Automatic vs manual mode.

All four generative AI features in SintoData default to automatic mode per feed — writes happen immediately when output is clean, only filling empty fields and never overwriting content the merchant already set. Each can be toggled to manual mode for merchant review instead. Here's how to choose, and what you trade off either way.

The Two Modes

What automatic and manual actually mean.

SintoData's AI features fall into two categories: safety gates (always automatic, never generate content) and enrichment features (configurable per feed).

Automatic Mode Manual Mode
AI writes to Shopify? Yes — applied immediately on each sync, no human review step. No — suggestions are held in a review queue. You approve or reject each one before it touches your store.
When is the AI called? Automatically on each feed sync, for eligible products (new or drifted rows). No manual step needed. Automatically on each feed sync — same as automatic mode. The difference is that output is held in your review queue instead of being written immediately. You can also trigger a scan manually via the "Run" button.
Time commitment Zero after initial setup. Runs unattended. Requires periodic review sessions. How often depends on your catalog size and feed update frequency.
Risk profile AI fills empty fields without human eyes. Mitigated by gap-fill enforcement (writes only where the field is empty), plus guardrails (vocabulary allow-lists, confidence thresholds, HTML-balance checks). You inspect every suggestion before it's applied. The gap-fill check runs on approval too, so even an explicit approval cannot overwrite content you already set. Good for building trust or when you want to read first.
Best for Recommended default. High-volume catalogs, features with strong guardrails, or after you've confirmed the AI's accuracy on your catalog. Clean output is applied immediately — only flagged results need your attention. Small-to-mid catalogs where you want to inspect every AI suggestion. Good for building trust in a new feature before switching it to automatic. Also useful for feeds with inconsistent data where you expect frequent guardrail trips.
Per-feed? Yes — one feed can be automatic while another is manual for the same feature. Yes — independently configurable per feed.

Quick Decision Guide

Go manual if

You have under 500 products and want to inspect every AI suggestion before it reaches your store, or you're setting up a feed with inconsistent product data where you expect frequent guardrail trips.

Stay automatic if

You have 500+ products making manual review impractical, or you trust the AI's guardrails (vocab allow-lists, confidence thresholds, HTML checks) to catch genuinely risky output. This is the recommended default for most feeds.

Mix modes

Keep automatic on features with strong guardrails (Attribute Extraction, Category Matching, SEO Metadata), switch to manual on Translation if brand voice matters or you want to review wording. Toggle per feed — your main supplier can run fully automatic while a secondary feed stays manual.

Per-Feature Breakdown

Automatic vs manual for each of the 7 AI features.

Three are always-on: Column Mapping (with a required approval gate on first map), Pricing Anomaly Detection, and Duplicate Detection. The four generative features default to automatic with gap-fill guardrails — they fill empty fields and never overwrite content the merchant already set. Each can be toggled to manual per feed.

AI Column Mapping

Always Automatic

What it does

Maps your supplier's file headers to standard catalog fields (SKU, price, quantity, title, etc.) on first setup. Detects header drift when suppliers change column names between syncs.

Why it's automatic: Column mapping is a prerequisite for sync — without it, no data flows. The AI only produces a mapping suggestion, never writes data directly. The mapping is always held for your approval the first time. Drift remapping also requires your review before being committed.
What to watch: If your supplier uses ambiguous headers, the AI may map them incorrectly. Always review the initial mapping before approving. A mis-mapped "price" column would flow through to your store.
Recommendation: No choice to make — column mapping is always automatic with a required approval gate on first map and drift. This is the safest design: you get the AI's speed but retain the final say.

Pricing Anomaly Detection

Always Automatic Deterministic

What it does

A deterministic (non-AI) safety net that compares every price change against the last-known price. Suspicious swings are held back and flagged for your review.

