PDID / Research resources

Public codebook

Political Deepfakes Incidents Database — Variable Reference

On smaller screens, scroll tables sideways to read all columns.

A reference for researchers, journalists, and others using PDID data.

Updated October 14, 2026. Identifier definitions have been aligned with the current coding guidance. Other field names follow the public variable reference and may differ across dataset versions. Not every field is displayed on the website or included in chart-summary downloads.

Use date_posted, rather than the numerical order of the identifiers, to analyze when reports appeared.

Database Overview

PDID documents political deepfakes, cheapfakes, and media alleged to be fake, including authentic media. The identifiers distinguish three levels:

Event (event_id): a broader political episode, campaign, or controversy connecting related incidents.
Incident (incident_id): one distinct image, video, or audio item. The same underlying media can appear in multiple reports.
Report (report_id): one coded appearance of an incident in a distinct post or upload. Ordinary native reposts are captured through engagement metrics rather than separate reports.

Unit of analysis guidance:

For analysis of coded posts, group by report_id.
For analysis of distinct media items, group by incident_id.
For broader political episodes, group by event_id where available.
Multiple coding rows can exist for a report. Use the validated row for one final record per report; do not treat every coder row as a separate post.

Important classification note: The type field (deepfake / cheapfake / real / unclear) reflects coder judgment at the time of coding. It is not machine-verified ground truth. The HiveModeration_Score provides a complementary AI-detection signal. external_verification also identifies whether news or fact-checkers have said whether the media is real or fake.

Variable Reference Table

Meta-Data

Variable	Type	Description
`event_id`	Linked identifier	Groups incidents connected to a broader political episode, campaign, or controversy.
`report_id`	Auto-number	Identifies one coded post or upload. Multiple coding rows may share this identifier; use the validated row for analysis.
`incident_id`	Linked identifier	Identifies a distinct media item across posts or uploads. Group by this identifier for distinct-media analysis.
`thread`	Number	Order of this item within a multi-item post (1 = first, 2 = second, etc.).
`date_coded`	Date	Date the row was entered into the database.
`Validated`	Checkbox	Whether a coder has reviewed and verified this entry.
`Title`	Text	Headline-style description of the incident, for website use.
`Tags`	Multi-select	Related tags, for website use.

Media & Content

Variable	Type	Description
`url`	URL	Link to the post where the content was found. May be unavailable if content was removed.
`source`	Single select	Platform where the content was posted. `Twitter/X`, `Facebook`, `YouTube`, `Reddit`, `Politifact`, `The New York Times`.
`file_with_watermark`	Attachment	Version of the file with a PDID watermark.
`screenshot`	Attachment	Screenshot of the post in context (showing account name, caption, engagement).
`date_posted`	Date	Date the content was posted at the URL.
`format`	Single select	Format of the manipulated content. Options: `image`, `video`, `audio`.
`transcript`	Long text	Transcript of spoken content. `NA` if no spoken content.
`summary_content`	Long text	1–2 sentence description of what the content appears to depict.

Verification & Classification

Variable	Type	Description
`type`	Single select	Coder's classification of the content.
`comment_type`	Long text	Coder's explanation for their `type` classification.
`HiveModeration_Score`	Text	AI-detection probability score from Hive Moderation tool
`sharer_faketype`	Single select	How the sharer presented the content.
`external_verification`	Single select	Whether external sources have verified the content's authenticity.
`links_verification`	Long text	URLs to fact-checks, news articles, or platform labels verifying authenticity.
`evidence_fake`	Long text	Description of specific evidence that the content is fake.
`watermark`	Single select	Whether a visible watermark or AI label is present. Options: `yes`, `no`, `unclear_unknown`.

type options:

Value	Definition
`deepfake`	AI/ML-generated or manipulated synthetic media
`cheapfake`	Low-tech manipulation (crop, speed change, out-of-context reuse)
`real`	Authentic media, possibly shared in a misleading context
`unclear/unknown`	Authenticity cannot be determined

sharer_faketype options:

Value	Definition
`real_authentic`	Sharer presented content as genuine
`fake`	Sharer acknowledged content is manipulated or AI-generated
`unclear`	Cannot determine how sharer framed it

external_verification options:

Value	Definition
`verified_fake`	External source confirmed content is fake
`verified_real`	External source confirmed content is authentic
`authenticity_disputed`	External sources disagree or are inconclusive
`unknown`	No evidence of verification attempts
`no evidence`	No external sources address authenticity

Engagement Metrics

Variable	Type	Description
`like_count`	Text	Likes/favorites at time of coding. `NA` if not displayed.
`view_count`	Text	Views at time of coding. `NA` if not displayed.
`comment_count`	Text	Comment count at time of coding. `NA` if not displayed.
`share_count`	Text	Shares/retweets at time of coding. `NA` if not displayed.

Original Source

The "original source" is the account or entity that first created or posted the deepfake. This may differ from the sharer (the account found at the coded URL).

Variable	Type	Description
`original_source_name`	Multi-select	Name of the original creator/poster.
`other_original_source_name`	Text	Free-text name if not in dropdown.
`original_source_username`	Text	Social media handle of original source.
`original_source_type`	Multi-select	Entity type of original source. Options: `individual`, `group`, `NA`, `unclear_unknown`.
`original_source_actor_type`	Multi-select	Political/institutional actor type.
`original_source_country`	Single select	Country of original source.
`other_original_source_country`	Text	Free-text country if not in dropdown.
`comments_coding_original_source`	Long text	Coder's notes on original source identification.

original_source_actor_type options:

Value	Definition
`political_leader_figure`	Elected official, candidate, or prominent political figure
`government_entity`	Official government body or institution
`private_ngo`	Private individual, company, or NGO
`individual`	Generic individual
`unclear_unknown`	Cannot determine

Sharer

The "sharer" is the specific account at the coded URL, not necessarily the original creator.

