Export your X bookmarks to a spreadsheet

Nine columns, ordered newest-saved-first, named bookmarks-2026-09-07.csv after the UTC day you asked for it. Every cell is quoted unconditionally, because post text contains commas, straight quotes and hard line breaks as a matter of routine and a writer that only quotes when it thinks it needs to eventually meets a post that proves it wrong. Embedded quotes are doubled, per the standard. Line endings are CRLF, again per the standard, and because Excel is the primary consumer.

Double-clicking the file works and produces something that looks correct, which is the problem. Two things go wrong on that path, both silently.

Encoding. The file is UTF-8 with no byte-order mark. Excel on Windows, handed a CSV with no BOM, falls back to the system code page — so emoji, accented characters and any CJK in a post turn into mojibake, and you find out weeks later when you search for a post you know you saved.

Numeric precision. Excel carries 15 significant digits. A post ID has 19. Opened as a number, 1795432198765432198 becomes 1.79543E+18 and the last four digits are gone — not hidden, gone. Formatting the column as text afterwards recovers a rounded number, not the ID. Every join, every deduplication and every attempt to reconstruct a permalink from that column is then subtly wrong.

1. 1

   Data → Get Data → From Text/CSV

   Pick the file. Do not open it from Explorer or from the recent-files list.
2. 2

   Set File Origin to 65001: Unicode (UTF-8)

   Check the preview pane for a post you know contains an emoji or an accent before going further. This is the whole encoding fix.
3. 3

   Transform Data, not Load

   In Power Query, click the post_id header, then Transform → Data Type → Text. If a “Change column type” step already appeared, replace it rather than adding another. Do the same for anything else you want kept literally.
4. 4

   Close & Load

   You get a real table with a refreshable query attached, which is convenient later: point it at next week’s export file and refresh instead of repeating all of this.

A post can begin with =, +, - or @. Quoting the field in the file, which we do, does not stop a spreadsheet from treating the cell as a formula once it has been parsed — the standard defines the file, not what Excel does with it afterwards. Importing as text, as above, is also what defuses this; it is a good reason not to turn formula conversion back on for convenience.

A pivot table on ai_category answers “what do I actually save” in about ten seconds, and it is usually not what people expect. Counting by author_handle answers a blunter question: which five accounts are producing most of your reading list.

The tags column is space-separated, which is deliberate — commas inside a CSV cell are survivable but tiresome. To fan it out, TEXTSPLIT in current Excel or SPLIT in Sheets will do it in one formula. If you would rather have the whole file arrive in a different shape altogether, the converter re-emits it as JSON or Markdown without uploading anything.

"post_id","post_url","author_handle","posted_at","saved_at","post_text","ai_category","ai_tags","ai_summary"
"1795432198765432198","https://x.com/rauchg/status/1795432198765432198","@rauchg","2026-08-14T09:12:44.000Z","2026-08-14T21:03:01.812Z","the only two hard problems in caching","Engineering","caching infra performance","A one-line note on cache invalidation."

Each download starts from your most recently saved bookmark and works backwards, so two exports taken on different days overlap wherever your library did not change. Paste the second file below the first, select the post_id column and use Remove Duplicates — that column is the only field in the file guaranteed unique per row, which is the other reason not to let it become a float.

If enforced field types and an actual merge key sound better than a column of strings you have to remember to protect, that is more or less the argument for loading the same file into Airtable, where the import can upsert on the post ID. And if you would rather read your saved posts than tabulate them, one note per bookmark in a vault is the other end of the same spectrum.