How Autofill Fills the Gaps in a New Range Without Inventing Facts
What AI Autofill fills in a new furniture range, how the preview, sample run, credit budget and undo work, and which facts you should keep it away from.
The supplier file is in. The range is grouped into products and variants. Now the gaps show: no long description on any of them, no SEO title, a blank depth on two rows, no care copy, and missing attributes required by your current Next setup. AI Autofill is the part of Furniture Connect that fills what can be filled. The useful question is which gaps those are, and how you keep it honest.
Two kinds of column
Autofill works column by column, and there are two kinds [1].
AI agent columns write copy: descriptions, key features, care instructions, packaging text, SEO fields and custom attributes. They can also look up information and create product images [1].
Formula columns calculate a value from your other columns, such as a unit conversion or a retail price derived from a markup. Formulas are free, run instantly and keep themselves up to date; AI agent runs use credits [1].
Keep the two separate in your head. A formula converting millimeters to centimeters is arithmetic on a fact you already have. An AI column writing a description is generating new content from the facts you have. The controls below are mostly about the second kind.
What you set before it runs
For an AI agent column you write the instruction, choose which of your existing fields it may read, and set the scope. The assistant setup shows a preview before anything runs. Under Advanced, Fill mode lets you choose Replace existing values; otherwise it only fills empty cells and existing content is kept [1]. On a new range that is the behavior you want: codes, names, dimensions and prices untouched, empty fields filled.
Before a full run you see an estimate of how many products it covers, the credits it will use and roughly how long it will take, and you can set a credit budget for the run [1].
How a run goes
Click Start 10 sample cells. Read the ten results against the products and the photos. If they're right, click Continue; if not, adjust the instruction and sample again [1]. During the run a progress bar above the table lets you pause and resume at any time [1].
Afterwards, Undo run restores the previous content while keeping any edits you've made since, and Revert this cell reverts a single filled cell [1]. That is the safety net for a column that read well in the sample and drifted later.
What to keep it away from
This is the part that matters for a furniture retailer, because a plausible wrong fact on a listing is worse than a blank one.
Don't point an AI column at a fact the supplier hasn't given you. Undocumented materials, measurements, country of origin, certifications, warranty terms, safety or performance claims, delivery promises: if it is not confirmed by the supplier's data or another approved source, the instruction should tell the column to leave the cell empty rather than fill it, and the sample run is where you check that it does. A photo can suggest a wood-like finish or textured upholstery. It cannot confirm solid wood rather than veneer or imitation, the wood species, or fabric composition; those come from a verified specification.
Keep prices, SKUs, barcodes and native dimensions out of AI columns entirely. A derived retail price belongs in a formula column, where it's arithmetic on a trade price you were given, not something generated.
Treat these as designed behaviors you set and check, not a guarantee of error-free output. The sample run and the undo are there because generated content needs review, and the controls reduce the risk rather than remove it.
What this looks like on a real range
A range of 40 products with 180 variants arrives with names, codes, dimensions and prices. You add an AI column for the long description, scoped to the 40 parents, reading from the name, the supplier's one-line description, the material attribute and the images, with an instruction that says to describe only what the supplier data confirms. You run 10 sample cells. The descriptions mention oak legs and bouclé upholstery where the supplier's specification confirms them, and say nothing about weight limits because nothing in the data states one. You set a credit budget and continue.
Later you have 40 long descriptions, 40 sets of SEO fields and a Next-specific title column filled from the same facts. You still have two blank depths and no origin on any product, because the supplier didn't give you them, and the instruction told the column to leave those alone. That is correct behavior. Those two emails to the supplier are the real remaining work.
Why this is the right split
One time-consuming part of preparing a range is writing distinct descriptions and adapting them to a retailer's house style. Autofill can reduce that drafting work; supplier clarification, image review and channel requirements can still be the limiting steps.
Missing supplier facts and the judgment about what goes live still need people who can verify and approve them. Keep both with the people who can answer them, using explicit source boundaries and reviewing what Autofill produces.
What to do this week
Pick the range with the most blanks. Add one AI column, run the ten-cell sample, read the output against the approved source data, and count how many remaining gaps are facts only the supplier can supply. That becomes the brief for your next supplier call. The step before this one is covered in what happens when you import a supplier spreadsheet. For the whole journey from supplier file to buyable, see getting a new furniture range online.
