Every product feed tool’s website says the same six things: connect your store, optimise your feed, publish everywhere, powered by AI. The feature lists are close to interchangeable, which is why choosing one from marketing copy is nearly impossible. The differences are real, but they sit one level down, in how each tool answers a handful of architectural questions.
Below are ten questions worth putting to any tool you are considering, including this one. Two of them are questions Verintra answers badly, and they are in the list for that reason: a checklist that only asks what we happen to be good at is an advertisement, not a checklist.
1. Where does the fix get written?
This is the first question and it decides more than any other. There are two designs. A tool can write improvements back into your store’s product records, or it can leave the source alone and apply changes when the feed is exported.
Writing back means one source of truth, and it means a tool has edit access to your catalog. If you leave, the edits stay, which sounds good until you consider that so does every mistake. Layering at export means your catalog is untouched, a change is reversible by deleting a rule, and the same product can go out differently to different channels. It also means the improvements live in the tool, so leaving costs you them.
Neither is wrong. But ask which one you are buying, because the sales page rarely says.
2. How does your data get in, and what does that require of you?
Three common answers: an app installed in your store, a URL the tool fetches, or a file you upload. They ask for very different amounts of trust.
This is the first question Verintra answers badly. There is no file upload. Ingest is by URL or by push, which means if your feed is behind authentication or produced by hand as a spreadsheet, this tool cannot read it and you should look elsewhere. There is also no store app: sixteen platform guides tell you where your platform publishes its feed, and the tool reads that URL. It works with any Shopify or WooCommerce feed URL, which is a weaker claim than being connected to the store itself, and the distinction is worth pressing on with any vendor whose wording blurs it.
3. Does it score, or does it only report errors?
An error list tells you what a channel rejected. A score tells you where the whole catalog stands, including the products that passed. The second is what lets you decide whether feed work is worth a quarter of your attention at all, and a tool that only mirrors Merchant Center is giving you something you already have for free.
If it does score, ask what the score is made of and whether it explains itself. A number with no breakdown cannot be argued with, and you will need to argue with it the first time it disagrees with your instinct. If you have not seen what a scored catalog looks like, the free scan grades every product and shows the distribution, read only.
4. Is the fix list ranked, and ranked by what?
Almost every tool produces a list of problems. Very few produce it in an order you can defend. Ranking by severity sounds right and is nearly useless: a critical problem affecting eight products matters less than a moderate one affecting eight hundred.
Ask whether each group tells you how many products it touches and what the score gain would be. That turns the list into a work plan, and it takes the order of work out of a meeting. The product feed optimization guide goes through the order that usually holds.
5. When it generates text, what is it allowed to use?
Every tool now writes titles and descriptions with a model. The question that separates them is what the model is permitted to draw on. If it can answer from its own knowledge, it will eventually state a material, a fit or a compatibility that your product does not have.
That failure is worse than an empty field, and the reason is worth being precise about: an empty field fails a check, so somebody sees it. An invention passes every check, so nobody does. It reaches a shopper as a fact.
Ask whether generation is restricted to attributes already in the feed, whether output goes to a separate field, and whether you can read and reject it in bulk before a channel sees it.
6. How does category matching avoid inventing an answer?
Google’s taxonomy has roughly five and a half thousand categories. The naive implementation hands all of them to a model and hopes. The result is confidently wrong categories, and a wrong category is expensive because it decides who you are benchmarked against rather than whether you appear.
A sound implementation narrows the list first, with something deterministic that runs on your own product text, and only then asks a model to choose from the shortlist. That makes the wrong answer structurally unavailable rather than merely unlikely. Ask which of the two you are getting; the answer is usually visible in whether the tool can explain how the shortlist is built.
7. Does a fix stay fixed?
A one-time bulk edit is a task. A rule is a permanent answer. Most feed problems are patterns rather than individual products, and the products you add next month will have the same pattern.
Ask whether you can express a condition and an action once and have it apply on every sync, how many fields are writable, and whether a rule can be turned off without unpicking anything. A tool where every fix is a manual pass produces a catalog that decays back to where it started within a season.
8. Are the channel mappings real, or is it one export with a different filename?
“Publish to fifty channels” can mean fifty prepared field mappings, or it can mean one generic file and a list of URLs. The difference shows up as disapprovals on the channels whose spec differs from Google’s.
Ask how many destinations arrive with a ready mapping rather than how many are listed, and whether each one gets a hosted URL the channel can fetch on its own schedule. Ask separately about the AI shopping destinations, because their formats are new and a tool that has not implemented them will still list them.
9. Does it read Merchant Center back, or only push to it?
Pushing a feed is the easy half. The useful half is reading the disapproval reasons back, grouping them, and mapping them to the products so the next fix is informed by what actually failed. Without that you are checking a browser tab by hand and typing the findings back in.
This is a specific capability to ask about by name rather than assume from an integration logo. The disapproval reasons and their fixes is a reasonable list to test any tool against.
10. What does it cost you to leave?
Ask this before you sign, not after. If the tool layers at export, leaving means losing the layer, so the question is whether you can get your improved feed out as a file you keep. If the tool writes into your store, leaving means the edits stay but so does anything you disagreed with, and there may be no record of what it changed.
The second question Verintra answers badly: pricing on this site is not settled, and no charge is currently made. That is honest rather than reassuring, and if a fixed annual cost is part of your decision today then that is a reason to wait or to look elsewhere. Be equally direct with any tool that will not put a number in writing.
What this list is not
It is not a comparison table of named products. Building one would mean asserting what two or three competitors do and do not do, from their marketing pages rather than from running them, and those assertions would be wrong often enough to be worthless to you. The ten questions are more useful precisely because you get to put them to each vendor yourself and hear how they answer.
If you want to see what the answers look like on a real catalog rather than in a sales call, run a free scan on your own feed URL first. It takes about two minutes, it needs no account, and it writes nothing back. Whatever you decide afterwards, you will be deciding with your own numbers.
Frequently asked questions
Do I need a feed tool at all?
If your catalog is small, stable and already complete, no. A spreadsheet and a supplemental feed will hold. Tools earn their place when the catalog changes faster than anyone can keep up with by hand, which for most stores is somewhere past a few thousand products or past a second sales channel.
What is the difference between feed management and feed optimization?
Management is the plumbing: getting data in, mapping it per channel, keeping the export fresh. Optimization is improving what the fields say. Most tools do both and the words are used loosely, so it is worth asking which half a given product is actually strong at.
Is a store app better than a feed URL?
An app can read data a public feed leaves out, which is a real advantage. It also needs permissions, needs updating, and breaks with platform releases. A URL asks for nothing and can read only what your platform publishes. Which is better depends entirely on whether the fields you need are in the published feed.
How long before feed work shows up in results?
Approved product counts move within a feed refresh. Impressions follow within days. Revenue takes a fortnight or more, because the feed refresh, the landing page crawl and the campaign learning do not finish together. Any tool promising same-week revenue is describing a coincidence.
Does the tool matter for AI shopping assistants?
It matters more there than for Shopping, because an assistant has no image and no listing grid to fall back on: it answers from the structured text alone. Ask specifically which AI destinations a tool exports to and in what format, since the formats are recent enough that listing them is easier than implementing them. The OpenAI ACP product feed is a concrete thing to ask about by name.
