Ask ChatGPT for a good pair of running shoes under $150 and it will give you a shortlist, with brands, prices, and reasons. Ask Perplexity which standing desk to buy and it will compare three of them. A growing slice of buying decisions now starts inside an AI assistant rather than a search results page, and the shopper often never sees the ten blue links you spent years optimizing for.
The numbers behind that shift are no longer small. ChatGPT referral traffic to publishers and retailers grew 206% year over year from January 2025 to January 2026, and those referrals convert at 14.2%, against 2.8% for conventional organic search (HiGoodie, AI Search Traffic Report 2026). Google's AI Overviews now show up on roughly 48% of queries, up from about 13% a year earlier. EMARKETER forecasts that 31.3% of the US population will use generative AI search in 2026.
So the question for any store owner is blunt: when an AI is asked what to buy, does it name your product or a competitor's? This guide explains what actually decides that answer in 2026, and what you can do about it this week.
Search has split into answers and agents
Two things are happening at once, and it helps to keep them separate.
The first is discovery. Assistants read the web and structured product data, then recommend specific items. This is where the real volume and the real opportunity sit today, and it is stable enough to optimize for.
The second is transaction, where the assistant completes the purchase for you. This is the louder, flashier story, and also the messier one. OpenAI launched Instant Checkout on February 16, 2026 through its Agentic Commerce Protocol (ACP), built with Stripe. By early March 2026 it had pulled back and revamped the experience, with only around 30 merchants live and complaints about scraped pricing and inventory that did not match reality (CNBC, March 24, 2026). In-chat checkout will matter, but it is early. Do not rebuild your store around it yet.
Meanwhile Google moved on the same week the new year started. On January 11, 2026 it announced the Universal Commerce Protocol (UCP), an open standard for product discovery and checkout inside Google AI Mode and the Gemini app, co-developed with Shopify, Etsy, Walmart, Target and more than twenty other partners (Google, January 2026). The protocols are still settling. The discovery mechanics underneath them are not, and that is the part you can win.
Your Google Merchant Center feed is your AI shopping visibility
Here is the most useful finding of the year, and the one most store owners have not heard. In March 2026, researchers analyzed 43,000 products across ten verticals and found that 83% of ChatGPT's product recommendations originate from Google Shopping and Google Merchant Center data. They confirmed it by finding base64-encoded Google Shopping parameters inside ChatGPT's own source code (reported by Search Engine Land; original analysis by Passionfruit).
Read that again, because it reframes the whole problem. You do not need a separate, mysterious "ChatGPT strategy." For the large majority of product recommendations, the feed you already maintain for Google Shopping is the feed ChatGPT is reading. If that feed is thin, stale, or missing fields, you are invisible in both places at once. If it is complete and accurate, you show up in both.
That is why we treat the Merchant Center feed as the foundation of ecommerce SEO for AI, not as a paid-media afterthought. The same data structure that powers your Shopping campaigns is now doing double duty as your organic AI shopping presence.
The four signals that decide whether an AI recommends your product
Across ChatGPT, Perplexity and Google AI Mode, the same handful of signals come up again and again. Get these right and you are doing most of the work.
1. Server-rendered structured data
Mark up every product page with Product and Offer schema in JSON-LD (JavaScript Object Notation for Linked Data, the format Google and the assistants prefer). The Offer must carry a price and an ISO 4217 currency code such as USD or GBP. Add AggregateRating and Review schema where you have genuine reviews. The critical detail: render this in the server HTML, not by injecting it with client-side JavaScript after the page loads. Many AI crawlers do not execute your scripts, so JavaScript-only schema is schema they never see.
2. A complete feed, not a minimal one
The feeds that win carry far more than title, price and image. Include product_review_count and average_rating, return_policy, shipping_speed and a free_shipping_indicator where it applies, plus GTIN (Global Trade Item Number) and brand. Assistants use these fields to answer the follow-up questions a shopper actually asks, things like "which of these ships free" or "which has the best reviews." A product the assistant can fully describe is a product the assistant is comfortable recommending.
3. Reviews and off-site reputation
AI answers lean on what the wider web says about you, not only what your own site claims. Reviews on your product pages help, but so do mentions on third-party review sites, roundups, and forums. A product with a real reputation footprint gets recommended more often than an identical product nobody talks about.
4. Real-time, consistent price and availability
This is where OpenAI got burned, and where you can quietly lose. If the price an assistant reads is wrong, or your "in stock" badge contradicts your feed, you are a liability the assistant learns to route around. Keep price and stock synced across your site, your schema and your Merchant Center feed. Contradiction is worse than absence.
How the tactics differ by platform
The signals above are shared, but each engine has its own access points worth knowing.
ChatGPT. Feed quality is the lever, given the 83% pipeline finding. Confirm OAI-SearchBot is allowed in your robots.txt, and keep your Merchant Center data clean and current. There is no secret beyond doing the feed properly.
Perplexity. Perplexity runs a free merchant program that lets retailers submit product information directly. It is no-cost to join and an easy win for stores that want presence in Perplexity's shopping answers. Allow PerplexityBot in robots.txt as well.
Google AI Mode and Gemini. Readiness here flows through the Universal Commerce Protocol, which leans on Shopify and Merchant Center. If you are on Shopify with a healthy feed, you are largely set up for it already. If you run a custom storefront, the work is making sure your product data reaches Merchant Center cleanly. This is a natural extension of an ecommerce marketing program rather than a separate project, and it is where a Shopify-focused build, the kind covered on our Shopify development page, pays off in cleaner structured data from the start.
