VariantLens — Shopify Feed & Variant Checker

Find exact-variant differences between a shopping feed and a Shopify storefront. VariantLens compares feed values with product data, variant-linked JSON-LD and supported rendered page signals, then produces an evidence report with screenshots. Optional GPT-6 Luna advice explains what to inspect next.

Cloud access requirement: Shopify may block cloud audits with HTTP 429. Store owners should create a signature in Shopify Admin → Online Store → Preferences → Crawler access, for the exact hostname being checked, and paste its Signature-Input and Signature values into the two encrypted Actor inputs. VariantLens adds the required Signature-Agent. The operator bot identity is signed but is not yet registered with Shopify; unrestricted cloud access is not promised. See Shopify’s crawler setup guide.

Early beta: tested on three public variants in Shopify’s official Dawn demonstration store using a constructed test feed, including deliberate price, stock and size errors. This is limited theme coverage, not a merchant-feed accuracy benchmark. A result is a point-in-time diagnostic, not Merchant Center approval or a complete policy audit.

Try the fictional demo

Run the default input with mode: demo. Its six fictional variants demonstrate a matching page, price mismatch, stock mismatch, missing markup, blocked page and wrong selected variant. No merchant is contacted. AI is off by default.

Check your store

1. Export up to 25 feed rows in CSV, TSV or JSON. Required fields: id, link, price, availability.
2. Select live, enter the exact HTTPS store hostname and paste the feed into feedText.
3. Confirm you own or have permission to audit the store. Links must identify the product and numeric Shopify variant, for example https://your-store.example/products/tee?variant=123.
4. For cloud access, supply a valid Shopify crawler signature for this hostname in the encrypted fields. These headers are never sent to other hosts or copied into reports. Expired or malformed credentials fail before store requests.
5. Keep rendering enabled to collect screenshots and visible evidence. Static checks cannot establish a landing-page match.
6. Optionally enable AI and supply your own OpenAI key in the encrypted input field. OpenAI charges are separate. Only diagnostic findings, counts and limitations are sent; HTML and screenshots are not sent to AI.

On supported Dawn markup, exact selected-variant data is read directly from the matching product component in the original page, avoiding a redundant product-JSON request. Other themes use the public Ajax product endpoint. The report records the source; ambiguous data is not accepted.

Optional columns include title, SKU, colour, size, currency, sale price and a timezone-aware sale interval. Prices are compared with decimal arithmetic. Ambiguous currencies and unsupported values require review instead of an assumed match.

Outputs

• OUTPUT: complete JSON report with row results, findings, timestamps, evidence and request-health diagnostics (counts, cache hits, retries and bounded error metadata).
• REPORT: readable HTML report containing screenshots, request health and any AI advice.
• ROWS: CSV summary, with formula-like cells neutralised.
• Dataset: one item per input variant.
• SCREENSHOT-N: available screenshot records.

Results are matched, mismatch, needs_review or blocked. A mismatch can coexist with incomplete evidence; inspect every finding. AI suggestions are linked to finding IDs and cannot change the results. If the AI request fails, the underlying report is still saved.

Coverage and limits

One store and anonymous market context per run; up to 25 rows and 1 MB feed text. Only public pages on the authorised hostname and selected Shopify asset hosts are requested. Robots rules and access restrictions are respected. No login, checkout, feed edits or bot-challenge bypasses are performed. Cloud data-centre requests can receive HTTP 429 even when a page works locally; affected checks are blocked/incomplete and carry no variant event fee. Direct connections are the default. Optional datacenter or residential modes use one fixed Apify session. Residential mode fetches public Shopify CDN assets directly to reduce transfer cost. Proxy tests also received persistent 429s; changing proxy mode does not replace crawler authorization. Proxy usage is included in the variant price; the selected proxy group must be available to your Apify account. Proxy setup must succeed before an audit starts. Requests are spaced, eligible public theme assets are cached for up to 120 seconds and product JSON for up to 30 seconds within the audit, and a 429 can receive one retry after Retry-After when the time budget permits. Retry waits share a 65-second budget; repeated rejection stops further requests to that host. HTML and private/no-store responses are not reused. No automatic session rotation or connection fallback occurs after an error.

Not covered: checkout totals, tax/shipping rules, customer discounts, subscriptions, every theme/locale, product images/descriptions or Google's crawl history. Unsupported visible selectors, blocked dependencies and unknown values need review. In the Dawn test, all baseline values agreed but the results correctly retained review warnings because tracking/checkout dependencies were blocked. A detected mismatch can still be reported alongside those warnings. JSON and HTML include the blocked-resource details. A matching row only means that the listed collected signals agree.

There is a 180-second batch admission budget, 45-second page captures, per-page request and byte limits, and bounded AI output with no automatic retry. Set a run timeout and spending limit in Apify. Use the default 2 GB memory setting initially and measure real runs before reducing it. Retain/share downloaded reports carefully; Apify output and input retention follow your account settings.

Charging

The checked-variant event applies only to live rows with an exact Shopify variant price and purchase availability, a rendered purchase-form selection, and one visible price in the feed currency. A qualifying row may still be needs_review or mismatch; this fee pays for collected evidence, not a guaranteed match or complete page load. Demo, static-only, blocked and incomplete captures have no variant event fee. Reports are saved before charging. Apify platform usage is included in the event price. Optional OpenAI usage is billed separately by OpenAI. The run spending limit must cover all submitted rows before any store is contacted; only qualifying results are then charged. See the Pricing tab for the current rate.

AI

Uses GPT-6 Luna with your key, strict structured output and store: false. There is no model substitution. Advice is optional and requires human review. OpenAI's applicable service data policies still apply. Do not include secrets in the feed.

Support

When reporting a problem, include the run ID, affected theme, currency, expected behaviour and a redacted example row. Never include API keys or private customer data. VariantLens is independent of Shopify and Google.