Photographs feel true in a way that words do not. When we read a claim, we instinctively ask ourselves whether it is accurate. When we see an image, something in our cognition says: this happened, this is real, the camera was there. For most of photography's history, that instinct was largely justified — a photograph required a camera, a photographer, and a real subject standing in front of a lens. That relationship has been transformed.

Images and videos can now be cropped, edited, captioned falsely, generated entirely by artificial intelligence, or synthesized into convincing video of events that never happened. The photograph that once said I was there may now say I was constructed or I was taken from somewhere else entirely. This chapter teaches you how to evaluate visual material with the same disciplined skepticism you already apply to written claims.

You will learn how context and captions shape what an image appears to document, how cropping and framing change meaning without altering a single pixel, how editing ranges from legitimate correction to deceptive manipulation, how to conduct reverse image and video searches, how AI-generated visuals and deepfakes differ from authentic recordings, and how to assemble all of these techniques into a complete visual verification workflow.

This chapter teaches students how to evaluate visual evidence critically and responsibly. Students will learn why images and videos can mislead even careful viewers, how context and captions shape visual meaning, how cropping and framing alter what a visual appears to show, how editing and digital manipulation are used to deceive, how to conduct reverse image and video searches, how AI-generated visuals and deepfakes differ from authentic recordings, and how to build a reliable visual verification workflow. By the end of this chapter, students should approach every image and video they encounter in research with the same disciplined skepticism they apply to written claims.

Visual evidence is everywhere in modern information. News articles include photographs. Social media posts circulate images and video clips as evidence of claims. Political arguments are illustrated with visuals. Scientific findings are accompanied by charts, microscope images, and data visualizations. Historical arguments rest on archival photographs and film footage. If you cannot evaluate visual evidence, you cannot fully evaluate the arguments that rest on it.

The skills in this chapter are among the most practically important in this entire textbook, and among the skills most people have never been formally taught. The central challenge of visual evidence is not that images always lie — most images are exactly what they appear to be. The challenge is that you cannot determine which images are authentic, which have been altered, and which have been taken out of context simply by looking carefully. The lie, when it exists, is usually invisible to the naked eye. Verification requires method, not just attention.

Understanding why visual evidence can mislead is the essential first step. Before learning verification techniques, you need to understand the specific ways in which images and videos fail as evidence — not because they are always fraudulent, but because they are always incomplete and always interpreted.

Every photograph and every video frame shows only a portion of what was present at a scene. The camera captures what falls within the lens's field of view during the fraction of a second the shutter opens. Everything outside the frame — to the left, to the right, above, below, and in the moments before and after — is invisible to the viewer.

This incompleteness is not a defect. It is the nature of the medium. But it means that a photograph is always a selection, not a complete record. A photograph of a crowd may look enormous or sparse depending on the angle. A photograph of a protest may make it appear peaceful or violent depending on which moment and which participants are captured. None of these photographs necessarily falsifies anything — but all of them reflect choices a careful viewer should recognize.

A photograph without context tells you relatively little. You see what appears to be a scene, but you do not know where it was taken, when it was taken, who the people in it are, or what was happening before and after the shutter clicked. Context — the information surrounding the image — is what allows the photograph to function as evidence of a specific claim.

An authentic photograph, one that has not been digitally altered in any way, can nevertheless be used as false evidence if it is presented with a false or misleading context. A real photograph from one country, one time period, or one event is presented as documentation of something that happened in a different country, time period, or event. The image is genuine. The claim the image is being used to support is not.

Human visual memory is imprecise. We remember general impressions of images rather than specific details. This means that when a familiar image is slightly altered — a face swapped, a sign changed, a crowd size adjusted — we may not notice the difference, especially if we encountered the original casually or some time ago. People also tend to focus on central or emotionally salient elements of an image and may not notice alterations at the edges or in the background, where skilled manipulators often make their changes.

Powerful images trigger emotional responses faster than written descriptions of the same events. This is one reason images are such effective communication tools, and also one reason they are effective misinformation tools. When an image triggers a strong emotional response — outrage, grief, fear, pride — that response can interfere with critical evaluation. We feel an image must be real because of how it makes us feel, even though emotions are not evidence of anything except our own reaction.

The caption is the most powerful element of any news photograph. Research on how people read news images consistently finds that captions shape interpretation more decisively than the image itself. An image without a caption is ambiguous; a caption resolves that ambiguity by telling the viewer who, what, where, and when. Once a viewer has read a caption, they tend to see the image through the frame the caption provides. Changing the caption — without changing the image at all — changes what the image appears to document.

A responsible caption for a news photograph answers the basic questions of journalism: who is depicted, what is happening, where the photograph was taken, and when. It also identifies the photographer and notes any significant limitations of the image, such as whether the photograph is drawn from a file archive rather than a current event.

Misleading captions may also identify people incorrectly, misidentify the nature of an event (describing a staged demonstration as spontaneous, or vice versa), omit context that would change the viewer's interpretation, or present file photographs as current without disclosure.

When you encounter an image with a caption that you plan to use as evidence in research, verify the caption independently. Ask:

Cropping removes portions of an image. Framing — the photographer's original choice of what to include in the lens — shapes the image before any cropping occurs. Both can significantly alter what an image appears to show, even though cropping is standard practice in photography, used to remove distracting elements, improve composition, and fit images into layouts. The real question is always whether a crop preserves or distorts the meaning of the original scene.

