How generative AI is changing product discovery

Shoppers now ask instead of search

Instead of typing "black ankle boots size 7" and scanning a page of links, a shopper asks an assistant: "I need comfortable boots for Galway in November that work for the office and weekend walks, around €120." The assistant reads reviews, prices, stock and style signals, then suggests three or four specific products with its reasons.

That is not a results page. It is closer to a personal shopper, and it changes what a shop's product data has to carry.

The channel is young but no longer theoretical. Adobe, which measures traffic across US retail sites, reported that during the 2025 holiday season, from 1 November to 31 December, traffic to retail sites from generative AI tools, counted as shoppers clicking a link to a retail site, rose 693.4% on the year before. Adobe adds the caveat that matters: "the base of users remains modest".¹ A year earlier it had measured a 1,300% rise for the 2024 season.² Big percentages on a small base, in the US. Useful as a direction, not as a forecast for a shop in Limerick.

The earlier version of this article quoted a string of further Adobe figures: a 3,300% rise on Prime Day 2025, 4,700% by July 2025, and AI referred shoppers converting 31% better, staying 45% longer and viewing 13% more pages. I could not reach a page from Adobe itself that confirms those figures, so they are gone.

This is what I tell every eCommerce client right now: the way shoppers find your products is changing, and your product data has to change with it. This article is about that data. How AI answers pick their sources in general, and what Search features look like around them, I cover elsewhere and link to below rather than repeat.

Where these answers appear for an Irish shopper

In Ireland, the AI answers that matter most sit inside Google. Google brought AI Overviews, a generated summary above the ordinary results, to Ireland on 25 March 2025, in English, as part of a wider European rollout that also covered Austria, Belgium, Germany, Italy, Poland, Portugal, Spain and Switzerland.³ AI Mode, a separate tab where the whole page is a generated answer with links and follow up questions, came later. Google announced it for more than 40 new countries and territories on 7 October 2025 and lists Ireland among the places it is available,⁴ and it went live here on 8 October 2025, on the results page and in the Google app for Android and iOS.⁵ What that does to clicks is in the article on AI Overviews and organic traffic.

Outside Google, ChatGPT and Perplexity both search the web and link to what they used. OpenAI runs a crawler called OAI-SearchBot to surface websites in ChatGPT's search features, separately from GPTBot, which gathers training data, and a site that blocks OAI-SearchBot will not be shown in ChatGPT's search answers.⁶ Perplexity and PayPal launched Instant Buy on 25 November 2025, letting US users pay inside Perplexity's chat while the retailer stays merchant of record.⁷ The earlier version described ChatGPT's shopping features in detail. OpenAI's own pages refused my requests, so I have left that description out rather than repeat it unchecked.

Each assistant works differently, and nobody outside them knows exactly how they rank, which I go into in how to rank in ChatGPT. What they share is simpler: they can only recommend a product they can read about.

An example: Ballintemple Threads is an invented fashion shop in Cork at ballintemplethreads.example. A shopper asks Google's AI Mode for "a relaxed black blazer for a smart casual office, under €50, delivered by Thursday". For Ballintemple Threads to be one of the products named, its blazer page has to be indexed and allowed a snippet, its price and stock have to be readable and current, and somewhere the page has to say "relaxed fit", "smart casual" and when it arrives. If all it says is "Blazer, black, €39.99, sizes 6 to 22", the assistant has nothing to match against most of the question.

The question describes a person, so the data must too

A keyword search matches words. A conversational question describes a person, a context, constraints and values. Compare:

• "womens black ankle boots"
• "comfortable black ankle boots for a teacher on her feet all day, waterproof if possible, about €80, from a brand that takes sustainability seriously"

The first needs a product name and a category. The second needs comfort, use, waterproofing, price and the brand's materials, all written down somewhere a machine can read. Most product feeds carry name, price, category, colour and size. That is enough for a Shopping advert. It is not enough for an assistant matching the second question.

Worked example: one blazer, two listings

The thin listing:

Name: Oversized single breasted blazer. Price: €39.99. Category: Women, jackets, blazers. Colour: black. Sizes: 6 to 22.

