🔍 Advanced Search on Twitter

These operators work on Web, Mobile, Tweetdeck.

There is some overlap, but largely these will not work for v1.1 Search, Premium Search, or v2 Search APIs.

Adapted from TweetDeck Help, @lucahammer Guide, @eevee Twitter Manual, @pushshift and Twitter / Tweetdeck itself. Contributions / tests, examples welcome!

Matching:

On web and mobile, keyword operators can match on: The user's name, the @ screen name, tweet text, and shortened, as well as expanded url text (eg, url:trib.al finds accounts that use that shortener, even though the full url is displayed).

By default "Top" results are shown, where "Top" means tweets with some engagements (replies, RTs, likes). "Latest" has most recent tweets. People search will match on descriptions, but not all operators work. "Photos" and "Videos" are presumably equivalent to filter:images and filter:videos.

Exact Tokenization is not known, but it's most likely a custom one to preserve entities. URLs are also tokenized. Spelling correction appears sometimes, and also plurals are also matched, eg: bears will also match tweets with bear. - not preceeding an operator are removed, so "state-of-the-art" is the same as "state of the art".

Private accounts are not included in the search index, and their tweets do no appear in results. Locked and suspended accounts are also hidden from results. There are other situations where tweets may not appear: anti-spam measures, or tweets simply have not been indexed due to server issues.

Twitter is using some words as signal words. E.g. when you search for “photo”, Twitter assumes you’re looking for Tweets with attached photos. If you want to search for Tweets which literally contain the word “photo”, you have to wrap it in double quotes "photo".

Building Queries:

Most "filter:type" can also be negated using the "-" symbol, with exceptions like filter:follows which can't be negated. exclude:links is the same as -filter:links. It's sometimes worth trying an alias like that in case the search doesn't work first time.

Example: I want Tweets from @Nasa with all types of media except images

from:NASA filter:media -filter:images

Combine complex queries together with booleans and parentheses to refine your results.

Example 1: I want mentions of either "puppy" or "kitten", with mentions of either "sweet" or "cute", excluding Retweets, with at least 10 likes.

(puppy OR kitten) AND (sweet OR cute) -filter:nativeretweets min_faves:10

Example 2: I want mentions of "space" and either "big" or "large" by members of the NASA astronauts List, sent from an iPhone or twitter.com, with images, excluding mentions of #asteroid, since 2011.

space (big OR large) list:nasa/astronauts (source:twitter_for_iphone OR source:twitter_web_client) filter:images since:2011-01-01 -#asteroid

To find any quote tweets, search for the tweet permalink, or the tweet ID with url eg: https://twitter.com/NASA/status/1138631847783608321 or url:1138631847783608321, see note for more.

For some queries you may want to use parameters with hyphens or spaces in it, e.g. site:t-mobile.com or app:Twitter for iOS. Twitter doesn’t accept hyphens or spaces in parameters and won’t display any tweets for this query. You can still search for those parameters by replacing all hyphens and spaces with underscores, e.g. site:t_mobile.com or app:Twitter_for_iOS.

Limitations:

Known limitations: card_name: only works for the last 7-8 days.

The maximum number of operators seems to be about 22 or 23.

All the Time operators have to be used in conjunction with something else to work.

Tweetdeck Equivalents:

Tweetdeck options for columns have equivalents you can use on web search:

• Tweets with Images: filter:images
• Videos: filter:videos
• Tweets with GIFs: card_name:animated_gif
• "Tweets with broadcasts": (card_domain:pscp.tv OR card_domain:periscope.tv OR "twitter.com/i/broadcasts/")
• "Any Media" (filter:images OR filter:videos)
• "Any Links (includes media)": filter:links

Notes:

Web, Mobile, Tweetdeck Search runs on one type of system (as far as i can tell), Standard API Search is a different index, Premium Search and Enterprise Search is another separate thing based on Gnip products. API docs already exist for the API and Premium but i might add guides for those separately.

Snowflake IDs:

All user, tweet, DM, and some other object IDs are snowflake IDs on twitter since 2010-06-01 and 2013-01-22 for user IDs. In short, each ID embeds a timestamp in it.

