AI shopping tools will only recommend a product they can read, and some stores hide the details inside JavaScript that a bot never receives. On November 2, Search Engine Journal published a recap of a webinar where seoClarity's Chris Sachs and Aris Vrakas walked through how to fix that. The advice fits any BC business that sells products online.
The webinar covered four moves: make product pages readable, describe products in the words shoppers use, keep feeds and pages in sync, and measure what changed. The presenters pointed at work a store can start now, without waiting for checkout inside AI services to mature. Here is what each move means for a BC store that wants AI shopping tools to find its products, in the order a store should do them.
Check what the bot receives
Vrakas started with access. Product content has to be available to bots as well as visible to a person in a browser. He described JavaScript-heavy pages where a shopper sees the details and an automated fetch does not receive them. If your price, sizes or materials load only after scripts run, an AI shopping tool may find your page and still leave with nothing to recommend.
The check is quick. Open the page source of a best seller (view source, not the inspector) and look for the product name, price and key specs. View source shows the HTML the server sends before scripts run, which is what a bot that skips JavaScript works from. If the details are missing, that is your first job, and it comes before any copywriting. Vrakas named prerendering and cached HTML as options where rendering is a bottleneck, and both are web development work. Google's own documentation calls dynamic rendering, which serves bots a prerendered version of the page, a workaround and not a long-term solution. It recommends server-side rendering, static rendering or hydration instead, so ask your developer which one fits your store. Rewriting descriptions first is wasted effort when the bot never receives them.
Describe products the way shoppers ask
The presenters used socks as their running example. Colour and fibre blend are useful attributes, but they say little about warmth, thickness, durability or intended use, and a conversational prompt can combine several of those needs in one request.
Say your store sells wool socks. A shopper asks an AI shopping tool for a pair warm enough for cold mornings and tough enough for work boots. A listing with a colour and a fibre blend answers none of that, so the tool has to guess, or it recommends someone else's socks.
Vrakas recommended tying product characteristics to the decisions shoppers are making, so the model does not have to infer a use case from a short specification. He also discussed using attributable review evidence to support claims about how a product performed. Even a broad request for a good sock can send an AI system looking for specific attributes, which means you can anticipate the questions before they arrive.
One limit applies. Write only what the product's attributes or your customers' reviews support. A warmth claim the sock cannot back is worse than a short description.
Keep the feed, the page and the markup identical
Sachs warned against treating the product feed as a separate project owned somewhere else in the company. What goes into a feed should also inform the product page and its structured data. Vrakas explained why: when a platform revisits a page, product identity, price and supporting details have to be clear and consistent. An enriched feed does not excuse a vague page.
Sachs called this closing the loop. It means SEO, merchandising, paid media and development work from one version of each product. The failure to avoid is easy to picture: say an agency manages your Google feed, a developer manages the site, and a price changes in only one of them. Google Merchant Center crawls landing pages, compares the price with your product data, and can disapprove a product when the two differ.
Local businesses face the same question from a different angle, covered in how AI recommends local businesses.
Measure three things separately
The presenters split measurement into three areas:
- Product-grid visibility: how often your priority products show up in shopping carousels for relevant searches.
- AI mentions and citations: whether an answer names a product or cites your content, and how it describes your brand.
- Paid-feed results: how many people saw, clicked and bought through feed-driven ads.
These are different signals, and no single score covers them. An AI mention does not establish a sale, and a strong carousel position does not mean AI shopping tools name you. Sachs tied the work to testing: “The best SEO teams are the ones that obsess about testing. The more testing you do, the better decision making you can have.”
In the Q&A, a viewer asked whether a before-and-after comparison is enough to judge generative engine optimization. Vrakas separated large catalogs, where product groups can be compared with each other, from smaller sites or broader brand questions. For the second case he described improving content around a defined topic and comparing its movement with topics left untouched. A smaller site falls into that second case. Without a comparison set, any visibility change can look like proof.
NG checks every month what ChatGPT, Gemini, Perplexity and Claude say about a client's business and which sites they cite. That check shows what the assistants say about your brand today, the subject of your AI salesforce is already describing your business. It also shows movement. A Lower Mainland metal building products company got a new website with its own order and quote forms, and by Aug 28, 2026, 94% of AI answers to product questions cited the site, up from 50% on Jul 2 (AI answer check, 3 assistants). The full numbers are on the metal building products case page.
Start with a small set of priority pages
Most stores will want to jump straight to rewriting descriptions, because writing is more pleasant than debugging rendering. Resist that. Access comes first, then content, then consistency, then measurement, and each step wastes effort if the one before it is broken. The webinar's own first review translates into this:
- Open the page source of a small set of priority products and confirm the name, price and key details are in the HTML.
- List the questions a shopper would ask about each product, and add answers that its real attributes or reviews support.
- Decide who owns the feed, and how approved changes reach the product page and its structured data.
- Record baselines for grid visibility, AI mentions and citations, and paid-feed results.
- Change one group of products and leave a similar group untouched for comparison.
Step one can be done today at no cost. Anything beyond it, such as server-side rendering, is development work, and the rest is search work that compounds slowly. NG's SEO & AEO plans run $800, $1,200 or $1,800 a month (CAD), and the pattern for SEO is first moves in 6 to 10 weeks and real traffic in 3 to 6 months. If you want a second pair of eyes on what your product pages show a bot, and whether AI shopping tools can find them, request a quote.