Generating content

Every piece of AI content is made by a “job”. Creating one takes two steps, and nothing is generated — and no credits are spent — until you press Start on the second.

The two steps exist so that you always see the price before you pay it. Step 1 is the wizard where you describe what you want. It ends with a draft. Step 2 is the draft’s own page, which tells you exactly how many credits the run will cost and holds the Start generation button.

Before your first job

Three things need to be in place:

• Your shop is connected and the connection test is green.
• The catalog has been synced at least once. Generation runs against the imported copy of your catalog, so products, categories, languages and media have to exist here first.
• Your shop has credits. You can build a draft with an empty balance, but you cannot start it.

Step 1 of 2: setting the run up

Where jobs start

You do not start from a generic “new job” button. You start from the task you want, because the wizard is different for each of them.

1. Open your shop.
2. In the left sidebar, pick the task under Products, Categories or Media — for example Products › Descriptions, Products › Meta texts, Categories › Category images or Media › Alt texts.
3. You land on that task’s own page, listing its previous runs.
4. Click the button in the top right: New Descriptions job, New Meta texts job, and so on.

What each task does is described on The generation modules. If a task you expected is missing from the sidebar, it has not been enabled for your shop — that is a setting only the platform operator can change.

The page header says which step you are on, for instance: “Step 1 of 2: choose model, languages and products. Nothing is generated yet — you get a summary with the exact cost first.”

What gets generated

The first card. It holds up to three things.

Template — a dropdown whose first entry is Shop configuration. That is the “no template” choice and means: use the settings from Settings › Configuration. Below it are your saved presets for this task. Picking one reloads the options on this page so you can see what it will do. See Templates.

The task’s own options. Image and alt-text tasks put their whole setup here — the media source, the image chain, the prompt list, the reference-image upload. Everything on The generation modules.

Also generate … — only on the product-description wizard. Tick it and a second task runs in the same job, using the description that was just written as its input (“Runs in the same job and uses the freshly generated content as input.”). This is how you get a description and its matching meta texts in one pass, and the meta texts are better for it, because they are written from the new text rather than the old one. The tick box gets its own Template dropdown.

If the checkbox is greyed out, no AI model available to your shop can do that second task.

On tasks that need to look at pictures you will also see: “This task looks at your product pictures, so only AI models that can read images are offered below.” That is a note, not a problem — the model list further down is already filtered for you.

How the AI should write (optional)

A collapsed section labelled — click to open. Inside are the instructions the AI is given: “These instructions come from your shop settings (or the template you picked above) and work as they are. Anything you change here applies to this one run only.”

You can ignore it. It is filled in and working. Open it if you want a one-off tweak for this run without changing your saved settings. If you have expert mode on, it starts open and shows the raw prompts as well.

The three tasks with their own options card — Product images, Image generator and Alt texts — do not have this section; their settings live in that card above instead. Category images does have it, which is why its watermark settings can be edited here. See The generation modules.

Writing style

A text box prefilled with your shop’s house style from Settings › Configuration. Change it here and the change applies to this run only.

It only appears on tasks that actually use it — a pure image run does not show it.

AI model

A dropdown: “Models differ in price and in how they write. If you are unsure, keep the recommended one.”

Each entry reads like Name — Provider · recommended · 3 credits per item, or · no credit cost when the task is free in your shop. So you can choose on price without leaving the page. The recommended model is preselected.

Only models that can do this particular job are listed. A picture task offers image models, an alt-text task offers models that can look at images, a text task offers text models. You cannot pick something that would fail.

Languages

One checkbox per language that came in with the catalog sync. Your shop’s main language is ticked and carries a default badge.

Every extra language multiplies the run: one product in one language is one item, and items are what you are charged for. Ten products in three languages is thirty items.

The Image generator task has no language picker at all — a picture has no language. FAQ suggestions goes the other way: it writes exactly one language per job, so its picker is radio buttons rather than checkboxes, and another language means a second run.

Choosing what to write about

Products

A two-column picker. On the left, Results; on the right, Selected for this job. Until you pick something the right side says: “Nothing selected yet — tick a filter and ‘Add all matching’, or click individual products in the results list.”

• Search by product number or name… — type and the results list updates. Click a result to add it.
• Manufacturers and Categories — two filter boxes above the search. The categories one is a tree you can expand; both have their own little filter field (Filter manufacturers…, Filter categories…) for long lists.
• Add all matching products — appears as soon as you tick a filter, together with a count of how many products match. This is the bulk tool: tick a category, press the button, and the whole category is in the job.

The counter at the top right of the card always tells you how many products are selected. Watch it — it is the number your cost estimate is built on.

Categories

Category tasks show a tree instead: “Pick the category landing pages to write for. Folders, links and inactive categories are not listed.” Tick the categories you want, then use the three options underneath:

• Include subcategories — ticking a parent also takes everything below it.
• Only categories that contain products — skips empty structural categories, which rarely need a text.
• Skip categories that already have a description — the safe way to fill gaps without overwriting work you already did.

FAQ suggestions has no third option, because an FAQ entry is not a description; instead Only categories that contain products starts out ticked there, and next to the filter box sits an All categories button that selects the whole tree (clicking it again clears the selection).

Media and freeform

The Alt texts task chooses its subjects through its Media source setting; when the source is “Product images” you also get the normal product picker. The Image generator has no picker at all — your prompts are the selection. Both are covered on The generation modules.