Variable	Type	Description
`sharer_name`	Single select	Name of the sharing account.
`other_sharer_name`	Text	Free-text name if not in dropdown.
`sharer_username`	Text	Social media handle of the sharer.
`sharer_type`	Multi-select	Type of entity sharing the content. Options: `individual`, `group`, `bot`, `high_impact_user`, `non-identifiable`.
`sharer_job`	Multi-select	Occupation/role of the sharer. Options: `politician`, `public_figure`, `journalist`, `researcher_academic`, `random_user`, `other`, `unclear_unknown`.
`sharer_country`	Multi-select	Country the sharer is associated with.
`other_sharer_country`	Text	Free-text country if not in dropdown.
`comments_sharer_coding`	Long text	Coder's notes on sharer identification.

Targets

A "target" is a person, group, or entity depicted in the deepfake. Up to 10 targets can be coded per incident (target_one through target_ten). The fields below apply to each target slot (replace [n] with one, two, etc.).

Variable	Type	Description
`target_[n]_name`	Single select	Name of the target.
`target_[n]_name_other`	Text	Free-text name if not in dropdown.
`target_[n]_sentiment`	Single select	How the deepfake portrays the target. Options: `Positive`, `Negative`, `Neutral`.
`target_[n]_type_macro`	Single select	Broad target category. Options: `individual`, `group`, `unclear_unknown`.
`target_[n]_type_micro`	Single select	Specific target role.
`target_[n]_country`	Single select	Country the target is associated with.
`target_[n]_country_other`	Text	Free-text country if not in dropdown.
`target_[n]_responded`	Checkbox	Whether the target publicly responded to the deepfake.
`target_[n]_response`	Long text	Description or quote of the target's response.
`comments_target_coding`	Long text	Coder's notes on target identification.

target_[n]_type_micro options:

Value	Definition
`politician`	Elected official or candidate
`public_figure`	Celebrity, influencer, or other prominent person
`journalist`	Reporter or media professional
`random_user`	Private individual with no public profile
`government_entity`	Government institution or body
`private`	Private company or organization
`ngo`	Non-governmental organization
`marginalized_group`	Group defined by race, religion, gender, sexuality, etc.
`unclear_unknown`	Cannot determine

Content & Topic

Variable	Type	Description
`content_depicted`	Multi-select	The topic or subject matter of the deepfake.
`harm_depicted`	Multi-select	Type of harm shown or implied within the content itself.
`comments_content_topic`	Long text	Coder's notes on content and harm depicted coding.
`social_policy_sector`	Multi-select	Policy domain the deepfake relates to.
`comments_social_policy_sector`	Long text	Coder's notes on sector coding.
`context_deepfake`	Long text	Broader political/social context at time of posting.
`text_around_deepfake`	Long text	Caption or text accompanying the post.

content_depicted options:

Value	Definition
`scandal`	Target in a compromising situation
`crime`	Target committing a crime
`sex_related`	Sexual content
`crisis_event`	Emergency or disaster
`election`	Electoral processes or campaigns
`government_legal_system`	Laws, courts, or government institutions
`war_violence`	Armed conflict or violence
`international_relations`	Diplomacy or cross-border relations
`marginalized_group`	Targets a marginalized community
`deepfake_regulation`	Deepfake laws or AI policy

harm_depicted options:

Value	Definition
`violence`	Physical harm, threats, or armed conflict depicted
`human_rights`	Fundamental rights violations depicted
`discrimination`	Bias or prejudice against a group depicted
`property`	Property damage or theft depicted
`environmental`	Environmental destruction depicted
`other`	Other harm depicted
`non-identifiable`	No specific harm depicted

Real-World Harm

Variable	Type	Description
`harm_evidence`	Multi-select	Types of real-world harm documented as a result of this deepfake. Only coded when evidenced.
`comments_harms_context`	Long text	Coder's description of harm evidence found (or not found).

harm_evidence options:

Value	Definition
`violence`	Physical violence occurred or was incited
`human_rights`	Human rights violations resulted
`discrimination`	Documented discrimination or harassment
`property`	Property damage or financial harm
`financial_loss`	Economic harm (fraud, scam, market manipulation)
`environment`	Environmental harm resulted
`political_interference`	Interference with democratic processes
`other`	Other documented harm
`non_identifiable`	No documented real-world harm found

Framing & Narrative

Variable	Type	Description
`communication_goal`	Multi-select	Sharer's apparent intent in sharing the content.
`comments_communication_goal`	Long text	Coder's notes on intent coding.
`core_frame`	Multi-select	Dominant rhetorical frame (Semetko & Valkenburg 2000). Options: `conflict`, `human_interest`, `economic_consequences`, `morality`, `responsibility`.
`comments_core_frame`	Long text	Coder's notes on framing.
`hero`	Text	Who is portrayed as the protagonist or positive force.
`villain`	Text	Who is portrayed as the antagonist or negative force.
`plot`	Text	The story being told (1–2 sentences).
`moral`	Text	The implied lesson or takeaway.
`comments_policy_narrative`	Long text	Coder's notes on narrative structure.