One Instruction vs One Product: What a Run Actually Is
You are not buying products. You are buying instructions.
One sentence you type — "rewrite the description for this product," "fix every listing in this collection," "go through the whole spring range" — is one run. One AI task. One line on your bill, whether it touches one product or a batch you pick.
That is the whole pricing model. Everything below is the table, one real run, and the part we have not measured.
The three plans, in credits and AI tasks
| Plan | Monthly allowance |
|---|---|
| Starter | 800 credits/mo (~80 AI tasks) |
| Growth | 4,000 credits/mo (~400 AI tasks) |
| Pro | 10,000 credits/mo (~1,000 AI tasks) |
Read the right-hand column and ignore the left one for now. An AI task is one instruction you gave: when you sit down and decide "today I want the seasonal copy sorted," that decision is the unit — not the number of rows it happens to touch. Credits are the meter underneath, and the conversion between them is in the FAQ.
What one real run looks like
Here is a run on our own demo store, start to finish.
The instruction was one sentence:
Rewrite the product description for one product only: Acacia Round Tray. Leave every other product untouched.
That is the entire input. No spreadsheet, no column mapping, no rules engine. The scope lives inside the sentence — "one product only," "leave every other product untouched" — because that is how you would say it to a person.
The catalog had 68 products before the run and 68 products after it. Nothing was created, nothing was deleted. Exactly 1 product came out different: the description on Acacia Round Tray went from 62 characters to 364 characters.
That before-and-after count is the part worth pausing on. A scoped instruction that says "leave every other product untouched" produced a catalog where every other product was, in fact, untouched. The blast radius matched the sentence.
Be clear about how much one run can carry, though. This is a single narrow instruction on a 68-product store we own, from a prompt we wrote ourselves. It shows you what the billing unit is and that the scope held on that run — it is not a sample, and nothing here tells you what a wide instruction costs, which is the gap named further down. The version of this that would convince us is the one you run on your own catalog.
When the run finishes, you get a card that restates what changed and links straight into your Shopify Admin, so you can look at the live product page yourself rather than taking our word for it. If you want the longer version of how a run is assembled and reported, we walked through the whole loop in a separate piece.
Why "per product" is the wrong mental model
Now change one word in that instruction. Instead of "one product only: Acacia Round Tray," write "every product in the Tableware collection."
It is still one instruction. One run. One AI task. What we cannot tell you is what that wider run draws off the meter in credits — that is the measurement we have not published, and it has its own section below rather than a footnote, because it is the number you would most want before choosing a plan.
From the product page itself:
One product, a batch you pick, or all of them — you choose the scope
This is the part that trips people up when they come from tools priced per row. Per-row pricing trains you to think about your catalog as a meter that ticks. You start rationing: maybe I only fix the top sellers, maybe I skip the variants, maybe I do half now and half next month. The rationing has nothing to do with what your store needs — it is an artifact of how somebody chose to charge you.
Arvio bills the decision, not the rows. The unit you get charged for is the unit you were already thinking in. You wanted the spring collection sorted; that is one thing you wanted; it shows up as one thing on the meter.
Practically, that changes how you should scope your work. Under per-row pricing, the instinct is to keep scopes tiny. Here the instinct should be the opposite: decide what "done" looks like for a whole slice of the catalog, then say that in one sentence. Splitting the same work into a dozen careful little instructions is a dozen AI tasks instead of one.
The measurement we have not published
If one instruction sweeps a whole collection instead of a single product, what does that cost in credits? We have not measured that in a way we are willing to print. The ~80 AI tasks on the Starter line is the conversion for a typical task, and a task that sweeps a large batch is not the same shape as one that touches a single product.
So there is no arithmetic here that turns 800 credits into 800 products. Any per-product cost you derive from that table is a figure we did not give you.
To find your own number, run one small scoped instruction, the way we ran the Acacia Round Tray one, and watch what it draws down. Then run a slightly wider one and watch again. That is measured on your store instead of on our hypothetical, and you can check the results in Shopify Admin between each step.
When you should not buy this
If your catalog is small and rarely changes, you don't need us.
That is the official line and we mean it literally. A store with a couple of dozen products where you wrote every description yourself and change them twice a year does not have a problem that an AI Store Operator solves. You will spend an afternoon and be done. Buying a monthly allowance of instructions to use twice a year is a bad trade, and there is no version of the plan table above that makes it a good one.
A few other cases where you should probably wait:
You do not yet know what "good" looks like for your store. Arvio executes decisions. If the decision itself is unsettled — you are still figuring out your voice, your positioning, how you want the category pages to read — the bottleneck is not execution speed. Work that out on a handful of products by hand first, then hand over the rest.
You need every single output reviewed before it goes live anyway. Some merchants operate in categories where each line of copy has to pass a compliance read. That review is still your job, and if reviewing takes as long as writing, the run is not buying you much.
Your problem is data you do not have. If the reason your listings are thin is that nobody ever collected the material — measurements, ingredients, sourcing details — then no instruction fills that in. It is a supplier problem wearing a copywriting costume.
The merchants this actually fits are the ones who already know exactly what needs to happen across a few hundred product pages and simply cannot sit down and do them one at a time. That is the gap. We have written about what the fix looks like when it lands in your store.
Arvio: AI Store Operator — install it on the Shopify App Store. One instruction, whatever scope you choose.
FAQ
If one instruction covers a whole collection, is that one AI task or one per product?
One instruction is one run. The scope lives inside the sentence you write — one product, a batch you pick, or all of them. What we have not published is how credit consumption scales as that scope widens, so if you are planning around a large sweep, run a small one first and watch your balance.
What is the difference between credits and AI tasks?
Credits are the underlying meter and AI tasks are the human-readable conversion of it. Starter carries 800 credits/mo (~80 AI tasks), Growth 4,000 credits/mo (~400 AI tasks), Pro 10,000 credits/mo (~1,000 AI tasks). The conversion is approximate, which is the reason for the tilde.
How do I know it only touched what I told it to touch?
Write the scope into the instruction and then check the result. In the run above, the sentence said "one product only" and "leave every other product untouched," and the catalog went from 68 products to 68 products with exactly 1 of them changed. The result card links into Shopify Admin so you can open the product and read what is live rather than trusting a summary.
Can I see what changed before it goes live?
The result card restates what the run did and points you at the affected products in your own Admin. Reading the output is part of the workflow, not an optional extra — especially on the first few runs, while you are calibrating how much scope to hand over at once.
Should I split a big job into several small instructions to be safe?
Splitting is a reasonable way to build confidence early, and it is how we would suggest starting. Just be aware of the trade: several careful instructions are several AI tasks, where the same work described in one sentence is one. Once you trust the scoping, widen it.
Why won't you just tell me the cost per product?
Because we have not measured it, and a number we invented would be worse than no number. We can tell you the plan allowances and we can tell you the billing unit is the instruction. The exchange rate between a wide instruction and a narrow one is a gap in what we have published, and filling it with division would be filling it with fiction.