Shopify Product Descriptions for AI Shopping Agents

How AI Agents Read Shopify Product Descriptions: Three Pathways

Before optimizing your descriptions, you need to understand what each AI shopping agent actually reads. There are three distinct pathways, and content visible in one may be invisible in another.

Pathway 1 — Direct HTML crawl

Agents like Perplexity's PerplexityBot and OpenAI's OAI-SearchBot crawl the rendered HTML of your product pages and extract text from the DOM. This pathway reads content that is present in the server-rendered HTML at page load. Content injected by JavaScript after DOM ready — including most Shopify app review widgets, upsell content, and dynamically loaded metafield blocks — is not reliably captured unless the crawler executes JavaScript, which most do not for product-level crawls.

Shopify's native body_html field is always in the server-rendered output. Product description apps that rewrite the DOM client-side are invisible to most crawlers.

Pathway 2 — /products.json API

Shopify exposes every store's product catalog at https://yourstore.com/products.json (paginated with ?page=N&limit=250). This endpoint returns the body_html field with all HTML tags intact. AI systems that consume this feed then strip tags themselves using their own HTML parser. The result: any content not stored in Shopify's native body_html column — content from third-party apps, metafield rendering, or theme injection — does not appear in this pathway at all.

Special HTML entities such as &, —, and ™ survive this stripping process as their Unicode characters. Decorative whitespace tags like <br> and <hr> collapse to spaces. Nested <div> wrappers add zero content value.

Pathway 3 — Google Merchant Center feed

Google AI Mode sources product data primarily through the Merchant Center feed, where the description attribute corresponds to your Shopify product description but with a hard 5,000-character cap. Shopify's Google & YouTube channel exports this field automatically, but it uses the raw body_html with minimal processing. HTML tags that survive into the feed description are flagged as feed errors. If you rely on heavily HTML-tagged descriptions (tables, divs, custom classes), the exported feed description may be significantly shorter than the actual visible text because tag overhead consumes the character budget.

Word-Count Thresholds That Determine AI Citation Rate

AI shopping agents calibrate their confidence in a product description based on its length. Below are the five operative tiers observed across ChatGPT Shopping, Perplexity, and Google AI Mode citation patterns.

These counts refer to visible text after HTML stripping, not raw body_html character length. A description with 600 characters of HTML tags and 80 words of actual text falls in the "50–149 words" tier, not the "500+" tier.

Description Source vs. AI Agent: What Each Platform Reads

The body_html Technical Requirements

Shopify's body_html field is the canonical source for all three pathways. Technical issues here compound across every downstream system.

JavaScript-injected content is invisible

Any content rendered by a Shopify app after page load — including dynamically inserted feature lists, fit guides, or ingredient panels — does not exist in body_html. It will not appear in /products.json and will be missed by crawlers that do not execute JavaScript. Move critical content into the native Shopify description field in the admin.

HTML tag overhead inflates character count without adding text value

A common pattern is wrapping each sentence in a <div class="desc-section"> block. This adds approximately 30 characters of tag overhead per sentence, consuming your Merchant Center feed's 5,000-character budget without contributing readable text. Use plain <p> tags, <ul>/<li> for feature lists, and <strong> only for genuinely critical terms.

Special characters survive /products.json stripping

HTML entities such as ™ (™), ® (®), and — (—) are decoded to their Unicode equivalents in the JSON output. This is expected behavior. However, malformed entities like ™ or unescaped bare ampersands cause JSON parse errors in some downstream consumers. Always use well-formed HTML entities or their Unicode characters directly.

JSON-LD Description Field: Liquid Implementation

The description property in your Product JSON-LD is the highest-confidence signal for ChatGPT Shopping and is cross-referenced against visible body text by Perplexity. It must be a plain-text string — no HTML tags, no newline characters, no unescaped quotation marks.

The following Liquid snippet produces a clean, safely JSON-encoded description from body_html. Place this inside your Product JSON-LD <script> block in your Shopify theme's product.liquid or product-template.liquid section.

{%- assign desc_clean = product.description
      | strip_html
      | strip_newlines
      | strip
      | truncatewords: 500 -%}

<script type="application/ld+json">
{
  "@context": "https://schema.org/",
  "@type": "Product",
  "name": {{ product.title | json }},
  "description": {{ desc_clean | json }},
  "sku": {{ product.selected_or_first_available_variant.sku | json }},
  "brand": {
    "@type": "Brand",
    "name": {{ product.vendor | json }}
  },
  "offers": {
    "@type": "Offer",
    "url": {{ canonical_url | json }},
    "priceCurrency": {{ cart.currency.iso_code | json }},
    "price": {{ product.selected_or_first_available_variant.price | divided_by: 100.0 | json }},
    "availability": {% if product.available %}"https://schema.org/InStock"{% else %}"https://schema.org/OutOfStock"{% endif %},
    "priceValidUntil": "{{ 'now' | date: '%Y' | plus: 1 }}-12-31"
  }
}
</script>

Key decisions in this snippet: strip_html removes all tags; strip_newlines collapses the output to a single line, avoiding JSON syntax errors from literal newline characters; strip removes leading and trailing whitespace; truncatewords: 500 keeps the field under approximately 3,500 characters, safely below the practical JSON-LD description limit. The | json filter at the end handles all necessary quote escaping.

Six Technical Signals AI Agents Extract from Descriptions

AI shopping agents do not treat descriptions as opaque text blocks. They extract structured signals using pattern matching. Including these signals explicitly — with consistent formatting — dramatically increases the probability that an agent uses your description in a product recommendation.

Per-Agent Implementation Priority

Technical Implementation Checklist

Further Reading

• Shopify Product Descriptions for AI Agents: Strategy Guide — the companion blog post covering content strategy and copywriting approach.
• Shopify XML Product Feed Format for AI Agents — how feed format affects downstream AI Shopping ingestion.
• AI Product Description Generator for Shopify — tools and prompts for scaling description quality across large catalogs.
• AI Shopping Agent Product Ranking Factors — the full ranking signal matrix across ChatGPT, Perplexity, and Google AI Mode.

Frequently Asked Questions

See Your Description Coverage Across Your Entire Catalog

Run a CatalogScan check on your store to see exactly which description signals are present across your catalog — word count per product, JSON-LD description presence, body_html cleanliness, and Merchant Center feed description length — all in one report.