Image and video editing ranges from adjustments so minor they are nearly invisible to alterations so significant that the resulting image no longer resembles the original scene. The general rule in photojournalism ethics is that editing may correct technical problems introduced by the camera or lighting, but may not add, remove, or alter substantive elements of the scene.

Video introduces additional possibilities for deception: selective editing that removes qualifying context from a speech; looping that makes a brief incident look like a sustained pattern; speed manipulation that makes movement appear unsteady or frenzied; out-of-sequence assembly that implies unrelated clips are connected; and audio manipulation that changes what a speaker appears to have said or alters the emotional tone of a scene.

Reverse image searching allows you to find other instances of an image on the web by submitting the image itself — rather than a text description — as your search query. It is one of the most powerful verification tools available to researchers: it can find an image's original source, surface earlier uses that predate its current context, reveal different captions attached to the same image, and locate higher-resolution versions.

When interpreting results, ask: What is the earliest date this image appears? What contexts and captions has it appeared with before? What source first published it? How does the current version compare to earlier versions in cropping or alteration? Are higher-resolution versions available that reveal more detail?

Reverse video search is less developed, but you can extract still frames at key moments and reverse-search those frames as images. The InVID / WeVerify browser plugin is built specifically for this purpose, extracting keyframes and metadata from videos on platforms such as YouTube. Checking platform metadata (upload date, view counts) and searching distinctive audio or speech content can also help establish a video's true origin.

Artificial intelligence systems can generate photorealistic images of scenes, people, and events that do not exist. These text-to-image systems are trained on enormous datasets of real photographs and learn statistical patterns that let them assemble new, internally consistent images — realistic lighting, plausible anatomy, natural textures — without any camera ever capturing a real scene.

This presents a fundamentally different verification challenge than manipulated photographs. With manipulation, there is an original to compare against, and inconsistencies between original and altered versions can be detected. With AI generation, there is no original. The image was not altered; it was generated as a complete whole, which is why classic manipulation-detection techniques (edge artifacts, inconsistent compression) are less reliable here.

Be especially alert to possible AI generation when an image depicts a dramatic, news-worthy scene not covered by reliable outlets; comes from an account with no established credibility; shows a public figure in an implausible context; cannot be found through reverse image search; or appears alongside a viral claim with no attribution to original news coverage.

Deepfakes are AI-generated or AI-manipulated videos in which a person's face, voice, or body is synthesized or replaced. The term combines "deep learning," the AI technique used, with "fake." At their most sophisticated, deepfakes are nearly indistinguishable from authentic video to untrained viewers at normal speed — the face moves naturally, expressions are consistent, and separately synthesized audio matches lip movement.

Deepfakes have been used to create false statements attributed to public figures, produce non-consensual intimate imagery, generate political propaganda, and fabricate evidence in disputes and legal proceedings. For researchers, deepfakes are a concern both directly — mistaking a deepfake for authentic video — and systemically, because awareness of the technology lets bad actors discredit genuine footage by simply claiming it is fake.

The previous sections describe specific forms of visual misleading and specific verification techniques. This section synthesizes those techniques into a comprehensive workflow you can apply to any image or video you encounter in research.

The MGU Misinformation Check applies directly to images and video. For any visual you are evaluating, ask each question in turn:

Scenario 1. Find a dramatic or emotionally compelling news photograph from a recent story published by a major international wire service (Reuters, AP, AFP, or comparable). Submit the image to Google Images, TinEye, and at least one additional reverse image search tool. Record: (1) What results did each tool return? (2) What is the earliest date the image appears in any result? (3) Does the image appear with different captions in any results? (4) What is the original source, based on your search? (5) Does the reverse search support or complicate the caption in the story you found?

Scenario 2. Find an image circulating on social media presented as documentation of a current event. Submit it to reverse image search tools and record: (1) Does the image have a prior history, and if so, what is the earliest date? (2) What contexts has it appeared in before? (3) Does the post's caption match those prior contexts? (4) What is your assessment of the image's evidentiary value for the claim the post makes?

Visit a publicly available, academic or non-profit AI image generation demonstration and generate three images from text descriptions. Examine each for the indicators in Section 30.6 (hands, teeth, eyes, backgrounds, edges, lighting). Submit each to an AI detection tool and record the result. Submit each to a reverse image search tool and record what you find. Write a brief assessment of each image's likely detection status, and note what would make it harder or easier to identify as AI-generated. Use only publicly available academic or demonstration interfaces, and do not generate images of real people or harmful content.

Practice Source Evaluation: Visual Evidence in a Research Context. You are researching the use of facial recognition technology by law enforcement agencies in major urban areas. You find this image and accompanying text in a news article:

Evaluation questions:

Model Entry: Visual Evidence Verification Session

Visual sources require specific citation elements: creator or photographer (if known), title or description, date, publisher or agency, and access information. Date is especially critical for visual evidence because date displacement is a major form of visual misinformation.

Several credible external organizations also specialize in visual verification methodology and are useful both as resources and as models for the verification process itself: First Draft (firstdraftnews.org), Bellingcat (bellingcat.com), Snopes (snopes.com), PolitiFact (politifact.com), and AFP Fact Check (factcheck.afp.com).