CSV export of iPhone messages: what every column means

A CSV export of an iPhone conversation is one row per message and fourteen comma separated columns, written as UTF-8 with a byte order mark and CRLF line ends so that Excel on Windows reads the accents and the emoji correctly. It is the format to choose when somebody is going to count, filter or sort the conversation rather than read it. Whether a double click also splits it into columns depends on your Windows, which is the next section but one. Getting to it is the same job as any other export, which how to export text messages from an iPhone to a computer covers from the backup onwards.

What are the fourteen columns?

In this order, which is the format’s contract and the thing a script depends on. One row from the sample published on this site, so you can check it against the file rather than against this page:

2026-02-06T19:04:00.000+01:00,+1 555 0142,false,"Picking Emily up Friday at 4, same as we agreed.",,,false,false,,,false,false,

That row has thirteen fields rather than fourteen, for the reason in the next section.

#	Column	What it holds
1	timestamp	ISO 8601 with the offset the phone recorded, so local time and the zone travel together
2	sender	The handle the message came from, or me for your own
3	is_from_me	true or false
4	service	iMessage, SMS or RCS as the source recorded it, empty when it recorded nothing
5	text	The words, with the attachment placeholder character stripped
6	attachments	The attachment filenames, several separated by ;
7	reactions	Each reaction as kind:sender, several separated by ;
8	edited	Whether this message was edited
9	deleted	Whether it is marked deleted
10	edited_at	When the edit happened, when the row records it
11	previous_versions	The superseded wordings, one per line inside the cell, each prefixed with its replacement time when known
12	retracted	Whether the sender unsent part of this message
13	recovered_deleted	Whether this row came from the Recently Deleted store, which holds a deleted message for about 30 days, rather than from the live thread
14	deleted_at	When it was deleted, when the row records it

Columns 9 to 14 are the ones a screenshot cannot give you. Column 8 is the exception and worth naming as one: the phone does show Edited under a message it has edited, so that fact is on screen, while the wording it replaced is not. Two of these columns need a switch turned on before the export runs rather than after: previous_versions fills only when edit history was requested, and recovered_deleted only when the Recently Deleted store was included, which holds a deleted message for the 30 to 40 days Apple documents and nothing older.

Why does the sample file have thirteen columns?

Because the sample on this site was generated before service existed, and the card beside it counts the file you can download rather than the program that wrote it. The published sample has thirteen columns and no service; the program ships fourteen.

One consequence matters if you are writing a script against the sample rather than against your own export: with service absent, every column after the third shifts left by one, so text is the fourth field in the sample and the fifth in the table above. The contract is the header row, which is why reading the header rather than counting positions is the habit that survives this kind of change.

Correcting the sample’s count to fourteen would make this site describe a file nobody can download, so the two numbers stay until the sample is regenerated. You can check both halves yourself: the sample exports page carries the file.

Why is the header in English when my document is in German?

Because the header row is machine identifiers rather than reading matter. is_from_me and recovered_deleted are names a formula, a filter or a script refers to, and translating them would break the only thing this format exists for. The values are raw data for the same reason: ISO timestamps, raw handles, booleans.

Exactly one thing in a CSV follows the document language, and only on an unlicensed run: the trial notice. The document language picker changes the PDF, the TXT and the HTML, and leaves the CSV’s structure alone on purpose. Which fourteen document languages exist, and why the choice is separate from the program’s own interface language, is on the download page.

Why does everything land in one column?

Because Excel splits a double clicked CSV on the list separator from your Windows regional settings, not on the comma in the file. Where the decimal separator is a comma, which covers German, French, Italian, Spanish, Portuguese and Polish Windows, the list separator is a semicolon, so a comma separated file arrives with every row sitting in column A. Microsoft’s own support answers say so plainly: opening a .csv this way relies on the Windows country or region settings of that PC.

Nothing is wrong with the file, and there are two ways round it. Either open it through Data, then From Text/CSV, which lets you set the delimiter before anything is loaded, or change the list separator in Windows regional settings. The first is the one to use when the file is going to somebody else, because it changes nothing on your machine.

