XLSX vs Google Sheets: What the File Format Actually Changes

By the XLsheetAI Team · Updated September 27, 2026 · 11 min read

The short answer

An .xlsx and a Google Sheet are not two formats for the same thing — they are two different kinds of object. The .xlsx is a self-contained file you possess: a zipped bundle of XML that any tool can read, copy, email, checksum or archive. A Google Sheet is a live record on a server that you reach through a URL, and downloading it produces an .xlsx that did not exist a second earlier. Once you see them as a file versus a service, the practical consequences stop being surprising: files fork and can be signed, services stay singular and can be co-edited, and converting between them costs something every time because each side holds concepts the other has no home for.

This article is about the artefact. If you are choosing which product a team should use, Excel vs Google Sheets covers pricing, scale limits, automation and AI features, and is the better starting point.

What is actually inside an .xlsx

Try this now, because it takes fifteen seconds and permanently changes how you think about spreadsheets. Take any .xlsx, make a copy, rename the copy's extension to .zip, and open it.

It is a folder. Inside:

Open a worksheet XML and a cell looks roughly like this:

<c r="B7"><f>SUM(B2:B6)</f><v>55</v></c>

Read that carefully, because it explains a behaviour that confuses people. Cell B7 stores two things: the formula as text, and 55 as the last value it computed. The file caches results. That is why a tool that cannot evaluate formulas — a Python script, a preview pane, an import routine — can still show you numbers, and why those numbers are whatever they were when the file was last saved by something that did calculate. A stale cached value in a file nobody has recalculated is a real failure mode, not a theoretical one.

The sharedStrings design is worth noticing too. Text is stored once and referenced, so a column of 100,000 rows containing one of five category names stores five strings and 100,000 short references. It is why .xlsx files are often far smaller than people expect, and why a workbook of unique text is far larger than a workbook of repeated text at the same row count.

The other Excel extensions, briefly

That .xls row is not a historical footnote. In 2020, Public Health England lost 15,841 positive COVID-19 test results because results were collated into templates saved in .xls, and rows past the 65,536 ceiling were dropped without a warning — a case covered in the costliest spreadsheet errors ever made. Truncation is the most dangerous kind of failure, because the output still looks complete.

A Google Sheet has no file

This is the part that makes the comparison click, and most explanations skip it.

When you create a Google Sheet, no spreadsheet file is written anywhere you can reach. The content lives as structured data in Google's storage, and what you are given is a URL. There is no .gsheet document containing your data — the thing that appears in Google Drive is a pointer to a record held server-side.

So when you choose File → Download → Microsoft Excel (.xlsx), Google does not hand you a stored file. It builds one, right then, by translating its internal representation into the Office Open XML structure described above. Download the same Sheet twice and you get two separate files, each a snapshot of a different instant, neither of which is the document.

Three consequences follow directly, and they are the answers to most questions in this area:

• A Sheet cannot fork. There is one authoritative copy, so there is never a "which version is current" question. An .xlsx forks the moment it is copied, and email makes copying the default action.
• A Sheet has no fixed identity. You cannot hash it, sign it, or attach it to anything — only a snapshot of it. If someone edits the Sheet after you cite it, your citation now points at different content, silently. For an audit pack or a contract schedule, that is disqualifying.
• A Sheet needs the network and the service. The .xlsx on your disk will open in fifteen years in any tool that can read a ZIP, because the format is a published standard. That is a real durability difference, whatever you think of either company's longevity.

Three things people mean by "sharing a spreadsheet"

The question "how does real-time collaboration work in a cloud tool versus a shared Excel file?" has a confusing answer because "a shared Excel file" describes two completely different situations. There are three cases, not two.

The middle column is the one people underestimate, usually because they last tried it years ago. Excel co-authoring is real: per Microsoft's own documentation, it requires the workbook to be stored in OneDrive, OneDrive for Business or SharePoint Online, in .xlsx, .xlsm or .xlsb format, with AutoSave keeping changes flowing. Note two exclusions in that list: the Strict Open XML variant is not supported, and SharePoint sites hosted on-premises rather than by Microsoft do not support co-authoring at all.

The thing to take from that table: what decides whether two people can edit a spreadsheet at once is where it lives, not what format it is in. The same .xlsx is a collaboration disaster as an attachment and a genuine co-authoring document in OneDrive. Most "Excel can't collaborate" complaints are really "we email attachments" complaints, and that is a workflow choice rather than a format limitation.

