The most instructive thing that happened in AI commerce this year wasn’t a launch. It was a retreat.
In March 2026, OpenAI abandoned the idea of completing purchases inside ChatGPT. Not because it couldn’t build it — because, as Search Engine Land reported, shoppers research products in the chat and then prefer to finish the purchase on the merchant’s own site.
OpenAI’s merchant documentation confirms the direction: away from a standalone Instant Checkout, toward better shopping discovery and checkout on the merchant side. It also states there are no commissions on purchases that begin in ChatGPT.
That reversal tells you exactly where to spend effort. AI is becoming the layer that decides what a buyer considers. Not the layer that takes their money.
Which is a much bigger deal than a checkout button.
What actually exists right now
ChatGPT Shopping is available to all users in the U.S., with Etsy and more than a million Shopify merchantsconnected.
Google has launched agentic checkout inside AI Mode with Wayfair, Chewy and Etsy.
So the infrastructure exists on both sides. But the behaviour underneath it is consistent: people use the assistant to narrow the field, then leave to buy.
Why the shortlist is the whole game
Think about what an assistant actually returns when someone asks for a product recommendation.
Not a page of results. Not twenty options with filters. Three, maybe four suggestions, with reasoning attached.
Every traditional discovery surface gave you a long tail to live in. Page two of Google still received traffic. Position eleven in a marketplace still made sales. A shopper scrolling a category page passed dozens of products they hadn’t planned to see.
A three-item shortlist has no long tail. You are in it or you are invisible — and the buyer never learns you existed to compare you.
That’s the shift. Ranking was a spectrum. Shortlisting is binary.
What determines whether you’re included
The uncomfortable answer is that a lot of it comes down to whether a machine can read your product data accurately.
Feed quality. Clean, complete, current. Missing attributes, stale pricing and inconsistent naming don’t just degrade the entry — they make the model uncertain, and uncertainty is a reason to recommend something else.
Structured descriptions. Specifics, expressed plainly. Dimensions, materials, compatibility, use cases, what it isn’t for. Marketing language that avoids stating facts gives a retrieval system nothing to match against.
Price accuracy across surfaces. If your feed, your site and third-party listings disagree, you look unreliable to a system whose job is to not embarrass itself.
Third-party corroboration. Reviews, comparisons, coverage, forum discussion. An assistant assembling a recommendation draws on what the wider web says about a product, not only what the seller says.
Availability data. Recommending something out of stock is a failure state. Systems learn to avoid sources that produce them.
None of this is glamorous, and none of it has traditionally sat with marketing. That’s precisely why it’s an opportunity — your competitors have it filed under operations.
The audit that costs nothing
Before changing anything, find out where you stand.
Take the ten questions a real customer asks before buying in your category. Not branded searches — the actual decision questions. «What’s the best X for Y?» «X versus Z?» «Affordable X that does Y?»
Run them through ChatGPT and Google’s AI Mode. Record what comes back:
- Are you named at all?
- Which competitors are, and how are they described?
- Is what’s said about you accurate?
- What criteria is the model reasoning from?
That last one is the valuable part. The assistant will often reveal the dimensions it considers decisive in your category — and they may not be the ones your product page leads with.
Most brands have never run this exercise. It takes twenty minutes and reframes the entire product-content brief.
Which assistant to prioritise
If you’re going to optimise for one, the referral data is unambiguous.
The Previsible «2026 State of AI Discovery» report, published via Search Engine Land, analysed 6.77 million LLM sessions across 166 sites over 19 months. ChatGPT accounts for 92.4% of tracked standalone LLM referral traffic, up from around 84% in December 2025.
But the same report notes that AI discovery inside Google — AI Overviews plus AI Mode — drives more AI-influenced traffic than all standalone LLMs combined.
So: Google’s AI surfaces first, ChatGPT second, everything else later. (Worth noting the report’s author is a Previsible co-founder, with the conflict of interest disclosed.)
The measurement problem, stated honestly
Here is the part nobody has solved.
When an assistant shortlists your product and the buyer arrives via a branded search two days later, your analytics record a branded search. The AI recommendation that caused it is invisible.
Practical partial answers:
Watch branded search volume as an outcome metric. Rising branded demand without a corresponding campaign is a signal something upstream is recommending you.
Track referral traffic from assistant domains separately. It undercounts badly, but the trend line is still information.
Ask buyers. A single «how did you first hear about us?» field on checkout will outperform your attribution stack for this specific question.
Re-run the shortlist audit quarterly. Presence in the shortlist is itself the metric. Measure it directly rather than trying to infer it from traffic.
The caveats
This is largely U.S.-centric today. ChatGPT Shopping’s full functionality and the agentic checkout integrations are concentrated in the U.S. market. For other regions this is a preparation window, not a live channel.
Commission-free won’t necessarily last. OpenAI states there are no commissions on purchases originating in ChatGPT. That’s a current commercial position, not a permanent property of the channel. Build on the assumption that the economics may change.
Nobody has a reliable playbook yet. The practices above follow from how retrieval systems work and from what the platforms have published. There is no equivalent of a decade of SEO case studies. Anyone claiming certainty here is selling something.
If you don’t sell products
The same logic applies with different objects.
A consultant, an agency, a clinic, a restaurant — all get shortlisted by the same mechanism, drawing on the same inputs: structured, accurate, corroborated information about what you do, who it’s for, and what makes you a fit.
The feed is different. The question the buyer asks the machine is identical.
About the author
Alina Palii — brand strategist, founder of ALPA Marketing.
She works with founders and leadership teams on the decisions that come before the marketing: what the brand stands for, who it is genuinely for, what it declines to be, and how that translates into everything the market eventually sees. Strategy first — the content, the channels and the campaigns follow from it.
10+ years in marketing and a master’s degree in the field. She has built brands from zero for AI startups and national companies, and shaped the positioning of personal brands whose audiences buy on trust rather than on price.
Ukrainian by origin, living between Dubai, Paris and Ukraine, and working across the UAE and European markets — a vantage point that matters when a brand has to hold its meaning across cultures rather than be rebuilt in each new one.
She works across categories rather than inside one. Positioning logic travels between industries even when the audience doesn’t, and the pattern recognition that comes from moving between them is often what a category-blind competitor is missing.
Alina takes on a limited number of strategy engagements at a time.
