This dataset contains raw tweets (posts) scraped from X (formerly Twitter) discussing the September 6th 2026 regional election in Saxony-Anhalt, Germany. It was collected to study partisan stance, narrative spread and disinformation around the vote, in which the Alternative für Deutschland (AfD) party won almost 44% of the vote. The corpus covers 33,964 tweets from 15,073 accounts, posted between August 31st and September 7th 2026, and includes full tweet text, engagement metrics, and rich author-level metadata for each post.

The dataset underlies the analysis in "The 2026 Saxony-Anhalt Election on X: Stance, Narratives and Disinformation."

Supported Tasks

Stance / political sentiment classification: classifying tweets as pro-AfD, anti-AfD/pro-other party, or neutral.
Disinformation and narrative detection: identifying election-fraud claims, coordinated messaging, and other misleading narratives.
Bot / inauthentic-behaviour detection: using author metadata (account age, posting volume, verification status) to flag automated or coordinated accounts.
Engagement and amplification analysis: studying how likes, retweets, quotes, views and bookmarks are distributed across political camps.

Languages

Tweets are predominantly in German (de), with some in English (en) and other languages. The lang field records the language detected for each tweet.

Dataset Structure

Data Instances

Each row represents a single tweet (or retweet/reply/quote) matching the collection keywords, along with metadata about its author at the time of collection.

Data Fields

Field	Type	Description
`id`	string	Unique tweet ID.
`text`	string	Full text content of the tweet.
`retweetCount`	int64	Number of retweets.
`replyCount`	int64	Number of replies.
`likeCount`	int64	Number of likes.
`quoteCount`	int64	Number of quote tweets.
`viewCount`	int64	Number of views.
`bookmarkCount`	int64	Number of bookmarks.
`conversationId`	string	ID of the conversation thread the tweet belongs to.
`displayTextRange`	list	Start/end character offsets of the displayed tweet text.
`edit`	unknown	Edit-history metadata, if the tweet was edited.
`entities`	unknown	Parsed entities (hashtags, mentions, URLs, etc.) found in the tweet.
`isNoteTweet`	bool	Whether the tweet is a long-form "Note" tweet.
`isQuoteStatus`	bool	Whether the tweet is a quote tweet.
`isReply`	bool	Whether the tweet is a reply.
`isTranslatable`	bool	Whether X offers a translation for the tweet.
`lang`	string	Detected language code of the tweet.
`source`	string	Client/app used to post the tweet.
`type`	string	Tweet type (e.g. tweet, retweet, reply, quote).
`url`	string	Canonical URL of the tweet.
`viewState`	string	View-count visibility state.
`createdAt`	string	Timestamp when the tweet was posted.
`author`	unknown	Nested author object as returned by the scraper.
`searchTerm`	string	Keyword/query used to collect this tweet.
`tweetUrl`	string	URL to the tweet.
`twitterUrl`	string	Alternate URL to the tweet (twitter.com domain).
`authorId`	string	Unique ID of the author.
`authorUsername`	string	Author's @handle.
`authorName`	string	Author's display name.
`authorFollowers`	int64	Number of followers.
`authorFollowing`	int64	Number of accounts followed.
`authorVerified`	bool	Legacy verification status.
`authorIsBlueVerified`	bool	Whether the account has X Premium/Blue verification.
`authorIsVerified`	bool	General verification flag.
`authorProfilePicture`	string	URL of the author's profile picture.
`authorCoverPicture`	string	URL of the author's cover picture.
`authorProfileBannerUrl`	string	URL of the author's profile banner.
`authorDescription`	string	Author's bio text.
`authorCreatedAt`	string	Date the author's account was created.
`authorStatusesCount`	int64	Total number of tweets posted by the author.
`authorMediaCount`	int64	Total number of media items posted by the author.
`authorProtected`	bool	Whether the author's account is protected/private.
`authorUrl`	string	URL listed on the author's profile.
`authorFavouritesCount`	int64	Number of tweets liked by the author.
`authorIsTranslator`	bool	Whether the author is part of X's translator program.
`authorPossiblySensitive`	bool	Whether the author's content is flagged as possibly sensitive.
`authorPinnedTweetIds`	list	IDs of tweets pinned to the author's profile.
`authorVerifiedType`	string	Type of verification (e.g. government, business).
`authorProfileBio`	unknown	Structured bio object.
`authorGrokTranslatedBio`	unknown	Grok-translated version of the bio, if available.
`authorHasGraduatedAccess`	bool	Whether the author has graduated developer/API access.
`authorParodyCommentaryFanLabel`	string	Parody/commentary/fan-account label, if any.
`authorProfessional`	dict	Professional account category metadata.
`authorProfileImageShape`	string	Shape of the profile image (e.g. circle, square).
`authorProfileInterstitialType`	string	Interstitial warning type shown on the profile, if any.
`authorProfileTranslatorType`	string	Translator status type for the profile.
`authorSuperFollowEligible`	bool	Whether the author is eligible for Super Follows.
`article`	unknown	Linked X Article object, if the tweet is/contains one.
`authorAffiliatesHighlightedLabel`	unknown	Highlighted affiliate organization label, if any.
`contentDisclosure`	unknown	Content disclosure/labeling metadata (e.g. state-affiliated media).
`noteTweet`	unknown	Full content object for long-form Note tweets.
`card`	unknown	Rich card metadata (e.g. link preview, poll).
`quotedTweet`	unknown	Nested object for the quoted tweet, if any.
`media`	list	List of media objects (images/videos/gifs) attached to the tweet.
`possiblySensitive`	bool	Whether the tweet is flagged as possibly sensitive.
`authorProfileDescriptionLanguage`	string	Detected language of the author's bio.
`mediaUrls`	list	URLs of attached images.
`videoUrls`	list	URLs of attached videos.
`authorLocation`	string	Free-text location listed on the author's profile.
`quotedTweetId`	string	ID of the quoted tweet, if any.
`imageUrls`	list	URLs of attached images (duplicate/alternate of `mediaUrls`).
`authorWithheldInCountries`	list	Countries where the author's account is withheld.
`conversationControl`	dict	Reply-permission settings for the tweet.
`previousCounts`	dict	Prior snapshot of engagement counts, if tracked over time.
`inReplyToUserId`	string	ID of the user being replied to.
`inReplyToUsername`	string	Username of the user being replied to.
`inReplyToId`	string	ID of the tweet being replied to.
`gifUrls`	list	URLs of attached GIFs.
`communityNote`	dict	Attached Community Note, if any.
`place`	dict	Geotagged place metadata, if any.
`communityId`	string	ID of the X Community the tweet was posted in, if applicable.
`unmentionedUserIds`	list	IDs of users tagged without an @mention (e.g. via media tags).

