commonly open an in-app browser that discards query parameters before your page loads. The result is a purchase filed under direct or unassigned.
Adobe found 34 percent of retailer homepage content is not machine-readable to AI models, while product pages average only 66 percent visibility.
So the fastest-growing, best-converting channel in US retail is also the one your analytics is least equipped to see. Stores conclude AI referrals don’t convert, when what actually happened is that the referral evidence never survived the trip.
You may be interested in: Your GA4 attributed zero AI-influenced buyers — they still came
Server-side capture closes the gap because it records the request at your infrastructure rather than asking a browser to report on itself. The user agent, landing path, timing and session sequence are all available at the server, before any tag fires or fails. That’s enough to classify AI-assistant traffic as its own cohort and watch it properly.
| Metric (AI-referred vs non-AI, US retail) | March 2025 | March 2026 | May 2026 |
|---|---|---|---|
| Conversion rate | 38% worse | 42% better | 54% higher |
| Revenue per visit | Below non-AI | 37% higher | 53% higher |
| Time spent on site | Not reported | 48% more | 53% more |
| Pages per visit | Not reported | 13% more | 23% more |
| Traffic growth (YoY) | Baseline | +393% (Q1) | +138% |
Read that table as a trend line, not a promise. It’s Adobe’s own client base in US retail, and it is not independently audited — which is precisely why the number that should drive your decisions is the one measured on your own store.
The Protocol Question: ACP, UCP and Small Stores
Five protocol families now compete for agentic commerce, and for most WooCommerce stores that is a reason to wait rather than to build.
The agentic commerce space currently holds several competing standards — the Agentic Commerce Protocol from OpenAI and Stripe, Google’s payments-oriented work, and card-network entries from Visa and Mastercard. All are moving.
The temptation is to pick one and integrate. For a store below enterprise scale that’s usually the wrong sequencing. Protocol integration is speculative work against a moving specification, while machine-readable product data and reliable server-side measurement pay off in every channel you already have.
Structured product data and first-party event capture are prerequisites for every protocol on that list. Do the prerequisite work and you’re ready for whichever standard wins, without having bet on it.
The same logic applies to measurement generally: durability comes from owning the collection layer rather than tracking each platform’s roadmap. That argument is made at length in our piece on why first-party server-side is the only durable signal after the Privacy Sandbox retirement.
A Ninety-Day Order of Operations
Three sequential steps: make the catalogue readable, capture sessions server-side, then segment the AI cohort and measure it.
Sequence matters here, because step three is meaningless without step two.
Weeks one to three — readability. Load your top twenty product pages with JavaScript disabled. Whatever vanishes is invisible to AI models. Move price, availability and specifications into server-rendered HTML and validate your Product markup against the visible page.
Weeks four to eight — server-side capture. Record sessions and conversions at your own infrastructure rather than depending on browser tags. This is the job the Transmute Engine™ handles for WooCommerce stores: it captures the event server-side and lands it in your own BigQuery instance, so the record exists whether or not a tag fired.
Weeks nine to twelve — segment and measure. Classify AI-assistant sessions by user agent and landing pattern, then compare conversion rate and revenue per visit against your other channels. You’ll get a number specific to your catalogue, which is worth more than any industry average.
At the end of that you can answer a question most store owners currently can’t: how much revenue did AI-mediated discovery actually produce last quarter?
Key Takeaways
The short version, for anyone deciding what to prioritise this quarter.
- The channel inverted: AI-referred US retail traffic went from converting 38 percent worse in March 2025 to 42 percent better in March 2026.
- Volume is compounding: AI referrals grew 393 percent year over year in Q1 2026 and 1,324 percent since October 2024.
- A third of your content is unreadable: Adobe measured 34 percent of retailer homepage content as not machine-readable, with product pages at 66 percent visibility.
- Checkout came back to your site: in-chat checkout was scaled back in early 2026 after converting roughly three times worse than click-throughs to the merchant.
- Attribution is the bottleneck: many AI-referred sessions arrive with no usable referrer and get filed as direct traffic.
- Prerequisites beat protocols: structured product data and server-side capture are required by every agentic standard, so they are the safe investment.
Frequently Asked Questions
Common questions about AI-driven discovery and measuring AI-referred sales on WooCommerce.
They read what they can parse. Product data expressed in server-rendered HTML and valid structured markup — price, availability, specifications, shipping, returns — is usable; the same facts locked inside JavaScript-rendered tabs or images often are not. Adobe found 34 percent of retailer homepage content is not machine-readable to AI models, with product pages averaging 66 percent visibility.
Aggregate data says yes and improving. Adobe reported AI traffic converting 42 percent better than non-AI in March 2026, rising to 54 percent higher by May 2026. That is aggregate US retail data from Adobe’s own client base, so treat it as direction rather than a guarantee for your store — which is why measuring your own AI-referred cohort matters.
Because many AI-referred sessions arrive without a usable referrer. Some assistants strip or omit it, some send users through a redirect, and app-based assistants often open an in-app browser that discards parameters. The visit lands as direct or unassigned, so the revenue gets credited to the wrong place even though the session happened.
For most small stores, not yet as a priority. OpenAI scaled back in-chat checkout in early 2026 after conversion inside the chat underperformed click-throughs to the merchant’s own site. The durable pattern is discovery in AI and purchase on your site, which makes machine-readable product data and reliable server-side measurement the higher-return work.
Check whether your product facts survive with JavaScript disabled. If price, stock status and specifications disappear, AI models are guessing about your catalogue. Fix that, then add server-side session capture so you can identify AI-referred visits by user agent and landing pattern rather than relying on a referrer that may never arrive.
References
Sources used for the figures and dates above.
- Adobe, “AI traffic grows but retail sites lag in AI search visibility” — conversion, engagement and machine-readability findings (2026) — business.adobe.com
- Adobe Analytics Q1 2026 AI traffic data, reported by Yahoo Finance (April 2026) — finance.yahoo.com
- Digital Commerce 360, “Adobe: AI-referred traffic to retail sites doubles in a year” (June 2026) — digitalcommerce360.com
- Elogic Commerce, ChatGPT commerce and agentic shopping statistics, citing OpenAI’s weekly active user disclosure (2026) — elogic.co
- Digital Applied, analysis of the in-chat checkout retreat and reported Walmart conversion comparison (2026) — digitalapplied.com
- Digital Commerce 360, “OpenAI expands agentic commerce push” on the Agentic Commerce Protocol (February 2026) — digitalcommerce360.com
If you want to know what AI-mediated discovery is worth on your own catalogue rather than in aggregate, that starts with measuring it — see how Seresa does it.