This dataset is discussed in far more detail in the corresponding paper, AMMeBa: A Large-Scale Survey and Dataset of Media-Based Misinformation In-The-Wild.

Background

The rise of convincing, photorealistic AI-generated images and video have heightened already intense concern over online misinformation and its associated harms. However, despite huge coverage in the press and interest by the general public, it's not clear if AI is widely used in misinformation. In fact, there is little systematic data available whatsoever about the forms misinformation takes online, the use of images and video in misinformation contexts, and what types of manipulations are taking place.

The AMMeBa (Annotated Misinformation, Media-Based) dataset seeks to provide a survey of online misinformation, allowing first-of-its-kind quantification of manipulations like deepfakes and photoshopped media as well as trends in how those populations are changing over time.

Recognizing the enormous value and work of fact checkers, AMMeBa uses publicly-available fact checks to identify misinformation claims, which were annotated by highly trained human annotators, providing detailed characterization of the misinformation claim. Media-based misinformation, which uses images, video and audio to bolster the claim, are a particular focus, especially images.

Annotations took place over two years. The resulting dataset comprises millions of individual hand-applied labels, applied to over a hundred thousand English-language fact checks published between 1995 and today. More than fifty thousand misinfo-associated images were identified and annotated.

Findings

• Online misinformation is popularly conceptualized as false claims and rumors rendered in text. Our data indicates that the majority of misinformation (recently, about 80%) involves media of some kind: images, video, or audio.
• Images are historically the most common type of media associated with misinformation. However, in the past two years, video-based misinformation has become increasingly common and is now the most common type of media associated with misinformation.
• Among images, screenshots are common, peaking at about 1/5th of misinformation-associated images. The majority of these are screenshots of social media posts, nearly 20% are screenshots of fake social media posts.
• While image-based misinformation is commonly thought of as consisting of photoshop-like manipulations, or, more recently, AI-generated content, our data show that the most common type historically is context manipulations without any pixel manipulation i.e. the original un-edited image is shown alongside a false claim about what that image shows.
  • The prevalence of technologically simple context manipulations underscores the fact that misinformation does not need to be sophisticated or elaborate to be effective.
• While widespread concern around the use of deepfakes in misinformation began in 2018, our data show that AI-generated content was a negligible proportion of overall image-based misinfo until early 2023, when it exploded in popularity. By the time data annotation ended, it accounted for nearly 30% of all fact checked content manipulations.

Dataset Notes

Image URLs

Image URLs were obtained in a best-effort manner. We provide them as a possible pointer to the correct image. However, URLs are absent for several reasons:

1. Attrition: The image has been removed from that location; see "Data Attrition" in the paper. We are working to identify other versions of the images, if available, and will make them available in dataset updates.
2. URL Dynamism: The images were obtained by following a fact check link to the original page or an archived version of it. Some pages, particularly archival services, dynamically generate image URLs on load or update the URLs periodically. This instability in the URL means collected URLs are soon useless for these images.

In the majority of cases, though, the URL under misinfo_source in all provided CSVs will point to the page where the image occurred, and in general they are still present (this is checked explicitly by raters when a fact check / source is passed to a subsequent stage, like Stage 1M → Stage 2M. If the entry is not "disqualified," then the image was present on the page at the time of subsequent annotation, and may still be fetchable by matching against the provided hashes.

Image Hashes

To allow users to fetch the images themselves, we provide three hashes of the image data. These hashes use the open-source "imagehash" Image Hashing Library from Github (README, with explanations of the hash functions).

• Available here: https://github.com/JohannesBuchner/imagehash
• Version used (git hash): 71db1d3324ec8344c507f5a89a529ff0c0156678

This library also requires Python packages Python Imaging Library PIL:

• Available here: https://github.com/python-pillow/Pillow.git
• Version: 9.4.0
• Version used (git hash): a5bbab1c1e63b439de191ef2040173713b26d2da

and Numpy:

• Available here: https://github.com/numpy/numpy
• Version: 1.24.1
• Version used (git hash): 6f491e08e8fa84d939f91f1404ea4cc8ccbf9bc1

Three hashes are provided: average_hash, phash (perceptual hash), and colorhash. Each is provided as a hexademical string of the hash. To compute hash distance (using Hamming Distance), the hashes must be converted from strings using a "Hash Rehydrating Function." The functions used to perform these are listed below:

Hash Name / Field	Hash Computation Function	Hash Rehydrating Function ("R")
average_hash	str(imagehash.average_hash(x, hash_size=8, mean=np.mean))	imagehash.hex_to_hash(x)
phash	str(imagehash.phash(x, hash_size=8))	imagehash.hex_to_hash(x)
colorhash	str(imagehash.colorhash(x, binbits=3))	imagehash.hex_to_flathash(x, hashsize=3)

Hash distances are computed for possible match image Y using function H on the rehydrated versions (using function "R", obtained from the table above based on the type of hash) of the query image's hash value (as a string) from the table and the hash of Y using:

R([query_image, hashes, H]) - H(Y)

Dataset Fields

The field descriptions are too long to put into the per-file information, and so are produced here.

