Skip to content

Writing many descriptions at once from a spreadsheet

Up to date as of 7 November 2026

Doing a catalogue in one go is mostly a spreadsheet problem, not an AI problem. If the sheet is clean, the drafts are easier to produce and much easier to check.

One product per row

A CSV file, the plain format most spreadsheets can export, is described by Wikipedia as plain text data consisting of one record per line.1 That's what you want: one product per row, with the facts from your fact sheet as columns.

Columns that work for most shops: product name, type, size, weight, materials, colours, what's in the box, care, origin, condition, and a notes column. Add a column called "unknown" or "to check" for gaps.

Wikipedia also lists some rules worth knowing. Each record should contain the same number of fields. A field with a comma, a double quote or a line break in it should be enclosed in double quotes.1 A materials cell reading "cotton, linen" will split into two columns if it isn't quoted, and the facts then shift sideways under the wrong headings. The page adds that care is required when importing a file whose first line may be a header.1

Run a small batch first

A batch run with checksBuild one clean row per productRun a small test batchRead every output against its rowFix the prompt, then run the rest
Small batch first, then the rest, with a person reading every row.

Pick a handful of rows that differ from each other: a simple item, an awkward one, one with missing facts. Run only those. Read the results against the rows before you touch the other products.

Problems show up early this way. A rule the tool ignores, a tone that drifts, a field it misreads. Fix the prompt, run the small batch again, and only then do the rest. The wording rules are in the prompt page.

Keep the prompt identical

Use the same prompt for every row, with only the row's facts changing. If you edit the prompt halfway, note where, so you know which products were written under which version. Why the example pair matters is covered in keeping your shop's voice.

Don't skip the reading

Speed is the temptation, and it's where control gets lost. Google tells site owners to fact-check and review everything an AI tool generates before it's published.2 Everything means every row, and a catalogue is a lot of rows.

The ACCC requires a business's claims about what it sells to be accurate, truthful and based on reasonable grounds.3 That applies to the last description in the sheet the same as the first. The rule is Australian, so check your own country's version.

So give each output a status column: draft, checked, approved, published. Nothing moves to published until a person has matched it to its row, using the checking steps. Spread the reading over several days if you must. Tired readers skim.

Watch for sameness

Mass-produced pages that add nothing for readers are what Google's spam policy on scaled content is aimed at.2 A big batch can produce similar openings. Scan a column of first sentences. If many start the same way, change the prompt and rerun those rows.

Keep the paper trail

Save the sheet with the facts, the prompt version, the raw output and the approved text. If a customer questions a claim, you can show where it came from. If a supplier changes a spec, you know which products to revisit.

Match output to row

However you feed the rows in, whether pasted into a chat or run some other way, copying by hand gives a row the chance to land against the wrong product. Put the product name at the top of each output and check it before you approve anything.

Where the facts come from

  1. Comma-separated values, Wikipedia
  2. Google Search's guidance on using generative AI content on your website, Google Search Central
  3. False or misleading claims, Australian Competition and Consumer Commission (ACCC)