Data Splits

The dataset is provided as a single unsplit collection of tweets. No train/validation/test split is defined.

Dataset Creation

Curation Rationale

The dataset was collected to analyze how partisan narratives, engagement asymmetries and disinformation circulated on X around a German regional election, and to study the prevalence of AI-generated imagery and coordinated/inauthentic accounts in that context.

Source Data

Tweets were scraped from X (Twitter) using keyword/hashtag search terms related to the "Sachsen-Anhalt" ("Saxony-Anhalt") election, over the period August 31st–September 7th 2026. The searchTerm field records which query matched each tweet.

Annotations

The raw dataset released here is unannotated (no stance, narrative or authenticity labels included). A separate hand-coded sample (520 tweets, stance-labelled) and narrative/disinformation tagging were produced for the accompanying analysis but are not part of this raw export.

Personal and Sensitive Information

This dataset contains publicly posted tweets and public profile metadata (usernames, display names, bios, follower counts, profile images, etc.) as made available via X at the time of collection. It may contain personal opinions, and author-level fields could be used to identify individuals. Users of this dataset should comply with X's Developer Policy and applicable data protection regulations (e.g. GDPR) when processing or redistributing it, and should consider the ethical implications of studying identifiable social media users.

Considerations for Using the Data

Social Impact of Dataset

This dataset can support research into election integrity, disinformation and platform dynamics. Care should be taken not to use it to target, harass, or deanonymize individual accounts, including those spreading misinformation.

Discussion of Biases

The dataset reflects only content matching the chosen search terms and only accounts active on X; it does not represent the full population of voters or the broader information ecosystem (e.g. other platforms, private groups). Engagement metrics (likes, views, retweets) are subject to platform ranking and algorithmic amplification effects, which the accompanying analysis discusses in more detail.

Other Known Limitations

Engagement counts (e.g. viewCount, likeCount) reflect values at scrape time and may have continued to change afterward.
Some fields (e.g. article, card, communityNote, quotedTweet, noteTweet) are only populated for tweets with the relevant feature and are otherwise empty/null.
Media URLs may expire or become inaccessible over time.

Additional Information

Dataset Curators

Collected and curated by the authors of "The 2026 Saxony-Anhalt Election on X: Stance, Narratives and Disinformation" using the UncovAI toolset.

Citation Information

Coming soon

Downloads last month: 19

Total file size:

297 MB