Fact Check Fields

Enumerates the fact checks used to sample misinformation claims in this study.

Fields:

• fact_check_url: The URL of the fact check.
• review_date: The date of the fact check, provided by the fact check publisher, rendered as YYYY.MM.DD.

Image Metadata Fields

Contains metadata for images referenced in the dataset, and links, where available.

Fields:

• image_id: The image ID. The Image ID is computed as a hash of the image bytes.
• mime: The image MIME type.
• height: The image height, in pixels.
• width: The image width, in pixels.
• image_size_bytes: The size of the image, in bytes.
• num_faces: The number of faces detected in the image.
• total_face_area: The sum of the number of pixels in the bounding boxes of the detected faces.
• has_text: Whether or not text was detected in the image.
• url: The URL to the image, or a close version of it. See Dataset Notes - Image URLs in the description.
• average_hash: The average hash of the image analyzed (not the image at url, though they may be the same). See Dataset Notes - Image Hashes in the description.
• phash: The perceptual hash of the image analyzed (not the image at url, though they may be the same). See Dataset Notes - Image Hashes in the description.
• colorhash: The color hash of the image analyzed (not the image at url, though they may be the same). See Dataset Notes - Image Hashes in the description.

Stage 1 Fields

No eligibility requirements exist for Stage 1.

Fields:

• fact_check_url: The URL of the fact check.
• submission_time: The time of the rating, rendered as YYYY.MM.DD HH.MM.SS, in UTC, regardless of rater location.
• disqualified: Whether or not this fact check or response was disqualified from further consideration.
• disqualification_reason: Indicates the reasons that the fact check or response was disqualified from further consideration. This is a categorical field, with values:
  • possibly_not_misinfo: The fact check indicates that the claim fact checked may be true or is otherwise possibly not misinformation.
  • multiple_claims_fact_checked: The fact check addresses multiple, distinct claims simultaneously (an "anthology" fact check).
  • non_english_fact_check: The fact check was primarily written in a language other than English.
  • fact_check_website_broken: The fact check website failed to load at the time of rating or the URL did not route to a fact check.
  • invalidated: The response was invalidated, which can occur for a variety of reasons including changes to data collection methodology and quality control.
• availability: Indicates whether or not the fact check provides a link to the misinformation, or a capture of it, as it existed on the open web. This is a categorical field, with values:
  • misinformation_accessible: A link to the misinformation, as it originally existed on the open web, is provided by the fact check (either as a direct link or a link to an archived version) and this link is loadable at the time of rating.
  • misinfo_inaccessible_screenshot_provided: No working link is provided, but a screenshot of the original misinformation is provided on the fact check page.
  • misinfo_inaccessible: No working link nor a capture of it is provided on the fact check.
• media_involvement: Indicates the type of media, if any, is involved and material to the misinformation claim. Materiality is defined as: "should this media be removed, the misinformation claim would be substantially less effective, nonsensical, or entirely absent." This is a categorical field, with values:
  • image_based: The misinformation claim relies on an image.
  • video_based: The misinformation claim relies on a video.
  • audio_based: The misinformation claim relies on an audio clip. Note that videos entirely without visual content are considered audio clips.
  • multiple_media_based: The misinformation claim relies on multiple media, possibly of mixed modality.
  • not_media_based: The misinformation claim does not rely on media at all.
  • insufficient_info: The fact check does not provide enough information to make a determination.
  • misinfo_unknown_other_reason: The media involvement of the misinformation claim being fact checked could not be determined for a reason that does not fit neatly into the above categories.
• misinfo_source: The URL to a webpage displaying the misinformation claim as it appeared on the open web, if provisioned by the fact check.
• misinfo_source_type: The nature of the misinformation claim source, if provisioned by the fact check. This is a categorical field, with values:
  • original_source: The misinformation claim source provisioned by the fact check is a direct link to the original source.
  • webpage_archive: The misinformation claim source provisioned by the fact check is an archived version of the original webpage.
  • screenshot_only_webpage_archive: The misinformation claim source provisioned by the fact check is an archived version of the original webpage, and this archive takes the form of a whole- or nearly whole-webpage screenshot.
• image_id: When the misinformation claim is materially related to an image, that image was identified by raters for further characterization in Stage 2 (see below). The Image ID is computed as a hash of the image bytes.