The byte order mark and the CRLF line ends do a different job: they are why the accents, the Polish characters and the emoji come out as themselves rather than as mojibake.

What else is worth knowing before you open it?

Two habits and one surprise.

Open a copy, not the file you will file. Excel offers to save in its own format and to reformat what it thinks are dates. Neither is reversible in a way you want to discover later, and the exported CSV is the thing whose checksum sits in the .sha256 file beside it.

Sorting is fine, but sort back. The rows come out in conversation order, and sorting by sender to group one person’s messages is the common and useful operation. A file that no longer reads as a conversation is not the file to attach to anything, so do that work on the copy.

Expect a leading apostrophe on some cells.

Excel and Google Sheets treat a cell beginning with =, +, - or @ as a formula, and OWASP’s note on CSV injection records that quoting the field correctly does not stop it. Four columns carry words the other side of the conversation wrote, so those four are the ones guarded: sender, text, attachments and previous_versions. Each arrives with Excel’s own treat-as-text apostrophe in front of it when it begins with one of those characters, which is why a phone number may look like '+1 555 0142 in a text editor.

What is never touched is what the program writes itself: the timestamps, the booleans, the transport name and the reactions column, which always begins with the reaction’s own kind.

Two notes on the apostrophe, because it is a mitigation rather than a fix. OWASP says the technique is unreliable once a CSV has been saved and reopened from Excel, which is a second reason to work on a copy. And a spreadsheet that imports rather than opens the file may show the apostrophe in the cell instead of hiding it.

What does a trial CSV look like?

Like a licensed one with an extra column. The unlicensed run adds a fifteenth column named trial_notice and repeats the notice on every row, rather than writing a comment line above the header.

A preamble line would break every script and spreadsheet that reads the file, turning “your trial output is watermarked” into “your CSV is corrupt”. Repeating the notice per row means it survives somebody sorting, filtering or copying a block of rows out, which a notice on the first row would not. The trial itself stops at the 50 most recent messages of a conversation.

What a CSV cannot do

• It cannot show the photographs. The attachments column names the files; it does not carry them. A document that needs the images in it is a PDF or an HTML export, which is exporting messages with photos and attachments.
• It cannot be handed to somebody as a document. It is a data file. A reader who is not going to filter anything is better served by the PDF, and most people filing a conversation want both: the PDF to read, the CSV to check.
• It cannot carry the integrity page. The source checksums, the device details and the scope sentence are part of the PDF’s front matter, which is the page a recipient reads to see where the document came from. The CSV gets the same .sha256 sidecar beside the file, which lets anyone check that this file is the one the program wrote, and says nothing about its origin.
• It cannot tell you what it never held. An empty edited_at means the row records no edit time, which is not the same as “was not edited”.

Questions people ask

Can I load a message CSV into a spreadsheet and still show it is unchanged?

Not after you save it from the spreadsheet, and that is the point of exporting once and working on a copy. The checksum belongs to the exported file byte for byte. Checking one is how to verify a PDF with SHA256, and the procedure is the same for a CSV.

Why are my own messages shown as me instead of my name?

Because me is unambiguous and a name is not: a filter matching on me cannot be fooled by somebody in your contacts who shares your first name. The column beside it, is_from_me, carries the same fact as a boolean, so a script never has to match on a word at all.

Does the CSV include group chats?

Yes, with the sender column holding the handle per row, which is what makes a group filterable by person. The participant list and the per sender counts are on the PDF’s front matter rather than in the CSV, because a data file has nowhere sensible to put them.

Is there a JSON export?

Not today. The five formats are PDF, TXT, HTML, CSV and RSMF, and RSMF is the one built for a review platform rather than for a reader. Anything else named on a roadmap is not something this page will promise.

Where to go next

• How to export text messages from an iPhone to a computer, the route the file comes out of.
• Exporting iPhone messages with their metadata, for what the other formats carry beside the words.
• The sample exports, to open a real CSV before installing anything.