Fix Shopify product titles for AI recommendation by putting the plain product type next to your creative name, then adding the use case to the description. Most catalogs under 500 SKUs can be corrected in an afternoon using a CSV export and a bulk edit.
Your brand team spent three weeks naming that product. An AI assistant reads the name, learns nothing, and recommends someone else.
The channel is small and growing fast, which is the worst combination for anyone deciding where to spend an afternoon. On Shopify’s first quarter 2026 earnings call, President Harley Finkelstein reported that AI-driven traffic to Shopify stores grew 8x year over year, orders from AI-powered search rose nearly 13x, and new buyer orders from AI search now arrive at close to twice the rate of traditional organic search.
Those are platform numbers. What makes them actionable is a much smaller observation about how the recommendation actually happens, and about a mistake that sits in most Shopify catalogs right now.
Assistants recommend by matching a product to a shopper’s stated situation, which means a product page that never states its own category cannot be matched to anything. This is not a ranking problem and it does not respond to the usual SEO levers. It is a data legibility problem.
Rob Langenback, President of Eight Oh Two, put it plainly on episode 480 of the podcast. His agency does search and nothing else, and he described clients arriving with catalogs full of lovely names where the product type had simply gone missing. One client named products things like “the Emma.”
If we don’t call it a shirt, how do we expect the tools to know that that’s a shirt?
Here is why this bites harder than it used to. A shopper typing into Google wrote three words and Google inferred the rest from your category structure, your breadcrumbs, and a decade of behavioral data about your store. A shopper talking to an assistant writes a full sentence with constraints attached, and the assistant resolves it against whatever text it can actually read on the page. There is no inference layer doing you a favor. If the page says “the Emma, $148, ships free,” the assistant has a price and a name and nothing to match.
Rob’s own client data puts LLM traffic at roughly 1% to 5% of sessions or revenue, growing 1,200% to 1,500% year over year for clients tracking it properly. That split is the argument for doing this now rather than later. The volume does not justify a project. The growth rate and the fact that the fix takes an afternoon do.
Export your product CSV from Shopify admin and count how many titles contain a plain product type, because that number is your actual starting position and it is usually worse than teams expect. Go to Products, select all, and export as CSV. Open it and look only at the Title column.
Add one column called “type present” and mark each row yes or no. You are asking a single question: if I deleted everything except this title, would a stranger know what the object is? “The Emma” fails. “Emma Longline Cotton Cardigan” passes. “Signature Blend” fails. “Signature Blend Medium Roast Coffee Beans” passes. Do not overthink the edge cases, mark them no and move on.
On a 200 SKU catalog this takes about 20 minutes. Sort by the new column and you have your work queue, ordered by nothing yet, which is the next problem to solve.
Then run the second half of the audit, which is Rob’s free process from the episode. Write ten prompts a real buyer would type before purchasing in your category, the kind with constraints attached rather than product names. Run each one across ChatGPT, Claude, and Perplexity, and log which brands come back. Rob is direct that you will never get the same result twice and that two different days produce two different results, so run each prompt more than once and read for patterns rather than treating any single response as a verdict.
What you are looking for is not your own absence. It is which competitor keeps appearing. Open their product pages next to yours and the gap is usually visible inside a minute.
Keep the creative name and add the product type immediately after it, because the branding asset and the machine-readable asset can occupy the same field. This is the part teams resist, and the resistance is worth naming: nobody wants to be told their naming work was a mistake. It was not. It was incomplete.
The pattern that works is creative name, then plain product type, then the single attribute buyers filter on most in your category.
“The Emma” becomes “Emma Longline Cotton Cardigan.” “Signature Blend” becomes “Signature Blend Medium Roast Coffee Beans, Whole Bean.” “The Coastal” becomes “Coastal Long Sleeve Swimsuit, UPF 50.” The brand voice survives. The category becomes explicit. The attribute that shows up in real buyer language is now present.