Stage 1M Fields

Stage 1M annotations were collected on fact checks that were rated as "misinformation_accessible" and "multiple_media_based" in Stage 1.

Fields:

• fact_check_url: The URL of the fact check.
• misinfo_source: The URL to a webpage displaying the misinformation claim as it appeared on the open web, as provided by a rater in Stage 1.
• submission_time: The time of the rating, rendered as YYYY.MM.DD HH.MM.SS, in UTC, regardless of rater location.
• disqualified: Whether or not this fact check was disqualified from further consideration. Note: in some cases, reaching a disqualification criteria did not prevent respondents from conducting further annotations.
• disqualification_reason: Indicates the reasons that the fact check was disqualified from further consideration. This is a categorical field, with values:
  • incorrect_s1_assessment: This fact check was routed to Stage 1M due to an incorrect response on the part of an annotator in Stage 1 (e.g., fact check was of the "anthology" type but was not marked as multiple_claims_fact_checked, or the claim being fact checked did not actually rely on multiple media).
  • fact_check_website_broken: The fact check website failed to load at the time of rating or the URL did not route to a fact check.
  • misinfo_source_website_broken: The URL provided as the/a original source of the misinformation claim, or an archive thereof, failed to load at the time of rating or routed to an unrelated page.
  • invalidated: The response was invalidated, which can occur for a variety of reasons including changes to data collection methodology and quality control.
• misinfo_source_type: The nature of the misinformation claim source, if provisioned by the fact check. This information should accord with the rating provided in Stage 1, although some Stage 1 annotations were completed before this criteria was added to the task. This is a categorical field, with values:
  • original_source: The misinformation claim source provisioned by the fact check is a direct link to the original source.
  • webpage_archive: The misinformation claim source provisioned by the fact check is an archived version of the original webpage.
  • screenshot_only_webpage_archive: The misinformation claim source provisioned by the fact check is an archived version of the original webpage, and this archive takes the form of a whole- or nearly whole-webpage screenshot.
• images_involved: The media involved in the misinformation claim includes at least one materially-related image.
• number_images: The number of images involved that are material to the misinformation claim, as an integer. If the rater could not determine this number confidently (including because not all assets on the source page loaded correctly), or their response was malformed, a value of 'Unk' is provided.
• videos_involved: The media involved in the misinformation claim includes at least one materially-related video.
• number_videos: The number of videos involved that are material to the misinformation claim, as an integer. If the rater could not determine this number confidently (including because not all assets on the source page loaded correctly), or their response was malformed, a value of 'Unk' is provided.
• audio_involved: The media involved in the misinformation claim includes at least one materially-related audio clip.
• number_audio: The number of audio clips involved that are material to the misinformation claim, as an integer. If the rater could not determine this number confidently (including because not all assets on the source page loaded correctly), or their response was malformed, a value of 'Unk' is provided.
• image_id: If the multiple media-based misinformation claim depends on at least one image, and at least one of the images is visible, one of the images is selected by the Stage 1M raters for further characterization in Stage 2M (see below). Raters were able to submit a single image at a time due to task constraints, and were shown images already submitted for this misinformation claim by other raters to prevent duplicates. Raters were able to indicate when all images had been identified. The Image ID is computed as a hash of the image bytes.

Stage 2 Fields

Stage 2 annotations were collected on fact checks that were rated as "misinformation_accessible" and "image_based" and where the misinfo-relevant image could be identified by the rater in Stage 1. During Stage 2, raters were presented with a link to the fact check, a link to the original misinformation source, and presented with the image identified in Stage 1.

• Raters were instructed to rely on content in the body of the fact check when making judgements that are not objective and criteria based; raters were instructed not to make personal assessments.
• Absence of value indicates that the field was not applicable given upstream rater answers and was not presented to the rater as an option for rating.
• Non-categorical fields may contain one of three values: 'true', 'false', or 'indeterminate.' Indeterminate indicates that the rater could not, or did not have enough information, to determine the value of this field.

