Everything your feed deserves.

All products is your single, up-to-date catalogue inside FeedGraph. Search, filter and sort across everything you sell, open any product to see and edit every field, and switch tabs to work through products that need fixing, restore something from the 30-day trash, or read the full change history. Because every feed you publish is built from this list, fixing something here fixes it everywhere.

The "Needs fixing" tab collects every product that would be rejected or would underperform, and groups the problems by type — all the missing descriptions together, all the missing sizes together. Fix them by hand, pick from the values the platform accepts, or let AI draft the corrections for you to approve. It works on a sample first so you can see the AI's output on a few products before spending credits on thousands.

Google Merchant Center disapproves products that break its rules, pulling them from your Shopping ads. The recovery panel inside All products collects every rejected product, explains the real reason for each rejection in plain language rather than Google's error code, and lets you fix and resubmit — turning a confusing error list into a simple to-do list.

Every platform has rules: a title can't be too long, a price must be present, an image must meet quality standards. Quality rules runs these checks across your whole catalogue, for Google, Meta and ChatGPT Ads. Some results are blocking — the product cannot go live until it is fixed — and some are warnings, meaning it will go live but may underperform. Crucially, a destination only blocks on the fields it actually requires: Google demands colour and size on apparel, OpenAI's commerce schema treats them as optional, and a ChatGPT feed is never refused for a rule that belongs to Google. The rules also change by category: fill in a product's category and FeedGraph tells you immediately what that category now requires. Advanced users can adjust the rulebook.

Two different problems, one screen. First, the same item ingested twice — once from Shopify, once from a spreadsheet — matched by confidence tier, with the connected source winning as the copy to keep and variants protected from being merged away. Second, products that are genuinely different but share a title, description or URL, which is a real SEO and Merchant Center problem. FeedGraph shows what differs between them and can draft a rewrite.

A single t-shirt comes in S, M, L and XL. If most sizes are sold out, advertising that product wastes money and annoys shoppers who cannot buy their size. Variants reads your size-level stock, flags products with critical gaps, and holds those products out of your live feeds. Because it is evaluated every time a feed is built rather than stored as a flag, a product comes back on its own as soon as sizes are restocked — nothing to remember or undo.

For Shopify, Google, Meta, ChatGPT Ads, Amazon, TikTok, the schema library lists the product details that platform expects, which are required versus recommended, and how they must be formatted. It is a reference: Google and Meta are live output channels today, ChatGPT Ads produces a validated feed file you connect yourself, and Amazon and TikTok are documented here ahead of their output shipping. It is also where the differences become obvious — ChatGPT Ads uses OpenAI's own commerce vocabulary (item_id, image_url, price as "79.99 USD") rather than Google's field names, and its product category is a free-form path rather than the Google taxonomy. When you are unsure why a platform wants something a certain way, this is where you look it up.

Most catalogues imported from Shopify or a legacy platform have gaps: missing brand names, vague descriptions, wrong category, no apparel attributes. AI enrichment fills those gaps in bulk, across thousands of products in a background job. Every run shows a preflight quote of what it will cost before it starts, runs on a sample you approve first, and records every change in the activity log so you can revert any field.

Strong product images move click-through rate. FeedGraph generates white-background heroes for Google Shopping and lifestyle context shots for Meta, from your own product photography. Product design and model identity stay locked to what you uploaded, while pose and camera are free — a frozen pose is what makes generated imagery look pasted in. Every image is checked against the product it belongs to, stored in your catalogue, and referenced automatically in your feeds.

Short-form video is the format paid social rewards, and shooting it per SKU does not scale. FeedGraph generates product video from your existing photography, with the same fidelity guards as images: the product itself is preserved, and you see the before and after. Three tiers let you match spend to the product — the cheapest for bulk catalogue coverage, the highest for hero products and paid social. Priced per 2 seconds so a longer video costs proportionally more.

Not every creative needs a model. Ad creatives composes platform-ready ad images on the server from templates: drop in a product, set the text and your brand colours, and export at the size each placement wants. It costs no credits at all, because nothing is generated — it is composition, not inference. If you do want the product lifted out of its background first, that step is available and metered separately.

The assistant is connected to your live workspace, so it answers from your data rather than in general terms. Ask it to analyse feed issues, compare product performance across channels, work out why specific products are blocked from export, or suggest what to enrich first. Three depths let you control cost: Fast for quick questions, Balanced for analysis, Thorough for multi-step strategy. Conversations are saved so you can pick one back up.

