Bulk run

Bulk run fills a column from the rows of your sheet without writing a formula in every cell. You pick the input columns, write one instruction with {{Header}} placeholders, preview it on a few rows with the exact cost, then let it run in the background while you keep working. Results land as plain values, so nothing recalculates or bills again when the sheet is opened later.

In Google Sheets, click Extensions ➤ GPT Workspace ➤ ⚡ Bulk run…. The sidebar reads the header row of the active sheet and lists its columns.

Field	What it does
Mode	Prompt answers your instruction for every row. Translate translates the input cells into a language. Classify picks one of your categories per row. Extract pulls one or more fields out of the input text. Web enrich searches the web to fill company details.
Input columns	The columns the instruction can reference. Each becomes a {{Header}} chip you can click to insert.
Output column(s)	Where results are written. Extract and Web enrich take one output column per field.
Instruction	For Prompt and Extract, the template, for example Write a one-line tagline for {{Company}} based in {{Country}}. For Translate, the target language. For Classify, optional context for the categories. For Web enrich, optional hints such as They are all in the fintech space.
Rows	The first and last row to process. Defaults to every row below the header.
Skip rows that already have a value	On by default: a row whose output cells are all filled is left alone, so re-running a job only fills the gaps.
Stop after … credits	A ceiling for the whole job. The run pauses when it is reached and you can resume it.

Preview runs the first five rows for real, shows their results next to their inputs, and extrapolates their cost to the whole range. The sample rows are billed like the matching formula (=GPT, =GPT_TRANSLATE, =GPT_CLASSIFY, =GPT_WEB) and are not written to the sheet: when you start the run, their results come from the cache for free. Adjust the instruction and preview again until the samples look right.

The estimate is an extrapolation: rows with longer text cost more, and rows whose inputs were already seen (by a formula or an earlier run) are served from the cache for free.

Run starts the job on our servers in batches of 20 rows. You can close the sidebar: the job keeps going and the progress bar catches up when you reopen it. Each batch is written to the sheet and checkpointed, so:

• Pause stops after the batch in flight. Resume continues from the next unprocessed row.
• Cancel stops the job for good; the rows already written stay.
• If the account runs out of credits or the job reaches its credit ceiling, it pauses by itself and the sidebar says why.
• If Google access expires while the job runs, it pauses and resumes the next time you open the sidebar.

Rows with identical inputs are run once and billed once. Failed rows are counted and skipped; a later run with Skip rows that already have a value on only re-runs those.

Web enrich uses the same web search as =GPT_WEB to fill, per row, any of: Website, Industry, HQ country, LinkedIn URL and Description. Give it a company name column, optionally a country or domain column to disambiguate, and one output column per field. Like =GPT_WEB, it is available on the Standard and Pro plans.

Plan	Bulk run
Free	Preview only, up to 50 rows
Starter	Up to 500 rows per job
Standard and Pro	Unlimited rows, Web enrich

Credits are consumed at the model’s actual cost, like formulas: see What is a credit?. Bulk runs count toward the Sheets credits of your plan.

The sidebar lists the last runs of the current spreadsheet with their status, rows processed and credits used, for 30 days. A paused job can be resumed from there.