Fields:

• fact_check_url: The URL of the fact check.
• misinfo_source: The URL to a webpage displaying the misinformation claim as it appeared on the open web, as provided by a rater in Stage 1.
• image_id: The ID of the image that was identified in Stage 1 and assessed by the rater in conjunction with the fact check and misinformation claim. Image ID is computed as a hash of the image bytes.
• submission_time: The time of the rating, rendered as YYYY.MM.DD HH.MM.SS, in UTC, regardless of rater location.
• disqualified: Whether or not this fact check was disqualified from further consideration.
• disqualification_reason: Indicates the reasons that the fact check was disqualified from further consideration. This is a categorical field, with values:
  • incorrect_s1_assessment: This fact check was routed to Stage 1M due to an incorrect response on the part of an annotator in Stage 1 (e.g., fact check was of the "anthology" type but was not marked as multiple_claims_fact_checked).
  • incorrect_s1_image: The image identified in Stage 1 was included erroneously. This can occur for a variety of reasons, including the Stage 2 rater regarding the image as irrelevant to the misinformation claim.
  • fact_check_website_broken: The fact check website failed to load at the time of rating or the URL did not route to a fact check.
  • invalidated: The response was invalidated, which can occur for a variety of reasons including changes to data collection methodology and quality control.
• image_type: The top-level image characterization. This is a categorical field, with values:
  • basic: The image consists of a single pictorial visual element, like a photograph or a painting, with no overlaid text or other elements except small watermarks or small annotations.
  • screenshot: The image contains elements that indicate it is a screenshot from a digital device, namely graphical UI elements or other features that identify it as such unambiguously.
  • complex: The image contains multiple elements that do not or are not made to appear to be a single visual element. It could be a collage of images, or contain overlaid digital text, etc.
  • indeterminate: The rater is unable to make a clear determination about the top-level image category.
• screenshot_type: The screenshot subtype. If this category is not applicable (i.e., the image is not a screenshot), this value for this field is empty. This is a categorical field, with values:
  • social_media_post: The screenshot depicts a post to a social media network.
  • fake_social_media_post: The screenshot depicts a post to a social media network that the fact check indicates is fabricated.
  • unspecified: The screenshot depicts something other than a post to a social media network.
• image_watermark: Indicates whether the image has an unobtrusive watermark (a subjective rater judgment).
• image_annotation: Indicates whether the image has an unobtrusive annotation (a subjective rater judgment), such as a red circle or small arrow.
• image_text: Indicates whether or not the image contains text. This is a categorical field, with values:
  • legible: The image contains legible text.
  • illegible_or_absent: The image does not contain text, or contains text that is small or distorted enough to be illegible.
• image_text_misinfo_relevance: Indicates the degree to which text present in the image is relevant to the misinformation claim. This is a categorical field, with values:
  • only_text_important: The image text is relevant, or contains, the misinformation claim. Other visual content in the image, if present, is not material to the misinformation claim.
  • relevant: The image text is relevant, or contains, the misinformation claim.
  • irrelevant: The image text is incidental to the misinformation claim.
  • indeterminate_foreign: The image text relevance to the misinformation claim cannot be determined because it is in a foreign language and the fact check does not provide a sufficient translation.
  • indeterminate: The image text relevance could not be determined by the rater.
• content_manipulation: The content of the image has been manipulated, or synthesized entirely, in a way that is relevant to the misinformation claim.
• text_manipulation: Indicates text occurring on an object in the image has been added or changed in a way that is not intended to appear as overlaid digital text in a way that is relevant to the misinformation claim.
• chyron_manipulation: Indicates a graphical element that appears as a lower-third graphic in a news broadcast has been added or manipulated in a way that is relevant to the misinformation claim.
• ai_manipulation: Indicates the image has been modified or synthesized using AI in a way that is relevant to the misinformation claim.
• premanipulation_image: Indicates that the fact check provides a version of the content manipulated image prior to content manipulation.
• premanipulation_image_id: The ID of the pre-manipulation image, if provided by the fact check. Image ID is computed as a hash of the image bytes.
• context_manipulation: The misinformation claim ascribes a false or misleading context to the image.
• self_miscontextualizing: Indicates that the false or misleading context is imparted onto the image by text present within the image itself.
• circumstance_manipulation: Indicates that the circumstances or events in the image are described misleadingly.
• identity_manipulation: Indicates that a person, group or object in the image is misidentified.
• location_manipulation: Indicates that the location depicted in the image has been misdescribed.
• datetime_manipulation: Indicates that the date the image was captured is misstated.
• atypical_manipulation: The context ascribed to the image has been manipulated in a way that is not well captured by the above categories, for instance by omission or incorrectly describing the originating party of the image.
• reverse_image_search: Indicates if a reverse image search was performed by the fact checker and its outcome, if known. This is a categorical field, with values:
  • successful: A reverse image search was performed and successfully identified evidence that was helpful for the investigation of the misinformation claim.
  • unsuccessful: A reverse image search was performed, but it did not successfully identify evidence that was helpful for the investigation of the misinformation claim.
  • performed: A reverse image search was performed, but the outcome is not determinable by the rater.
  • not_performed: A reverse image search is not mentioned in the body of the fact check.
• photograph_of_screen: Indicates that the image contains a screen or other digital display, which is itself displaying content relevant to the misinformation.
• fake_official_document: Indicates if the image is or contains a "fake official document," that is, it is made to appear as though a communication from an official or widely-recognized organization, government, or corporation.