The rich listing adds what a good shop assistant would say out loud:

Style: oversized, relaxed fit, dropped shoulders. Occasions: smart casual office, weekend. Styling: over a T shirt or a roll neck, with wide leg trousers or jeans. Material: recycled polyester blend, lightweight, unlined. Fit: true to size in the body, deliberately roomy in the shoulders, size down for a sharper line. Delivery: next day if ordered by 10pm. Reviews: 4.2 stars from 847 reviews, customers mention comfort and that it runs slightly large.

Every number in that listing is the example's, and every claim in it has to be true. The second version answers the blazer question above: relaxed, smart casual, under €50, by Thursday. The first answers almost none of it. Write the richer text on the product page itself, where shoppers read it too, not only in a feed.

Premium shops need their editorial knowledge on the product page

A premium shop gets harder questions. "A birthday present for my wife, she likes understated luxury and finds the big logo brands too obvious. Who should I look at?" That needs taste, positioning and context, and a good premium shop already has that knowledge. The problem is where it lives: in buyer's notes, newsletters and lookbooks, not on the product pages an assistant reads.

An example: Hightown and Co is an invented premium gift boutique in Kilkenny at hightownandco.example. Its buyer wrote a lovely newsletter about a small leather goods maker in Thomastown, Co. Kilkenny, invented too: hand stitched, vegetable tanned, no visible logo, pieces from €180 to €450. None of it is on the product pages, which say "Card holder, tan, €180". The fix is plain: put the maker's story, the materials, the method and the price position on each product page and on a maker page they link to, in words a shopper would use. "Hand stitched in Thomastown from vegetable tanned leather, no visible branding" is a sentence an assistant can lift into an answer to that birthday question. A newsletter in someone's inbox is not.

Structured data states the facts, it does not create a ranking

Google is explicit that there are no extra requirements and no special schema.org markup needed to appear in AI Overviews or AI Mode: a page has to be indexed and eligible to show with a snippet. It also asks that structured data matches the visible text, and that Merchant Center information is kept up to date.⁸ So markup is not a trick for AI answers. It is how you state price, stock, delivery and returns unambiguously, which every channel that reads your page benefits from.

For a product page, Google's merchant listing documentation names the properties: on the product a name, an image and an offer are required, and brand, colour, material, size, GTIN and ratings are recommended. On the offer, price and currency are required, and availability, shipping details and the return policy are recommended.⁹ Here is a generator for a Rails shop that sticks to those:

module SEO
  class ProductSchema
    def self.for(product, variant)
      {
        "@context": "https://schema.org",
        "@type": "Product",
        name: product.name,
        description: product.description, # the rich text that is visible on the page
        image: product.image_urls,
        sku: variant.sku,
        gtin13: variant.gtin,
        brand: { "@type": "Brand", name: product.brand_name },
        color: variant.colour,
        material: product.material,
        size: variant.size,
        offers: {
          "@type": "Offer",
          url: product.canonical_url,
          price: variant.price.to_s,
          priceCurrency: "EUR",
          availability: variant.in_stock? ? "https://schema.org/InStock" : "https://schema.org/OutOfStock",
          shippingDetails: {
            "@type": "OfferShippingDetails",
            shippingDestination: { "@type": "DefinedRegion", addressCountry: "IE" },
            shippingRate: { "@type": "MonetaryAmount", value: product.shipping_cost.to_s, currency: "EUR" },
            deliveryTime: {
              "@type": "ShippingDeliveryTime",
              handlingTime: { "@type": "QuantitativeValue", minValue: 0, maxValue: 1, unitCode: "DAY" },
              transitTime: { "@type": "QuantitativeValue", minValue: 1, maxValue: 2, unitCode: "DAY" }
            }
          },
          hasMerchantReturnPolicy: {
            "@type": "MerchantReturnPolicy",
            applicableCountry: "IE",
            returnPolicyCategory: "https://schema.org/MerchantReturnFiniteReturnWindow",
            merchantReturnDays: 30
          }
        }
      }.compact
    end
  end
end

Two changes from the earlier version. It put occasions, style and sustainability into additionalProperty fields, which Google's merchant listing documentation does not list,⁹ so that text now goes into the visible description instead. And it wrote availability as a bare word. The full schema.org address is the documented form.⁹ Review stars carry their own rules, which I take apart in the article on SERP features, together with product snippets and merchant listings.

