The Spreadsheet Was Correct. Nobody Could Read It.

Here's a failure mode I didn't appreciate until it happened to us: you ask Claude to build a spreadsheet, it writes every value correctly, and the result is unreadable.

Not wrong. Unreadable. Writing values into Google Sheets gives you no formatting at all. The untouched default is columns 100 pixels wide, and in a filled table a long cell clips to a slit. A directory sheet with names, descriptions, and links renders as a wall of truncated fragments, and here's the part that makes it worse than a normal bug: the reader never learns there was more text. The person who made the sheet saw correct data go in. The person who opened it saw slivers. Neither knows what the other saw.

That's why formatting tools aren't polish. They're the difference between a sheet that transfers information and one that technically contains it.

As of August 26, four of them are live.

One call to readable

sheets_format_table does the whole readable pass at once: real column widths, wrapped and top-aligned cells, a styled and frozen header row, hairline borders. It's the tool to run on any sheet a person is going to open, and in practice you don't name it. You say "make this readable" and Claude reaches for it.

Before, the same result took a pile of raw batchUpdate requests with 0-based, end-exclusive grid ranges, which is exactly the kind of arithmetic that produces a border on the wrong row. I know because our first formatted sheets were built that way, by hand, and the off-by-one rate was not zero.

Specific styling, one atomic pass

sheets_format_range handles the styling that's yours rather than generic: fonts, colours, alignment, padding, number formats, merges. It takes a list of instructions and applies them in one atomic call.

Atomic matters more than it sounds. If one instruction is invalid, nothing applies and the error names which one. The alternative is a sheet that's half-styled in a way you now have to diff by eye.

sheets_batch_update stays underneath both as the escape hatch: banding, copy and paste, inserting or deleting columns, duplicating sheets, anything the job-shaped tools don't cover. Full pass-through, same atomicity.

The one that's easy to undersell

sheets_find_rows looks minor and isn't. It finds rows by value and returns their row numbers.

Without it, "update the row for this customer" means reading the whole sheet into the conversation and filtering it yourself. That burns context on hundreds of rows nobody asked about, and it gets worse every week the sheet grows. With it, a lookup is a lookup: you get back the row number and the exact range to write to.

It also takes many values at once, so looking up twenty customers is one call, not twenty. And values that matched nothing come back listed separately, which sounds like a detail until you've debugged the other version: an empty result that could mean "not there" or could mean "your search was broken", with nothing to tell you which.

The before and after

We ran this on a real sheet: a public directory we maintain, values written first, formatting second.

Before: default grid, every column 100 pixels, description text clipped mid-word, the header row scrolling away as soon as you moved. To a reader it looked like a data export someone forgot to finish.

After one sheets_format_table call and one sheets_format_range pass for the header colours: sized columns, wrapped text, a frozen tinted header. It looks like a person laid it out. Nobody touched the sheet by hand at any point, which is the proof point I care about, because "Claude did most of it and then I fixed the layout myself" was the old workflow and it defeated the purpose.

Zapier and n8n, honestly

Zapier's Google Sheets integration has real formatting actions, including Format Cell Range and conditional formatting rules, plus row lookups. n8n's Google Sheets node covers row operations: append, update, get, delete. Both checked against their docs on August 26, 2026. So no, "other tools can't format spreadsheets" would be false, and I'm not going to write it.

The difference is the same one we wrote about for formatted email: those are workflow steps. You decide in advance which range gets formatted and wire it into a zap or a node graph, and it runs the same way every time. That's automation, and for a recurring report it's a fine shape.

Connected is different. The sheet Claude just built for you is different every time, so the formatting has to be decided fresh, by something that can see what it made. "Widen the description column and freeze the header" is a sentence about this sheet, now. There was nothing to build and there's nothing to maintain.

Try it

Connect Google Sheets at datatorag.com and ask for a table you'd actually send to someone. Then ask Claude to make it readable, and open the result. The gap between those two sheets is the entire argument of this post.