To use these with since_id / max_id as time delimiters, either pick a tweet ID that roughly has a created_at time you need, remembering that all times on twitter are UTC, or use the following (This works for all tweets after Snowflake was implemented):

To convert a Twitter ID to millisecond epoch:

(tweet_id >> 22) + 1288834974657 -- This gives the millisecond epoch of when the tweet or user was created.

Convert from epoch back to a tweet id:

(millisecond_epoch - 1288834974657) << 22 = tweet id

Here's a use case:

You want to start gathering all tweets for specific search terms starting at a specific time. Let's say this time in August 4, 2019 09:00:00 UTC. You can use the max_id parameter by first converting the millisecond epoch time to a tweet id. You can use https://www.epochconverter.com.

August 4, 2019 09:00:00 UTC = 1564909200000 (epoch milliseconds)

(1564909200000 - 1288834974657) << 22 = 1157939227653046272 (tweet id)

So if you set max_id to 1157939227653046272, you will start collecting tweets earlier than that datetime. This can be extremely helpful when you need to get a very specific portion of the timeline.

Here's a quick Python function:

def convert_milliepoch_to_tweet_id(milliepoch):
    if milliepoch <= 1288834974657:
        raise ValueError("Date is too early (before snowflake implementation)")
    return (milliepoch - 1288834974657) << 22

Unfortunately, remember that JavaScript does not support 64bit integers, so these calculations and other operations on IDs often fail in unexpected ways.

More details on snowflake can be found in @pushshift document here.

Quote-Tweets

From a technical perspective Quote-Tweets are Tweets with a URL of another Tweet. It's possible to find Tweets that quote a specific Tweet by searching for the URL of that Tweet. Any parameters need to be removed or only Tweets that contain the parameter as well are found. Twitter appends a Client-parameter when copying Tweet URLs through the sharing menu. Eg. ?s=20 for the Web App and ?s=09 for the Android app. Example: twitter.com/jack/status/20/ -from:jack

To find all Tweets that quote a specific user, you search for the first part of the Tweet-URL and exclude Tweets from the user: twitter.com/jack/status/ -from:jack.

Geo Searches

Very few tweets have exact geo coordinates. Exact Geo coordinates are phased out for normal tweets, but will remain for photos: https://twitter.com/TwitterSupport/status/1141039841993355264

Tweets instead can be tagged by Place

How did I find these in the first place?

Reading Twitter Documentation and help docs from as many sources as possible - eg: Developer Documentation, Help pages, Tool-specific help pages, eg: Tweetdeck help etc. Using Share feature on tweetdeck to copy the search string. Searching google and pastebin and github for rarely documented ones together to find other lists of operators others have compiled.

Known Unknowns and Assumptions:

I have no idea how Twitter decides what should match filter:news, my guess is that it's based on a list of whitelisted domain names, as tweets from anyone can appear as long as they link to a news site, no idea if this list is public. No idea if or how this filter changed over time. But we can try to retrieve tweets and see. lang:und will match most empty tweets or tweets with a single number or link. filter:safe presumably uses the User setting "Contains Sensitive Content" - but may also apply to specific tweets somehow.

It would be great to be able to reliably find Promoted tweets - this may be possible with some of the card searches. Tweets composed in Twitter Ads are available with source:twitter_ads but other promoted tweets may not have been created with that app.

I'd also like to search for Collections (Timelines) and Moments, but this seems to work ok with just url: searches. eg: url:twitter.com/i/events and url:twitter.com/i/moments (I think the difference is events are curated?) but url:twitter.com url:timelines has many false positives.

In Search Settings, "Hide Sensitive Content" equivalent is filter:safe - is there an equivalent to "Remove Blocked and Muted Accounts"?

