For Shopify stores with more products than you can edit by hand. SEO titles, meta descriptions, image alt text, attributes and categories, written from each product's own description. You approve everything before it is written.
Built for a Shopify catalogue of about 50,000 SKUs, and run on it before it was sold to anyone. Why we built it
One product with a blank SEO title is easy to fix by hand. A thousand of them, with meta descriptions and image alt text as well, is not.
The hard part isn't the storefront. It's getting found — and a thin product page with no meta description and no alt text gives a search engine very little to rank.
Shopify will not write those fields for you, and it gives you no bulk editing path that fills them in. So the usual alternative is exporting a CSV, pasting it into a chatbot, pasting the answers back, and hoping nothing landed in the wrong row. Altfield proposes every value inside your Shopify admin, records what each one replaced, and puts the whole lot back in one click if you change your mind. One run covers your whole catalogue, and each product counts once.
Every proposal sits beside the product it belongs to. Tick what you want, leave the rest. Anything the model is unsure of is flagged and not pre-selected.
Pick the fields that matter to you. Unticked fields are never sent to the model, so a narrower run is a cheaper one. Product names and descriptions stay off unless you ask for them — shoppers read those.
Every change records its previous value before it is replaced. Reverting restores those exact values, makes no AI calls, and keeps working on the free plan. Every run keeps its own record of what it changed, so you can show exactly what was done to the catalogue and when.
Switch on autopilot and products are filled in as you add them, so a catalogue stays complete instead of drifting back out of date. Off by default, chosen per field, and it only ever fills fields that are empty.
It fills gaps. Replacing a value you wrote yourself is a separate choice you make on a run.
Everything Altfield wrote stays in your store if you downgrade, cancel or uninstall. It is your data.
It asks for product and file access only. No customer scopes, no order scopes, nothing to leak.
Ran it on about 180 products. Most of the suggestions were actually usable, a couple of the jewellery ones were a bit flat but nothing a quick edit didn't fix. The revert button worked when I realised I'd left product attributes on by mistake.
We had a lot of products with no alt text after a supplier feed update. It filled them in from the existing descriptions without me having to export anything. Approval step is necessary — I wouldn't have been comfortable letting it write straight to the store.
It's not magic and the output still needs a human eye, especially if the product descriptions are thin to begin with. That said, it's cut the time I spend on the boring meta work by a good bit. I like that nothing gets written until you tick it.
Testing turned up the same weak spots they mention, and two more worth knowing before you install: category suggestions are wrong often enough on specialised products to be worth checking every time, and on unusual products the meta descriptions come out generic and need an edit. Standard products need far less. Both are an edit, not a rerun.
Used it and want to say something, good or bad? info@nkd.ie.
The only difference is how many products you can process. A product counts once per run.
25 products a month
750 products a month
6,000 products a month
25,000 products a month
750 products a month is 750 fields you didn't edit by hand, for less than one hour of freelance work.
Paid plans include a 7-day trial with a smaller allowance, so you can judge the output before being charged. Billing is handled by Shopify and appears on your Shopify invoice — Altfield never sees a card.
To generate a proposal, Altfield sends a product's own text to Anthropic's API — title, description, type, vendor, tags, existing SEO fields and existing metafields. Where a description is too thin to work from, it also sends one product photograph.
Install it, run it on your own catalogue, and look at what it proposes before a single field changes.