How Do I Add Conditions in Bulk for Shopify Products? (Real Workflow)

This is a build-in-public post — a real workflow we ran for a client, the actual approach, and what it cost. Part of our AI workflows pillar.

The customer pain point

The client sold a mix of new, refurbished, and used inventory. The condition was buried inside each product description as free text ("This unit is refurbished and tested…"). It wasn't a structured field, so:

• It couldn't be shown consistently on the product page
• It couldn't be filtered or merchandised
• It couldn't be used in feeds (Google Shopping requires a condition field)
• Editing ~800 products by hand to add a structured condition was a multi-day manual job nobody wanted

They needed condition as a structured, displayable, filterable field across the whole catalog — without manual editing.

Table of Contents

• The Approach
• The Custom App + Prompt
• Rendering It Dynamically on the PDP
• Cost & Time
• What to Watch For

Key Takeaways

The Approach

Three pieces:

1. Extraction — read the existing product description, identify the condition (new / refurbished / used).
2. Storage — write that condition into a structured metafield (custom.condition) so it's a real field, not free text.
3. Display — reference the metafield in the theme's product information block as a dynamic variable so it renders on every PDP automatically.

The key insight: don't edit products manually and don't hardcode condition into the description. Extract once into a metafield, then let the theme read the metafield. New products get processed by re-running the app.

The Custom App + Prompt

We built this with Shopify's AI app generation mode (the prompt-to-app builder inside the Shopify dev tooling). The prompt we used to generate the app logic:

Build a Shopify app that:

1. Loops over all products in the store (paginate via the Admin GraphQL API,
   250 products per page).
2. For each product, reads body_html (the description).
3. Determines the product condition by scanning the description text for
   condition signals:
   - "refurbished", "renewed", "reconditioned" → "refurbished"
   - "used", "pre-owned", "second-hand", "open box" → "used"
   - otherwise → "new"
4. Writes the result to a product metafield:
   namespace: "custom", key: "condition", type: "single_line_text_field"
5. Skips products that already have custom.condition set (idempotent —
   safe to re-run).
6. Logs every product handle + the condition assigned, and a summary
   count at the end.

For products where the description was ambiguous, we added a second pass that sent just the description to the Claude API with this prompt:

You are classifying a Shopify product's condition. Read the product
description below and respond with EXACTLY ONE word: "new",
"refurbished", or "used". No punctuation, no explanation.

Description:
{body_html stripped of HTML}

The deterministic keyword pass handled ~85% of products instantly (near-zero cost). The Claude fallback handled the ambiguous ~15%.

Rendering It Dynamically on the PDP

Once custom.condition exists on every product, the theme reads it. In the product information section (Liquid), reference the metafield as a dynamic variable:

{% if product.metafields.custom.condition %}
  <div class="product-condition">
    Condition:
    <strong>{{ product.metafields.custom.condition | capitalize }}</strong>
  </div>
{% endif %}

If the store uses a metafield-aware theme block (Online Store 2.0), you can add the metafield as a dynamic source directly in the theme editor — no Liquid editing required. Add a text block to the product information section → click the dynamic source icon → select custom.condition.

The result: condition shows on every PDP, pulled live from the metafield. Add a new product, run the app, condition appears. Zero manual work after the initial build.

Cost & Time

The honest takeaway: this is mostly a build cost, not a token cost. The AI is doing classification on short text — that's cheap. The value is in never touching 800 products by hand and having a re-runnable system.

What to Watch For

• Idempotency matters. The app must skip products that already have the metafield set, or a re-run will re-process everything and waste calls. We made the skip check explicit in the prompt.
• Google Shopping condition values are constrained. Google's feed spec only accepts new, refurbished, used. Map to those exact strings if the metafield feeds your Shopping feed.
• Ambiguous descriptions exist. "Like new" could be new or used depending on the brand's policy. We had the client define the rule once; the app applied it consistently.
• Theme dynamic source vs Liquid. On OS 2.0 themes, prefer the dynamic source approach — it survives theme updates better than custom Liquid edits.

Talk to Branva

We build custom Shopify apps and AI workflows like this as part of transparent monthly Shopify operations. Book a free call — we'll look at what manual work is eating your week and what we'd automate first.

Frequently Asked Questions

Can't I just use Shopify's bulk editor for this?

Only if the condition data already exists in a structured form. The bulk editor edits existing fields — it can't extract condition from inside free-text descriptions. That extraction step is what the app does.

Why a metafield instead of a tag or the product type?

Metafields are purpose-built for structured attributes, render cleanly as dynamic sources in OS 2.0 themes, and don't pollute your tag taxonomy or product-type filtering. Tags work but get messy at scale.

Do I need Shopify Plus for this?

No. Metafields and the Admin API are available on all Shopify plans. Shopify's AI app generation tooling is also available to standard dev accounts.

How do I keep it updated as I add products?

Re-run the app. Because it's idempotent (skips products that already have the metafield), a re-run only processes new products and takes minutes.

Related reading

• The AI Workflows pillar — every workflow we run, where AI fits, where humans review.
• How Do I Add Alt Text to Shopify Images in a Block? — same custom-app pattern, applied to image SEO.
• What an AI Marketing Agency Actually Does — where this kind of build fits in the broader model.
• The Shopify Operator Stack 2026 Edition — the tools behind workflows like this.