The new KPI is citation share, not rank
You cannot manage what you do not measure, and the old metric does not fit. When the output is a paragraph naming three products, "what position did I rank" is the wrong question. The metric that matters is citation share, often called Share of Model: how often an assistant names your brand or product for the queries you care about, and how that compares with rivals.
The opportunity is in how few people are doing this. Conductor reported in 2026 that only about 14% of marketers measure AI search performance at all. That gap is your head start. Tools in this category include Profound, Semrush and Conductor; the point is less which one you pick and more that you start tracking citation share now while most competitors are flying blind. Profound's position gives a sense of the money flowing in: it raised a $96 million Series C at a $1 billion valuation in February 2026, backed by Lightspeed, Sequoia and Kleiner Perkins. McKinsey projects agentic commerce will drive $3 to $5 trillion globally by 2030.
Because so much of AI shopping visibility runs on the same feed as your paid Shopping campaigns, organic AI presence and performance marketing are no longer separate disciplines. The feed that wins you a free ChatGPT recommendation is the feed that makes your paid Shopping ads cheaper and more relevant. Measuring them together is how a full-funnel marketing program should work in 2026.
A practical checklist you can start today
None of this requires waiting for the protocols to settle. Work through this list and you will already be ahead of most stores in your category.
- Audit your Merchant Center feed. Check coverage of average_rating, product_review_count, return_policy, shipping_speed, GTIN and brand. Fill every gap you can.
- Audit your schema. Confirm Product, Offer (with ISO 4217 currency), AggregateRating and Review are present and server-rendered, not JavaScript-injected.
- Check robots.txt. Make sure OAI-SearchBot and PerplexityBot are allowed, not blocked by a blanket rule.
- Build review velocity. A steady stream of genuine reviews beats a one-time burst, on your site and on third-party platforms.
- Sync your prices and stock. Reconcile site, schema and feed so an assistant never reads a contradiction.
- Join Perplexity's free merchant program. Low effort, direct submission, real upside.
- Start measuring citation share. Pick a tool, track your branded and category queries, and watch the trend.
Where this leaves you
The discovery layer of AI shopping is no longer speculative. The pipeline from your product feed to a ChatGPT recommendation has been documented, and the signals that move it are known. Yet only about 14% of stores are measuring any of it, so there is real room to get ahead before the rest catch up.
The catch is that the work is technical and cross-channel: feed engineering, schema, crawler access, review systems, price sync, and a measurement layer that did not exist two years ago. Most stores will lag simply because no one owns the whole picture. That is the gap we close. DigiRocket sets up AI shopping optimization end to end, from a clean Merchant Center feed and server-rendered schema to citation-share tracking, as part of a connected ecommerce SEO and performance marketing program. If you would rather be recommended than overlooked, that is where to start.
Frequently asked questions
What is AI shopping optimization?
AI shopping optimization is the practice of structuring your product data, schema, reviews and pricing so that AI assistants like ChatGPT, Perplexity and Google's Gemini recommend your products when a shopper asks them what to buy. It overlaps heavily with Generative Engine Optimization (GEO), but focuses specifically on commerce signals such as the Google Merchant Center feed, Product and Offer schema, and real-time price and availability.
Does my Google Merchant Center feed affect what ChatGPT recommends?
Yes. A March 2026 analysis of 43,000 carousel products across ten verticals, reported by Search Engine Land, found that 83% of ChatGPT product recommendations trace back to Google Shopping and Google Merchant Center data. Researchers found base64-encoded Google Shopping parameters inside ChatGPT's source code, confirming the pipeline. In practice, a clean, complete Merchant Center feed is the single biggest lever you have over your visibility inside ChatGPT shopping.
Is OpenAI's Instant Checkout live for my store?
Probably not yet. OpenAI launched Instant Checkout on February 16, 2026 through its Agentic Commerce Protocol (ACP) with Stripe, but only around 30 merchants were live. In early March 2026 OpenAI revamped the shopping experience after problems with scraped pricing and inventory data. Treat in-chat checkout as early and in flux. The discovery side, getting recommended in the first place, is what is stable and worth optimizing today.
What is Google's Universal Commerce Protocol (UCP)?
The Universal Commerce Protocol (UCP) is an open standard Google announced on January 11, 2026 to power product discovery and checkout inside Google AI Mode and the Gemini app. It was co-developed with Shopify, Etsy, Walmart, Target and more than 20 other partners. If you sell on Shopify or keep a healthy Merchant Center feed, you are already most of the way to being UCP-ready.
How do I measure whether AI search is sending me sales?
Rank position is the wrong KPI here. The new metric is citation share, sometimes called Share of Model: how often your brand or product is named in AI answers for the queries you care about. Tools such as Profound, Semrush and Conductor track this. Conductor reported in 2026 that only about 14% of marketers actually measure AI search performance, so building the habit early is a genuine first-mover advantage.
Should I let AI crawlers access my site?
If you want to be recommended, yes. Allow OAI-SearchBot (ChatGPT's search crawler) and PerplexityBot in your robots.txt, keep your product schema server-rendered rather than injected by client-side JavaScript, and make sure prices and stock are accurate. Blocking these bots, or hiding your structured data behind JavaScript an assistant cannot execute, is the most common way stores quietly remove themselves from AI recommendations.