Supported Languages:

Language is specified as 2 letter ISO codes. Language is tagged automatically from the tweet text, nad not always accurate, see here for notes on accuracy. The list from TweetDeck dropdown menu has all of them:

lang:am Amharic (አማርኛ)
lang:ar Arabic (العربية)
lang:bg Bulgarian (Български)
lang:bn Bengali (বাংলা)
lang:bo Tibetan (བོད་སྐད)
lang:ca Catalan (Català)
lang:ch` Cherokee (ᏣᎳᎩ)
lang:cs Czech (čeština)
lang:da Danish (Dansk)
lang:de German (Deutsch)
lang:dv Maldivian (ދިވެހި)
lang:el Greek (Ελληνικά)
lang:en English (English)
lang:es Spanish (Español)
lang:et Estonian (eesti)
lang:fa Persian (فارسی)
lang:fi Finnish (Suomi)
lang:fr French (Français)
lang:gu Gujarati (ગુજરાતી)
lang:hi Hindi (हिंदी)
lang:ht Haitian Creole (Kreyòl ayisyen)
lang:hu Hungarian (Magyar)
lang:hy Armenian (Հայերեն)
lang:in Indonesian (Bahasa Indonesia)
lang:is Icelandic (Íslenska)
lang:it Italian (Italiano)
lang:iu Inuktitut (ᐃᓄᒃᑎᑐᑦ)
lang:iw Hebrew (עברית)
lang:ja Japanese (日本語)
lang:ka Georgian (ქართული)
lang:km Khmer (ខ្មែរ)
lang:kn Kannada (ಕನ್ನಡ)
lang:ko Korean (한국어)
lang:lo Lao (ລາວ)
lang:lt Lithuanian (Lietuvių)
lang:lv Latvian (latviešu valoda)
lang:ml Malayalam (മലയാളം)
lang:my Myanmar (မြန်မာဘာသာ)
lang:ne Nepali (नेपाली)
lang:nl Dutch (Nederlands)
lang:no Norwegian (Norsk)
lang:or Oriya (ଓଡ଼ିଆ)
lang:pa Panjabi (ਪੰਜਾਬੀ)
lang:pl Polish (Polski)
lang:pt Portuguese (Português)
lang:ro Romanian (limba română)
lang:ru Russian (Русский)
lang:si Sinhala (සිංහල)
lang:sk Slovak (slovenčina)
lang:sl Slovene (slovenski jezik)
lang:sv Swedish (Svenska)
lang:ta Tamil (தமிழ்)
lang:te Telugu (తెలుగు)
lang:th Thai (ไทย)
lang:tl Tagalog (Tagalog)
lang:tr Turkish (Türkçe)
lang:uk Ukrainian (українська мова)
lang:ur Urdu (ﺍﺭﺩﻭ)
lang:vi Vietnamese (Tiếng Việt)
lang:zh Chinese (中文)

Searching for lang:chr, lang:iu, lang:sk seems to fail, as tweets matching the keywords are returned instead of the language.

There are also some special language codes that work. For example:

• lang:und for undefined language
• lang:qam for tweets with mentions only (works for tweets since 2022-06-14)
• lang:qct for tweets with cashtags only (works for tweets since 2022-06-14)
• lang:qht for tweets with hashtags only (works for tweets since 2022-06-14)
• lang:qme for tweets with media links (works for tweets since 2022-06-14)
• lang:qst for tweets with a very short text (works for tweets since 2022-06-14)
• lang:zxx for tweets with either media or Twitter Card only, without any additional text (works for tweets since 2022-06-14)

Common clients:

source: should work for any API client, try putting the client name in quotes or replace spaces with underscores. This is the App name field that you can alter in the developer app configuration page, so anyone can set anything here and appear to tweet from a made up client.

You cannot copy an existing name. This operator needs to be combined with something else to work, eg: lang:en These are some common ones:

twitter_web_client
twitter_web_app
twitter_for_iphone
twitter_for_ipad
twitter_for_mac
twitter_for_android
twitter_ads
tweetdeck
tweetdeck_web_app
twitter_for_advertisers
twitter_media_studio
cloudhopper (tweets via sms service)

facebook
instagram
twitterfeed
tweetbot.net
IFTTT

"LG Smart Refrigerator"
"GUCCI SmartToilet™"