Test generation

A small panel: “Preview the output for one selected product before starting the job — nothing is saved.”

1. Select at least one product above (category tasks preview the first ticked category, so they need no product).
2. Pick a Product and a Language.
3. Press Run test.

The result appears right there. Nothing is stored, nothing is sent to your shop, and it costs no credits. Use it liberally: it is the cheapest way to find out that your writing style is too formal, or that a product is missing the information the AI would need.

FAQ suggestions answer this panel with a compact table of proposals, marked “Unverified — the real run checks category entries against every product.” A test run does the writing pass only; the checking pass that drops or narrows a category entry is what the real run adds. So read the table as “this is the kind of thing it would write”, not as the set of suggestions you will get.

What this will cost

The last card before the button. It tells you the balance (“Your shop has … credits”), and if some credits are tied up in runs that have not finished, how many can actually be spent right now. Then the rate: “<model>: N credits per item”, and the promise “You will see the exact total for this run on the next page, before anything is generated.”

If the task is free in your shop it simply says This task currently costs no credits.

Owners and Managers get a link from here to the credit ledger, or to How to get more credits when the balance is empty.

Continue to summary

Press Continue to summary. Next to it the app repeats the reassurance: “No credits are used yet — the next page shows the total and has the Start button.”

A draft job is created and you are taken to it, with the message “Job saved. Review it below, then start generation.” Nothing has been generated. Nothing has been charged.

Step 2 of 2: the draft page

This is the job’s own page. Its status badge reads Draft and the progress bar sits at zero with the badge Not started.

What this run costs

A big number: “about N credits”, and below it your balance and how much of it is free to spend. The footnote matters: “This is an upper estimate — items that are skipped are not charged.” If you ticked “skip items that already have …”, the real bill is usually lower.

If you cannot afford the run, the card turns amber and says by how much: “That is N credits short, so this run cannot start yet. Add credits, or discard this draft and pick fewer items.” Owners and Managers get a link to How to get more credits.

What will be generated

Counts as badges (Products, Categories, Image prompts, Languages) plus the actual list of products, categories or prompts. Scan it. This is your last chance to notice that a filter caught 400 products instead of 40.

Start generation

Three buttons:

• Start generation — asks to confirm first: “Start this run now? About N credits will be used.” Confirm and the job is queued.
• Edit job — reopens the Step 1 wizard with everything filled in as you left it. Change what you like and press Save changes; you come back here with a new estimate. Use this rather than starting over.
• Discard — throws the draft away. It asks “Discard this draft? This cannot be undone.” Nothing was charged, so nothing is lost but the setup.

A run starts as a whole: there is no way to start half a draft. If you want to try a handful of items first, discard this draft, build a small one, and create the big job afterwards.

If you only have viewing rights, the page tells you: “This job is a draft. Someone with job permissions must review and start it.”

While the job runs

The page updates itself — leave it open and watch, or come back later. The progress bar shows done / total items and, if anything went wrong, N failed.

Items are processed one by one and each row’s Status badge moves along:

You are charged per completed item. Failures and skips cost nothing.

Stopping a run

While a job is running, the top right of the page has Stop job. It confirms first, and the confirmation is worth reading: “Stop this run? Items that have already been generated are kept and stay charged. Everything not started yet is skipped and cannot be resumed — you would need to start a new run for the rest.”

So stopping is safe for what is already done, but it is a full stop, not a pause.

Working with the results

When the run has finished, the item table is your workbench. Each row has a Review button that opens a side-by-side comparison of Current shop content (as of last sync) and Generated — for picture tasks, a grid of the generated images next to the Current cover. A Prompt button shows exactly what the AI was asked, which is the fastest way to understand a disappointing result.

Above the table, a toolbar: “Tick the items you are happy with, then send them to your shop or ask a colleague to approve them.” Ticking rows enables its buttons.

The Module column of each row states where the text is destined to go — Goes into the product description or Goes into the custom field …. An item that writes several fields at once, as Multi-field text does, lists them all: Goes into: Name, Description, Custom field “…”. If a job produced several tasks, tabs above the table let you filter: All modules plus one tab per task.

Reviewing, approving and sending are covered on their own pages: Review & approval and Sending content to your shop.

Generating again

Not happy with some of the output? You have two options, and both cost credits again.

• Regenerate selected — tick the rows you want redone and press it. It confirms: “Generate the N selected items again? This replaces the texts and pictures already generated for them and costs credits a second time.”
• Generate everything again — top right, for the whole job. Same warning, bigger number.

Two kinds of item are skipped, and you are told how many: items a worker is sending to your shop right now, and items waiting for an approval decision. An item that has finished being sent can be regenerated: it goes back to Not reviewed and Not pushed. That does not undo anything — the text already in your shop stays there until you send the new version, which costs the push credit a second time. See Sending content to your shop.

The jobs list

Two views on the same thing:

• All jobs in the sidebar — every run in this shop, newest first.
• Each task page (Descriptions, Meta texts, …) — only that task’s runs.

The table shows Created, Modules, Model, Progress (done / total) and Status, with a View button on every row.

What the job statuses mean

Where to go next

• The generation modules — every task and every option, one by one.
• Review & approval — checking results and getting them signed off.
• Sending content to your shop — the last step, and the only one that changes your live shop.
• Credits — what costs what.