Where the file model still gives ground is the shape of the history. OneDrive versioning keeps a series of saved states, so you can go back to 4:15pm. A Sheet's history is a continuous stream of individual attributed edits, so you can ask who changed this cell and when. For anything where accountability per value matters — the exact question behind pay-review controls — that is a materially different capability, not a nicer interface.

Why each round trip costs something

Converting is not copying. Excel and Google Sheets hold different sets of concepts, so a conversion is a translation, and anything with no equivalent on the far side gets approximated or dropped. Nothing restores what a previous translation removed, which is why losses compound rather than repeat.

The rule of thumb that predicts almost all of this: anything typed into a cell probably survives, anything configured in a dialog probably does not, and anything written in code definitely does not.

Two asymmetries are worth knowing. Excel to Sheets loses more, because Excel simply has more machinery that Sheets has no home for. And in the other direction the casualties are specifically the Google-only functions — every QUERY, IMPORTRANGE, GOOGLEFINANCE and ARRAYFORMULA arrives in Excel as an error with no automatic substitute. A Sheet built without those exports to .xlsx with mostly cosmetic differences.

The practical damage is rarely the first conversion. It is the file that bounces — exported to Excel for a board pack, re-uploaded for the team to update, exported again next quarter. Each pass sheds a little, nobody is watching, and the loss surfaces as a wrong number rather than as an error. Convert once, deliberately, then commit to a side.

Doing each conversion with the least loss

Excel to Google Sheets. Upload to Drive, then File → Save as Google Sheets rather than working on the uploaded file in Office compatibility mode, which is a half-state that pleases nobody. Before you do: note every macro and Power Query step somewhere outside the file, because they will be gone and you will not be told. After: check any date column, spot-check totals against the original, and look specifically at conditional formatting rules.

Google Sheets to Excel. File → Download → Microsoft Excel (.xlsx). Before you do: search the Sheet for IMPORTRANGE, QUERY, GOOGLEFINANCE and IMPORTHTML, and convert those results to static values if the Excel version needs to work. After: check for #NAME? errors, which is how Excel reports a function it has never heard of.

When you only need the numbers. Export CSV and accept the loss on purpose. A deliberate, complete loss of formatting and formulas is far easier to reason about than a partial, silent one — and if something downstream only wants values, the fidelity question disappears entirely.

Rebuilding a formula that did not survive the conversion?

Describe what the original did in plain English and XLsheetAI writes the Excel equivalent, explains each argument, and lets you practise it — including the Google-only functions like QUERY and ARRAYFORMULA that arrive as errors.

Different ceilings, counted differently

The limits are not comparable units, which causes a specific and avoidable surprise.

Excel gives every worksheet 1,048,576 rows and 16,384 columns, fixed since the .xlsx format arrived. Google caps a whole spreadsheet at 10 million cells or 18,278 columns.

Per sheet versus across the file is the trap. A 20-column table in Sheets runs out at roughly 500,000 rows. A 40-column export dies around 250,000. Teams who read "10 million" as "10 million rows" discover this on their first real dataset, usually mid-migration. And both slow down long before their ceilings — Sheets in a browser, where you cannot add hardware, and Excel on a machine you can upgrade.

Choosing which one is the artefact of record

This is the decision that actually matters, and it is not about which product is better. It is about what the thing has to be.

The .xlsx should be authoritative when something other than a person consumes it — an upload endpoint, an import routine, a payroll system. When the artefact itself must be fixed, hashable and archivable: regulatory filings, audit packs, contract schedules. When it depends on VBA, Power Query or Power Pivot. When it has to work with no network. Or when you need to hand someone a spreadsheet that will still open in a decade.

The Sheet should be authoritative when several people update it routinely, when you need per-cell attribution of who changed what, when the readers are outside your organisation and you do not want to manage attachments, or when most editing happens on phones.

The failure to avoid is treating both as authoritative at once. A Sheet the team edits and an .xlsx that finance updates guarantees divergence, and the divergence is discovered late and by accident. Pick one as the source of truth and make the other an explicitly disposable export — regenerated, never edited, ideally with a generated-on date written into cell A1 so nobody mistakes a stale snapshot for the live record.

For the product-level decision, see Excel vs Google Sheets. If a converted file has left you with broken formulas or misread dates, cleaning messy data covers fixing a column at a time, and dynamic arrays covers the modern Excel functions that replace many Google-only ones.