Why it's automatic: This feature never generates or modifies a price — it only decides whether to trust the supplier's price. It's a gate, not a generator. False positives (legitimate price swings flagged as anomalous) are easy to approve in one click. False negatives (missed anomalies) would be far worse.
What to watch: If your supplier has genuinely volatile pricing (e.g. market-based pricing), you'll get frequent alerts. You can adjust the ratio threshold per feed to match your supplier's normal price range. Anomalous prices are held back, so they don't reach your store until you approve them.
Recommendation: Leave on. It can't hurt your catalog — it only withholds suspicious prices for your review. The only scenario to disable it is if your supplier prices swing wildly by design (e.g., live market feeds) and you want them applied without any gate.

Duplicate Product Detection

Always Automatic | Read-Only

What it does

A daily shop-wide catalog scan using string-similarity algorithms to surface likely duplicate product listings. Makes zero Shopify writes — ever.

Why it's automatic: This feature is pure read-only. It never writes to Shopify. It only surfaces candidate duplicates for you to review in your admin. There is no "approve" button — you fix duplicates yourself. There's no downside to leaving it on.
What to watch: The similarity threshold is calibrated at 0.96 to avoid false positives (e.g., "Wireless Mouse" vs "Wireless Mouse Pro" won't trigger). If you see false positives, you can dismiss them. Dismissed pairs won't re-trigger on future scans.
Recommendation: Always leave on. It costs nothing, writes nothing, and occasionally surfaces real duplicates you'd otherwise miss — especially duplicates that span two different supplier feeds.

Attribute Extraction

Automatic Default

What it does

Extracts color, size, and material from product titles and writes them as structured Shopify metafields so your storefront filters work automatically.

Automatic mode pros: Filters populate without any manual work. Every new product gets attributes extracted on its first sync. Ideal for high-SKU catalogs where manually tagging 2,000+ products is impractical.
Automatic mode cons: AI may misidentify color/size/material from ambiguous titles (e.g. "Apple" as a color, "XL" in a product name that isn't a size). Off-vocab values are always flagged and never written, but in-vocab misidentifications could reach your filters.
Manual mode pros: You approve each extraction before it's applied. Catch misidentifications before they reach your storefront. Build trust in the AI's accuracy on your specific catalog.
Manual mode cons: Requires periodic review sessions. If you add 50 new products per week, you'll need to review 50 extractions weekly. Doesn't scale well past a few hundred products.
Recommendation: Leave on automatic. The vocabulary allow-list provides a strong safety net — in-vocab values are auto-written, off-vocab values are always flagged regardless of mode and never reach your storefront filters. Switch to manual only if your product titles use highly ambiguous language.

Product Translation

Automatic Default

What it does

Translates product titles and descriptions to any Shopify Markets locale, writing to Shopify's native translations API. The AI fills empty locale fields — if you already have a translation for a product, it is never overwritten, whether you wrote it yourself or paid a translator.

Automatic mode pros: International catalogs stay translated without manual effort. Every new product or updated description gets translated on sync — but existing translations, including human or paid ones, are never overwritten. No backlog of untranslated products.
Automatic mode cons: AI translation errors (mistranslated technical terms, brand names, or idiomatic phrases) go live without review on newly translated products. An HTML-balance guardrail blocks translations that introduce broken tags, flagging them for your review, but mistranslated text has no automated safety net. Existing translations are always protected.
Manual mode pros: You review every translation before it reaches international shoppers. Edit awkward phrasing, fix mistranslated product specs, and ensure brand voice is preserved. The gap-fill check still protects existing translations even on approval.
Manual mode cons: High time commitment. Translating 500 products into 3 locales = 1,500 manual reviews. Best suited for small catalogs or when you only need a few locales.
Recommendation: Translation is the feature where reading first matters most, since mistranslations have no vocabulary safety net. But under gap-fill, automatic mode cannot overwrite translations you already have — so the risk is only on new products, not your existing catalog. Leave it on automatic if your product titles are straightforward and you don't have existing translations to protect. Switch to manual if brand voice or technical accuracy is critical for your international shoppers.

Category Matching

Automatic Default

What it does

Suggests a Shopify product type for each product based on its title and description. The AI fills empty product type fields — if you already set a product type, it is never overwritten.

