Answers
How do I get a tweet's posting time from its ID?
Short answer
Shift the post ID right by 22 bits and add 1288834974657, X's Snowflake epoch in Unix milliseconds; the result is the creation time to the millisecond, computed offline for free. It holds for every post since November 2010. When you also want the text and counts, okdata X post detail returns published_at with them for $0.0003.
Every post on X has a numeric ID like 1987654321098765432, and that number carries its own timestamp. X generates IDs with Snowflake, a scheme it built in 2010: the high bits are the milliseconds since a fixed starting point, so you can read the creation time straight out of the ID.
The layout
| Bits | Holds |
|---|---|
| 1 (top) | Always 0 |
| 41 | Milliseconds since the X epoch, 1288834974657 (4 November 2010, 01:42:54.657 UTC) |
| 10 | Machine number |
| 12 | Sequence within that millisecond |
So: unix_ms = (id >> 22) + 1288834974657.
Worked example
Take 1987654321098765432.
- Shift right by 22 bits (integer division by 4,194,304):
473893718981. - Add the epoch:
473893718981 + 1288834974657 = 1762728693638milliseconds. - That is 9 November 2025, 22:51:33.638 UTC, or 06:51:33 on 10 November in Beijing.
We checked the formula against a real post from an API response: the decoded time and the post's createdAt agreed to the second.
Things that trip people up
- Precision. IDs exceed what a JavaScript number holds exactly, so
Number("1987654321098765432")silently changes the last digits. Use strings for storage andBigIntfor maths, as X's own documentation says. - Old posts. IDs from before Snowflake went live in late 2010 are small sequential numbers, and the formula gives a meaningless time for them. The largest pre-Snowflake ID the TweetedAt researchers could find is 29700859247; treat IDs up to that as undated.
- User IDs. Accounts created in recent years also have Snowflake IDs; one we decoded landed seven seconds before the account's listed creation time. Older accounts have short IDs that do not decode.
- Sorting. Because time is in the high bits, sorting posts by ID sorts them by time. You do not need the date column for ordering.
When to call the API
X post detail takes an ID or an x.com link and returns published_at along with the text, media, views, likes, reposts and replies.
| You need | Do | Cost |
|---|---|---|
| Posting time only | Decode the ID | free |
| Time plus current engagement | One detail call | $0.0003 |
| 1,000 posts from a dataset | 1,000 calls | $0.30 |
Example
import os, requests
from datetime import datetime, timezone
X_EPOCH_MS = 1288834974657
def posted_at(post_id: str) -> datetime:
ms = (int(post_id) >> 22) + X_EPOCH_MS
return datetime.fromtimestamp(ms / 1000, tz=timezone.utc)
pid = "1987654321098765432"
print(posted_at(pid).isoformat(timespec="milliseconds")) # 2025-11-09T22:51:33.638+00:00
body = requests.get("https://api.okdata.online/v1/x/posts/get", timeout=60,
headers={"Authorization": f"Bearer {os.environ['OKDATA_API_KEY']}"},
params={"id": pid, "fields": "id,published_at,metrics.views,metrics.likes"}).json()
print(body["data"]["item"] if body["code"] == "ok" else body["code"])
In JavaScript: new Date(Number((BigInt(id) >> 22n) + 1288834974657n)).
Limits
- A deleted or protected post still decodes, but detail returns
not_foundfor it, which is still charged as one call. - Each edited version of a post has its own ID (X's data dictionary lists them as
edit_history_tweet_ids), so the ID of an edit dates the edit, not the original.
To read the post itself from a link, see how an AI agent can read a tweet from an x.com link.
Sources
- X API docs: X IDs (64-bit Snowflake IDs)
- X API data dictionary: Post and user fields
- Snowflake ID (Wikipedia)
- twitter-archive/snowflake
- TweetedAt: finding tweet timestamps for pre and post Snowflake tweet IDs
Where people asked this (read 2026-11-12)
Updated