Stage 2M Fields

Stage 2M annotations were collected on fact checks that where the original misinformation was rated as accessible in both Stage 1 and Stage 1M, the multiple media were rated as including at least one image in Stage 1M, the misinfo-relevant image(s) could be identified by the Stage 1M rater, and where the at least two Stage 1M raters have indicated that all misinformation relevant images have been identified. During Stage 2M, raters were presented with a link to the fact check, a link to the original misinformation source, and presented with all images identified in Stage 1M. For all images present in a multiple media misinformation claim, at least one rating was conducted; one image was presented as the "primary" image and all the other images were presented as secondary images, provided for context. Certain fields applied exclusively to the primary image, others apply to all images presented.

• The labels "primary" and "secondary" relate purely to the selection of a single image from among the pool of images associated with this misinformation claim for the purposes of rating and does not reflect its overall importance. All images are treated as the primary image across different ratings for the misinformation claim.
• Raters were instructed to rely on content in the body of the fact check when making judgements that are not objective and criteria based; raters were instructed not to make personal assessments.
• Absence of value indicates that the field was not applicable given upstream rater answers and was not presented to the rater as an option for rating.
• Non-categorical fields may contain one of three values: 'true', 'false', 'indeterminate,' or 'non_primary.' Indeterminate indicates that the rater could not, or did not have enough information, to determine the value of this field; non_primary indicates that the fact check indicates that the field in question can apply to some images presented but either not the primary image or the fact check is not explicit about which image the field applies to. Fields with this property are denoted explicitly in the descriptions below.

Fields:

• fact_check_url: The URL of the fact check.
• misinfo_source: The URL to a webpage displaying the misinformation claim as it appeared on the open web, as provided by a rater in Stage 1 and affirmed in Stage 1M.
• primary_image_id: The ID of the primary image presented, one of the images identified in Stage 1M and assessed by the rater in conjunction with the secondary images, fact check and misinformation claim. Image ID is computed as a hash of the image bytes.
• secondary_image_ids: The ID(s) of the secondary images presented. Different image IDs are separated by colons ':'. This field may be empty if the multiple media misinformation claim involved only one image. Image ID is computed as a hash of the image bytes.
• submission_time: The time of the rating, rendered as YYYY.MM.DD HH.MM.SS, in UTC, regardless of rater location.
• disqualified: Whether or not this fact check was disqualified from further consideration.
• disqualification_reason: Indicates the reasons that the fact check was disqualified from further consideration. This is a categorical field, with values:
  • incorrect_s1m_image: The image identified in Stage 1M was included erroneously. This can occur for a variety of reasons, including the Stage 2M rater regarding the image as irrelevant to the misinformation claim.
  • fact_check_website_broken: The fact check website failed to load at the time of rating or the URL did not route to a fact check.
  • invalidated: The response was invalidated, which can occur for a variety of reasons including changes to data collection methodology and quality control.
• standalone_image: Indicates that the secondary images are not necessary for the primary image's material relevance to the misinformation claim. The primary image being standalone does not imply that the secondary images are not also standalone.
• image_type: The top-level image characterization. This is a categorical field, with values:
  • basic: The image consists of a single pictorial visual element, like a photograph or a painting, with no overlaid text or other elements except small watermarks or small annotations.
  • screenshot: The image contains elements that indicate it is a screenshot from a digital device, namely graphical UI elements or other features that identify it as such unambiguously.
  • complex: The image contains multiple elements that do not or are not made to appear to be a single visual element. It could be a collage of images, or contain overlaid digital text, etc.
  • indeterminate: The rater is unable to make a clear determination about the top-level image category.
• screenshot_type: The screenshot subtype. If this category is not applicable (i.e., the image is not a screenshot), this value for this field is empty. This is a categorical field, with values:
  • social_media_post: The screenshot depicts a post to a social media network.
  • fake_social_media_post: The screenshot depicts a post to a social media network that the fact check indicates is fabricated.
  • unspecified: The screenshot depicts something other than a post to a social media network.
• image_watermark: Indicates whether the image has an unobtrusive watermark (a subjective rater judgment).
• image_annotation: Indicates whether the image has an unobtrusive annotation (a subjective rater judgment), such as a red circle or small arrow.
• image_text: Indicates whether or not the image contains text. This is a categorical field, with values:
  • legible: The image contains legible text.
  • illegible_or_absent: The image does not contain text, or contains text that is small or distorted enough to be illegible.
• image_text_misinfo_relevance: Indicates the degree to which text present in the image is relevant to the misinformation claim. This is a categorical field, with values:
  • only_text_important: The image text is relevant, or contains, the misinformation claim. Other visual content in the image, if present, is not material to the misinformation claim.
  • relevant: The image text is relevant, or contains, the misinformation claim.
  • irrelevant: The image text is incidental to the misinformation claim.
  • indeterminate_foreign: The image text relevance to the misinformation claim cannot be determined because it is in a foreign language and the fact check does not provide a sufficient translation.
  • indeterminate: The image text relevance could not be determined by the rater.
• content_manipulation: The content of the primary image (or at least one of the images presented in the case of "non_primary") has been manipulated, or synthesized entirely, in a way that is relevant to the misinformation claim.
• text_manipulation: Indicates text occurring on an object in the primary image (or at least one of the images presented in the case of "non_primary") has been added or changed in a way that is not intended to appear as overlaid digital text in a way that is relevant to the misinformation claim.
• chyron_manipulation: Indicates a graphical element that appears as a lower-third graphic in a news broadcast has been added to or manipulated in the primary image (or at least one of the images presented in the case of "non_primary") in a way that is relevant to the misinformation claim.
• ai_manipulation: Indicates the primary image (or at least one of the images presented in the case of "non_primary") has been modified or synthesized using AI in a way that is relevant to the misinformation claim.
• premanipulation_image: Indicates that the fact check provides a version of the content manipulated image prior to content manipulation. This field applies exclusively to the primary image.
• premanipulation_image_id: The ID of the premanipulation image, if provided by the fact check. Image ID is computed as a hash of the image bytes.
• context_manipulation: The misinformation claim ascribes a false or misleading context to the primary image (or at least one of the images presented in the case of "non_primary").
• self_miscontextualizing: Indicates that the false or misleading context is imparted onto the image by text present within the image itself. This field applies exclusively to the primary image.
• circumstance_manipulation: Indicates that the circumstances or events in the primary image (or at least one of the images presented in the case of "non_primary") are described misleadingly.
• identity_manipulation: Indicates that a person, group or object in the primary image (or at least one of the images presented in the case of "non_primary") is misidentified.
• location_manipulation: Indicates that the location depicted in the primary image (or at least one of the images presented in the case of "non_primary") has been misdescribed.
• datetime_manipulation: Indicates that the date the primary image (or at least one of the images presented in the case of "non_primary") was captured is misstated.
• atypical_manipulation: The context ascribed to the primary image (or at least one of the images presented in the case of "non_primary") has been manipulated in a way that is not well captured by the above categories, for instance by omission or incorrectly describing the originating party of the image.
• reverse_image_search: Indicates if a reverse image search was performed by the fact checker and its outcome, if known. This is a categorical field, with values:
  • successful: A reverse image search was performed and successfully identified evidence that was helpful for the investigation of the misinformation claim.
  • unsuccessful: A reverse image search was performed, but it did not successfully identify evidence that was helpful for the investigation of the misinformation claim.
  • performed: A reverse image search was performed, but the outcome is not determinable by the rater.
  • not_performed: A reverse image search is not mentioned in the body of the fact check.
• photograph_of_screen: Indicates that the image contains a screen or other digital display, which is itself displaying content relevant to the misinformation.
• fake_official_document: Indicates if the image is or contains a "fake official document," that is, it is made to appear as though a communication from an official or widely-recognized organization, government, or corporation.