Automatic mode pros: Products are categorized automatically as they enter your catalog. No backlog of uncategorized products. Gap-fill enforcement ensures the AI never overwrites a product type you already set.
Automatic mode cons: A miscategorized product gets the wrong product type without review. The 0.5 confidence threshold discards weak suggestions, but confident misclassifications can still happen (e.g., "monitor arm" classified as "desk accessories" vs "monitor mounts").
Manual mode pros: You verify each product type match before it's applied. Ensure every product gets the right type. Reject weak matches easily. The gap-fill check still protects existing product types even on approval.
Manual mode cons: Moderate time commitment. If you add 50 products per week, you'll review 50 category suggestions weekly. Faster than reviewing translations (one suggestion per product vs one per locale).
Recommendation: Leave on automatic. Gap-fill enforcement ensures the AI never overwrites product types you already set, and the 0.5 confidence threshold discards weak suggestions. A miscategorized product is unlikely and easy to spot. Switch to manual only for feeds with highly ambiguous product titles.

SEO Metadata

Automatic Default

What it does

Generates page titles and meta descriptions for search engines. Never touches your visible product copy — only the metadata that appears in Google/Bing results and browser tabs.

Automatic mode pros: Every product gets SEO-optimized metadata without manual effort. Ideal for large catalogs where hand-writing meta descriptions for 5,000 products is unrealistic. Safe because it never modifies your visible product title or description.
Automatic mode cons: AI-generated meta descriptions may not capture your preferred keywords or brand voice. SEO is strategic — you might want different keyword emphasis than the AI chooses. No vocabulary guardrail (each description is unique).
Manual mode pros: Tailor each meta description to your SEO strategy. Add target keywords, adjust tone, and ensure brand consistency before publishing. Edit the AI's suggestions to perfect them.
Manual mode cons: Moderate time commitment. Reviewing 50 SEO descriptions per week is manageable, but 500 is substantial. SEO metadata only affects search engines, so errors are less visible than translation or attribute errors.
Recommendation: Leave on automatic. This is the safest of the four generative features — it only touches search engine fields (never your visible product copy), and Shopify clamps over-length values server-side. If SEO is a core strategy and you want precise keyword control, switch to manual and review each suggestion. Otherwise, automatic mode with periodic spot-checks works well.

Strategy

A recommended progression.

All four generative features default to automatic — they start working the moment your feed syncs. You can dial back to manual on specific features or feeds as needed, or stay fully automatic. Here's a typical onboarding path.

Phase What to do Timeframe
Day 1 Set up your first feed. Review the AI column mapping. Run a dry run. All enrichment features are automatic by default — they'll apply on the first live sync. The 3 always-on safety features (Column Mapping, Pricing Anomaly Detection, Duplicate Detection) require no configuration. ~15 minutes
Week 1 Spot-check what the AI wrote after your first few syncs. Check extracted attributes on a few products, verify product type assignments, and glance at SEO metadata. If anything looks off, switch that feature to manual for that specific feed — no need to go fully manual on everything. ~5 minutes per check
Week 2 Review Translation output for your target locales. While gap-fill protects existing translations, mistranslations on new products still go live without review. If brand voice or technical accuracy matters, consider switching Translation to manual. ~10 minutes per locale
Week 3 Add your remaining feeds. Each inherits the same default (automatic on all four features). If a new feed has inconsistent supplier data, flip individual features to manual on that feed only — your main feed stays automatic. Adjust per-feed modes as your trust level differs per supplier. ~5 minutes per feed
Ongoing Spot-check automatic feature output periodically (once a month, or when you notice something off in your store). Review your Alerts panel — flagged rows, blocked writes, and anomaly alerts surface there. Most merchants settle into full automatic with monthly spot-checks. ~5 minutes per check

Ready to configure your feeds?

Walk through the step-by-step setup guide, or jump straight to pricing.

Setup Guide → View Plans →