An example: the Ballintemple Threads blazer in size 12 is in stock at €39.99, ships for €3.95, is dispatched the same or next day and arrives one or two days later, with 30 days to return. The generator turns that into one block of JSON on the page. The day it sells out, availability changes to out of stock on the next render, and a shopper asking for "delivered by Thursday" is not sent to a page that cannot deliver.

The checklist for a shop

1. Write descriptions a personal shopper would. Who it is for, when it is used, what it goes with, how it fits, what it is made of. True, specific, on the visible page.
2. Mark up the facts. Product and offer structured data that matches the page, with price, stock, delivery and returns. Keep Merchant Center in step.⁸
3. Let reviews speak. Detailed reviews that mention fit and use give assistants honest context. Never write them yourself.
4. Keep it current. A wrong price or stale stock is worse than no answer, for the assistant and for the shopper.
5. Answer the questions people actually ask. A size guide that says "runs large, size down" answers a question an assistant gets every day.
6. Decide who may read the shop. Allowing OAI-SearchBot and Google's search crawler while deciding separately about training crawlers is covered in the AEO and GEO article.
7. Measure what arrives. Referrals from AI assistants show up in your consented analytics by referrer. Count them and their orders separately, small as they are.
8. Remember the rest of the web. Assistants read reviews, comparisons and press, not just your site. Being described well elsewhere matters too.

The bigger picture

The old model was to optimise for Google's ranking, pay for adverts and compete on keywords. That model has not gone away, and by Adobe's own description the base of AI shoppers is still modest.¹ What has been added is a reader who asks in full sentences and answers on the shopper's behalf, which AI agents take a step further.

The work it asks for is not exotic. Describe your products the way a good shop assistant would, state the facts so machines cannot misread them, and keep both true. Shops that do that are easier to recommend for any reader, human or not.

Sources

Footnotes

1. Adobe, Adobe: Holiday Shopping Season Drove a Record $257.8 Billion Online with Consumers Embracing Generative AI Tools, press release, 7 January 2026, https://news.adobe.com/news/2026/01/adobe-holiday-shopping-season

2. Adobe, Adobe forecast for the 2025 holiday season (press release), 6 October 2025, https://news.adobe.com/news/downloads/pdfs/2025/10/100625-adobe-holiday-forecast-2025.pdf

3. Hema Budaraju, We're bringing the helpfulness of AI Overviews to more countries in Europe, Google, 25 March 2025, https://blog.google/feed/were-bringing-the-helpfulness-of-ai-overviews-to-more-countries-in-europe/

4. Hema Budaraju, AI Mode is now available in more languages and locations around the world, Google, 7 October 2025, https://blog.google/products-and-platforms/products/search/ai-mode-expands-languages-locations/ (Ireland is listed on Google's help page, Get AI powered responses with AI Mode in Google Search, https://support.google.com/websearch/answer/16011537)

5. Brian O'Donovan, Google launches new AI mode search tool in Ireland, RTÉ News, 8 October 2025, https://www.rte.ie/news/ireland/2025/1008/1537357-google-ai-ireland/

6. OpenAI, Overview of OpenAI crawlers (GPTBot, OAI-SearchBot, ChatGPT-User), OpenAI developer documentation, accessed November 2026, https://developers.openai.com/api/docs/bots

7. PayPal, PayPal and Perplexity Launch Instant Buy, PayPal newsroom, 25 November 2025, https://newsroom.paypal-corp.com/2025-11-PayPal-and-Perplexity-Launch-Instant-Buy

8. Google, AI features and your website, Google Search Central documentation, accessed November 2026, https://developers.google.com/search/docs/appearance/ai-features

9. Google, Merchant listing (Product, Offer) structured data, Google Search Central documentation, accessed November 2026, https://developers.google.com/search/docs/appearance/structured-data/merchant-listing