People are entitled to know when what they are looking at was made by a machine, and ad platforms increasingly expect it. Every image and video FeedGraph generates carries a "Generated by AI · FeedGraph" mark composited into the pixels, plus the IPTC digital source type that identifies synthetic media in the file itself, so it travels with the asset. AI-written product fields are flagged at field level against the value they carry, and you can attach an AI declaration to a feed export per destination.

A feed is the list of products you send to a platform. Channel feeds is where you create and manage them — one for Google, one for Meta, one for ChatGPT Ads, or a custom output shaped however you need. Describe the feed you want in plain language ("all in-stock products over $20 with a photo") and FeedGraph selects the matching products and formats them correctly. Every export path blocks an empty feed and enforces a per-product validation floor, so an incomplete product cannot slip into a live channel. The ChatGPT Ads tab is labelled "feed file only", because FeedGraph builds and hosts that feed while you connect it in OpenAI's Ads Manager — the tab says so rather than letting you assume otherwise.

If Channel feeds is the quick way, Advanced feed tools is the precise way. It walks you through building a feed step by step — choose your products, map your fields to the platform's required attributes, apply any final transformations, and get a finished file. The same screen converts a feed built for one platform into another's format, previews how your fields will project into a channel, and manages per-channel adapters. It is also where a ChatGPT Ads feed is built and published to a hosted URL, and where the OpenAI crawler check and the context-hint drafter live. Use it when you want exact control.

Every platform arranges product information differently — Google wants particular column names, Meta wants different field names and different image rules, and OpenAI's commerce schema for ChatGPT Ads shares almost no field names with either. Format conversion, in Advanced feed tools, takes a feed built for one platform and reshapes it for another, so putting your existing Google feed on ChatGPT — or moving it to Meta — does not require rebuilding it from scratch. Columns come from the destination's own published specification rather than a renamed copy of the source's, and before you convert anything it tells you which columns carry across, which get rewritten, which are derived, and which nothing can supply — naming the last group instead of shipping them empty. When the destination is a feed-file-only channel, it says so up front.

A Segment is a saved group of products. Define it by rule — "everything with a margin above 30%", "this season's new arrivals" — and it updates itself as products come and go. Or hand-pick the products for a specific campaign, described in plain English, by attribute, or one by one. Either kind drops into any feed, campaign or report without rebuilding the selection, and any segment can be published as its own live feed. Segments also drive the custom labels on your products, which is how you target a subset in Google Ads — and the same labels populate the ads metadata in a ChatGPT Ads feed, which is what its ad-group product filters read.

Sometimes you need to hand a specific product list to a platform, a partner or a colleague. From the Feeds tab in Segments, publish any saved segment as a live, token-protected URL. Whoever has the link always sees the current list, so there are no spreadsheets going back and forth. Refresh the token or revoke the link whenever you need to.

Push to channels is the single place products go out. Run it as a dry run to see exactly what would be sent and what would be rejected, then push for real. Retries are handled with backoff, every attempt is recorded, and — because pushing is a write to an outside platform — it goes through the same approval gate as every other platform write. Nothing leaves without you approving it. Pushing is Google and Meta only: ChatGPT Ads is served from a hosted feed URL instead, because OpenAI's Ads API is not open to self-serve accounts.

ChatGPT Ads runs on a product feed, and OpenAI’s commerce schema is its own: item_id, image_url, price written as "79.99 USD", and a free-form category path rather than the Google taxonomy. FeedGraph builds that feed from the same catalogue that feeds Google and Meta, validates it against OpenAI’s spec rather than Google’s, and publishes it to a hosted URL — which matters, because ChatGPT expires a product 14 days after it drops out of a snapshot, so a one-off CSV upload quietly decays. Before you launch, FeedGraph checks that your robots.txt and your image host actually let OpenAI’s crawlers through, and drafts the ad-group context hints that are the only relevance lever ChatGPT gives you. FeedGraph produces and hosts the file; you connect it in OpenAI’s Ads Manager.

Over time your store, your ad platforms and FeedGraph can drift apart — a price changes in one place but not another. Sync Center compares them continuously, shows you the mismatches field by field, and helps you bring everything back into line. It is the feature that stops "my website says one price and Google says another", and it is where you control how often each connection refreshes.

FeedGraph registers Shopify webhooks for instant product updates, runs hourly incremental polling to catch anything webhooks missed, and performs a daily Bulk Operations reconcile that catches deletions and gaps. All three run together, so no single failure loses a change. FeedGraph is the source of truth: local edits flow back out to Shopify automatically every two minutes, and FeedGraph wins any conflict.

TikTok is next on the channel roadmap alongside Amazon. Catalogue field mappings exist and are documented in the platform schema library, but the channel stays closed until the output path ships. Feeds for Google and Meta, and custom output in any format, are available today.