Two rules keep this from degrading into keyword stuffing. Use the words your buyers use rather than your internal category taxonomy, because “longline cardigan” is a shopper phrase and “outerwear layer 2” is a merchandising phrase. And stop at one attribute in the title. Everything else belongs in the description, which is where the situational language does its work.
That situational language is the second half of the fix. Rob’s swimwear example is the template: rather than generic copy, the page states the UPF 50 rating, says the garment suits sun-sensitive skin, and names the situation of several hours on a beach. He points out this is a phrase nobody would have targeted as a keyword a few years ago, and it is now close to verbatim how people ask. If you want the structural version of this argument, our answer-first strategy for Shopify brands covers how to build pages assistants can quote cleanly.
Fix your top 20 revenue-generating products first and ignore the long tail entirely on the first pass, because the recommendation surface you care about is narrow and your time is not free. Pull the last 90 days of sales by product from Shopify Analytics, take the top 20, and work only on those.
At this catalog size you do not need tooling. Edit the titles directly in Shopify admin, or make the changes in your exported CSV and re-import. Budget roughly two to three hours for 20 products if you are also rewriting descriptions, which you should be, since the title alone gets you matched and the description is what gets you chosen.
For each of those 20, write two or three sentences naming the person and the situation the product is right for. Not benefits language. Situations. “Best for someone with sun-sensitive skin planning several hours outdoors” is a situation. “Premium sun protection you can trust” is not, and an assistant can do nothing with it.
The reason to stop at 20 is the pattern I have watched play out repeatedly at the $500K to $2M stage, which is premature complexity. Teams at this size reliably choose the comprehensive catalog overhaul, stall out at product 60, and abandon the project with two thirds of the catalog half-edited and inconsistent. Twenty finished products beat 200 started ones, and the finished twenty tell you whether the approach is working before you commit the rest of the quarter to it.
Past roughly 2,000 SKUs the constraint stops being writing and becomes governance, which means the fix has to live in a system rather than in a person. Manual editing does not scale, and more importantly, manual editing does not stay fixed. New products arrive, seasonal ranges launch, and the naming convention drifts back within two quarters unless something enforces it.
Two practical paths. For bulk editing against an existing catalog, Matrixify handles large CSV imports and exports with variant-level control, and it is the standard answer for one-time restructuring work at this scale. For ongoing enforcement, this is where a product information management system earns its cost, because the naming convention becomes a field rule rather than a habit. Plytix, which sponsored the episode this article draws on, sits in that category, as do Akeneo and Sales Layer. Evaluate them on how well they handle Shopify variants and how much of your team can actually work in them, not on the AI feature list.
Whichever path you take, write the convention down before you touch anything. A single page naming standard specifying the order of elements, the approved product type vocabulary, and the one-attribute limit is what makes the fix durable. Without it you are buying a tool to enforce a rule nobody has written.
Rob is candid that even at the mid-market and enterprise level, his own agency is still testing tools for the visibility side of this and that his clients often show a disconnect between what brands think they are known for and what they actually rank for. Do not wait for the tooling to mature. The catalog work is the part that is fully within your control today.
Add variant markup for sizes, colors, and patterns, because Rob identifies it as the field most Shopify stores skip and it feeds both assistants and the product grids now crowding Google results. Most themes ship Product and Offer markup by default and stop there, which leaves a real gap.
The implementation is ProductGroup markup wrapping your variants, and it tells the assistant that one product exists in seven sizes and four colors rather than presenting twenty-eight loosely related items. If a shopper asks for something in a specific size, that distinction decides whether you are eligible for the recommendation at all. Our guide to structured data and ProductGroup markup for variants covers the JSON-LD implementation alongside the return policy and merchant listing fields worth adding at the same time.
Beyond variants, the extractable facts worth surfacing in clean page text are dimensions, materials, fit, and care. Rob makes the case with a domestic example: he and his wife photographed a room they were renovating, asked ChatGPT for decor ideas, and the assistant came back specifying a couch 80 to 84 inches wide. A product without assembled dimensions on the page cannot be matched to that constraint no matter how good it is.
