visibilai

Home / Guide

Guide · updated 29 October 2026

How AI assistants choose which products to recommend

What actually decides whether ChatGPT, Claude, Perplexity, Gemini or Copilot name your products, what the evidence says helps, what is a placebo, and how to measure it without fooling yourself. Written for Shopify merchants.

The short version.

  1. Product cards in AI answers come mostly from catalog feeds. For Shopify stores that means Shopify Catalog and Google Merchant Center. Being in the feed, with complete data, is the gate.
  2. Product names in the text of an answer follow the web: what reviewers, forums, videos and retailers say about you, and how you rank in Google and Bing.
  3. AI crawlers do not run JavaScript. Reviews and specs that load with scripts are invisible to them.
  4. llms.txt, extra schema and keyword-stuffed copy have little or no evidence behind them.
  5. Answers vary from run to run. Measure how often you are named over many runs, never one answer.

1. Two ways products get into an answer

When a shopper asks an AI assistant something like “a waterproof jacket for hiking under $200”, two different systems can put a product in front of them.

They reward different work. Feeds reward complete, accurate product data. The written answer rewards being talked about on other people’s sites.

2. Get into the feeds Strong evidence

For a Shopify store, most of the plumbing already exists:

Do this: open the Agentic channel, confirm each AI channel is on, fix every eligibility reason, connect the Google & YouTube channel to Merchant Center, and clear any disapprovals there.

3. Make your product data complete Strong evidence

AI shopping agents filter and match on structured fields. A product without a material, size or use case cannot match a question that asks for one. Google Merchant Center’s AI insights list popular attributes that may be missing from your product data, which says a lot about what the platform values.

Shopify’s channel mapping for Shopify Catalog lets the title and description it shares come from metafields (the product category cannot be mapped). That means you can give agents a plain, factual version without rewriting the storefront copy your customers read.

4. Reviews, and why AI often cannot see them Strong evidence

Review volume and rating line up with which products get the top shopping cards, and OpenAI’s product feed format has fields for review count and star rating.

The catch: the major AI crawlers do not execute JavaScript. A crawler study by Vercel found that bots such as GPTBot and ClaudeBot download script files but do not run them. Most Shopify review widgets, FAQ accordions and size charts load with JavaScript, so to those crawlers the page has no rating at all.

Test it: open a product page, view the page source (not the inspector), and search for your rating. If it is not there, crawlers do not see it.

5. Let the right crawlers in Strong evidence

Each AI company runs several bots with different jobs. Blocking the wrong one can quietly remove you from AI search.

AI crawlers and what they do
CrawlerUsed forSuggestion
OAI-SearchBotChatGPT search resultsAllow. Sites that block it are not shown in ChatGPT search answers.
GPTBotOpenAI model trainingYour choice. It does not affect search.
PerplexityBotPerplexity search indexAllow.
Claude-SearchBotClaude searchAllow.
ClaudeBotAnthropic model trainingYour choice.
Google-ExtendedGemini training and grounding (not Google Search or AI Overviews)Your choice.
Bingbot, GooglebotWeb search, which also grounds Copilot and Google’s AI featuresAllow.

Shopify’s default robots.txt is fine for most stores. Problems usually come from a customised robots.txt.liquid or a CDN or firewall in front of a custom domain that blocks AI bots by default. Note that blocking crawlers does not stop Shopify Catalog from sharing your products with AI channels; the two are separate.

6. Keep prices and stock consistent Strong evidence

When the price on your page, in your structured data and in your feeds disagree, shopping systems may drop the product, and an assistant may quote the wrong price. Check that sale prices, currencies and stock status match everywhere, and re-check after a sale ends.

7. Earn mentions off your site Correlational evidence

For brand names in the written part of an answer, the strongest correlate found so far is how often a brand is mentioned across the web. A study of 75,000 brands found web mentions correlated with AI Overview visibility more strongly than backlinks did; that is a correlation, not proof that more mentions cause more visibility. Academic work found AI search engines lean towards earned third-party sources over a brand’s own pages. And assistants often cite pages that rank in Google and Bing, though frequently for related sub-questions rather than the exact question asked.

This is the slowest work, it happens outside your store, and its effect is the hardest to prove. Look at which sources the assistants cite for your competitors; that list is your outreach plan.

Plausible, not yet proven Plausible

8. What does not work Weak or none

9. How to measure it honestly

10. Checklist

  1. Agentic Storefronts on, every AI channel enabled, no “Ineligible” reasons.
  2. Store policies complete; no password page.
  3. Google Merchant Center connected with no disapprovals.
  4. Every product has the right category, filled attributes, barcode where one exists, and a factual title.
  5. Rating and key specs visible in the page source.
  6. OAI-SearchBot, PerplexityBot, Claude-SearchBot, Bingbot and Googlebot allowed.
  7. Prices and stock match across page, structured data and feeds.
  8. A list of the third-party sites assistants cite in your category, and a plan to be on them.
  9. A measurement routine based on real traffic: AI sessions and orders, Search Console and Bing data, compared with the day you made each batch of changes, 7 days after.

Where Visibilai fits. Visibilai helps with steps 4 to 9 on Shopify: it audits your store for the reasons above, proposes improvements, applies them only when you approve (with undo), and then shows what your real traffic and search data did, 7 days after you apply changes. It cannot promise recommendations, and neither can anyone else. See how it works or check your store yourself.

Questions

Do I need to submit my products to ChatGPT?

Usually not. For eligible Shopify stores, Shopify’s Agentic Storefronts channel shares your catalog with AI channels, and it is on by default for eligible stores. Check its status under Sales channels in your Shopify admin.

Does adding schema markup get me cited by AI?

There is no good evidence that it does on its own. Google says no special structured data is needed for its AI features. Keep your Product structured data correct because it feeds shopping results, but do not expect it to change AI answers by itself.

Why can’t AI see my reviews?

Most review widgets load with JavaScript after the page arrives, and the major AI crawlers do not run JavaScript. If your rating is not in the page source, those crawlers do not see it. A server-rendered review summary fixes that.

How do I know if my changes are working?

Look at real traffic: visits and orders that arrive from AI assistants, and the clicks and impressions Google and Bing report for your pages. One answer from an assistant means little because lists change from run to run and from person to person. Combine your own numbers with the free reports from Shopify, Google Merchant Center and Bing Webmaster Tools.

Sources

Checked on 29 October 2026. Platform rules change often; follow the links for the current version.