The Advertising overview connects ad spend to sales, product by product, across Google and Meta. It sorts your catalogue into clear groups — the products earning well, the ones quietly draining budget, and everything between — and recommends concrete next steps: spend more here, pull back there, fix this listing. The buckets work as filters, so you can go straight from a finding to the products behind it.

Ad performance is the detail behind the overview: per-SKU impressions, clicks, cost, revenue and ROAS, segmented by day and joined to your live catalogue on the real SKU. Because it is day-level rather than a single total, you can see when something changed rather than only that it did. Products are gated on eligibility, so a SKU that was never advertised is not counted as a poor performer.

Performance Max is Google's most powerful ad type, but it only works well if your product data is strong. Google Ads Studio scores your readiness across titles, images, pricing, labels and more, gives you an overall score, and lists exactly what to fix. It can generate a supplemental feed that enriches your main Google feed automatically, and it executes real changes against the Google Ads API — pausing or enabling campaigns and asset groups — each behind an approval, each reversible.

Meta has its own requirements for good ad performance: high-quality images, correct pixel tracking, complete product details. Meta ads checks your catalogue against all of them, shows how ready you are by field and by category, and points out exactly what to fix so your products perform in dynamic product ads and Instagram Shopping.

Not every product deserves the same ad budget. Smart bidding reads per-SKU ROAS, margin, stock cover and return rate, then sorts products into actionable groups and derives a target return for each from that group's own measured performance. From there you can act three ways: tag products with custom labels so your existing campaign rules pick them up, hold the money-losers out of your feeds, or set target ROAS and daily budgets directly on your live Google Ads campaigns. The last one shows you the current value against the new one, limits any single change to 50%, needs your approval, and records the previous value so it can be rolled back.

Amazon is next on the channel roadmap. The field mappings and listing-completeness scoring already exist in FeedGraph and are documented in the platform schema library, but the channel is deliberately closed until the full output path and Amazon Advertising integration are ready — surfacing it early would promise an export we cannot honour. Feeds for Google and Meta, and custom output in any format, are available today.

The AI report generator writes a business report across ads, feed and inventory — customise what it covers and schedule it to arrive on a cadence. Alongside it: Sales velocity shows how fast each product is selling and when it will run out; Quality check history tracks whether fewer products are getting rejected over time; Sync activity logs every time data moved in or out, so you can confirm everything is flowing.

Every edit, import, export, push, AI run and settings change is recorded with the person behind it, the values before and after, and the reason. AI-originated changes are labelled as such and kept distinct from human edits. Any field change can be reverted, and the revert is itself logged, so there is no way to quietly undo history. Entries are hash-chained: each one is hashed together with the one before it, so changing or removing any entry breaks every hash after it — one click checks the whole chain, and the log exports to CSV with the hashes included so it can be verified independently.

A tool with your Google Ads and Meta credentials can do real damage, so FeedGraph starts unable to write to them at all. Turning Write Actions on is a deliberate, per-workspace decision, and even then nothing goes out on its own: each change is staged with a plain-English summary of what will happen, and reaches the platform only when someone approves it. Every outbound write in the codebase is forced through one enforcement point, so there is no side door — if the plumbing is missing, the request fails rather than proceeding.

The external API gives your own tools read access to your catalogue: products, a product's full graph, channel projections and adapter runs, plus the ability to trigger an adapter run. API keys are stored as SHA-256 hashes, carry explicit permission scopes, and can be given an expiry, rotated or revoked without disrupting other integrations. Webhooks are signed with HMAC-SHA256 so your systems can verify authenticity and reject spoofed requests. Use it to connect BI tools, automate catalogue workflows, or feed AI agents live product data.

Permissions are per module: catalogue, feeds, segments, creatives, advertising, reports, connections, billing, team and settings, each with view and manage levels. The boundaries are enforced server-side, so a hidden menu item is not the only thing standing between someone and an action they should not take. Sensitive changes — team access, connected accounts, billing — require a second factor.

The Security screen sets up a second factor, required before changes to team access, connected accounts or billing. The Privacy screen shows what this workspace has agreed to and when, how long each class of data is kept and by what rule, and where requests people make about their own data are handled. Both are driven from a single declared data inventory, so what the screen says is what the retention job actually does.

FeedGraph uses a credits system for AI operations. The billing screen shows your plan, your remaining balance and where credits have been spent, down to the individual operation. Before any large AI job you get a preflight quote of what it will cost, and per-run and per-day ceilings mean a batch job cannot quietly empty your balance. Top up or change plan whenever you need to.