Validate everything with Google’s Rich Results Test before you consider the work finished. Inaccurate markup is worse than absent markup, because it breaks trust with the systems you are trying to earn trust from.
Rerun the same ten prompts 30 days after the changes ship and compare brand appearance rates, because no tool currently reports AI recommendation share reliably enough to use as your primary measure. This is unsatisfying and it is the honest state of measurement in 2026.
Keep the prompts identical between runs. Log the date, the assistant, the prompt, whether your brand appeared, and which competitors appeared. Run each prompt three times per assistant, because Rob’s point about two different days producing two different results means a single run tells you almost nothing. What you are measuring is a rate, not an event.
Alongside that, watch the two signals in tools you already have. In Google Analytics, segment referral traffic from assistant domains and track conversion rate and revenue per session rather than session count, since the whole argument for this channel is that fewer visitors arrive further along. In Google Search Console, watch impressions on your product pages, since improved product data tends to show up in Google’s product grids before it shows up anywhere else.
Set expectations honestly with whoever approved the time. Rob’s anonymous cookware client went from essentially no organic traffic to close to 30,000 visits a month, and that took about 18 months of sustained work across product and collection pages rather than one afternoon of title edits. The title fix is the cheapest, highest-certainty step in that sequence. It is not the whole sequence.
Write the creative name first, then the plain product type, then the single attribute buyers filter on most in your category. “The Emma” becomes “Emma Longline Cotton Cardigan.” The reason is that AI assistants match products to a shopper’s stated situation using the text on the page, with no inference layer filling in what you left out. A distinctive name carries brand value and tells a model nothing about the object. Use shopper vocabulary rather than your internal merchandising taxonomy, and stop at one attribute in the title itself. Everything else about who the product suits belongs in the description.
The most common reason is that your competitor’s product pages state the product type and the use case explicitly while yours do not. Assistants match a recommendation to a shopper’s situation, so a page that never says what category the item belongs to or which situations it suits cannot be matched against a constrained query. Run ten buyer-style prompts across ChatGPT, Claude, and Perplexity, log which brands appear, then open the winning competitor’s product and collection pages beside your own. The difference is usually visible within a minute, and it is usually specificity rather than authority or backlinks.
Budget about two to three hours for your top 20 revenue-generating products if you are rewriting descriptions alongside the titles, which you should be. A full catalog under 500 SKUs is generally an afternoon of work using a CSV export and bulk re-import. Past roughly 2,000 SKUs the job shifts from writing to governance and needs tooling, either a bulk editor like Matrixify for the one-time restructure or a PIM for ongoing enforcement. Start with the top 20 regardless of catalog size, because finished products teach you more than started ones.
Variant schema, implemented as ProductGroup markup, tells search engines and AI assistants that one product exists in multiple sizes, colors, or patterns rather than presenting them as separate loosely related items. Most Shopify themes ship Product and Offer markup by default and skip this entirely. It matters because a shopper asking for a specific size or color is filtering on exactly the attribute the markup describes, and because Google’s product grids draw on the same structured data. Add it alongside accurate availability and return policy markup, then validate everything with Google’s Rich Results Test before treating the work as complete.
Rerun an identical set of ten buyer-style prompts 30 days after your changes ship and compare how often your brand appears, running each prompt three times per assistant because results vary day to day. No tool currently reports AI recommendation share reliably enough to serve as a primary metric, which makes repeated sampling the honest approach. Alongside that, segment assistant referral traffic in Google Analytics and track conversion rate and revenue per session rather than session count, since the value of this channel is that visitors arrive further along in the decision.
Keep the names and add the product type beside them, because this is not a choice between branding and discoverability. The creative name and the plain category can occupy the same title field, and the fix is additive rather than a replacement. Distinctive naming builds recognition with humans who already know your brand, and it does nothing for an AI assistant trying to determine whether your product answers a stranger’s question. Adding three words of plain description after the name preserves everything the naming work achieved while making the product eligible for recommendation in the first place.
