Lesson 20: Can the crew post this image as true? Educator guide

The flood nobody can confirm

The story: An image with no source

It is Wednesday at 3:10 p.m. For twenty minutes, an image has been going around the group chat. It looks like a photo of the campus courtyard, with water over the paving and the benches. The caption says, “Courtyard flooded today, stay away.” It showed up at 2:48 p.m. with no link, no date and no name.

Jordy is putting together the crew’s recap, which goes out at six. Sami zooms in on a hand in the picture. The fingers seem to merge at the knuckles. Someone runs the image through a free online detector. It says “12% likely AI-generated.” Someone else says their roommate’s friend saw water near the library.

Jordy checks the campus facilities alerts page. It shows no active alerts, but its last update was Monday morning.

This is the last case in the casebook, and it uses almost every skill you have learned. Your role is to write the verification note Jordy posts at six.

Where did it come from?

In Lesson 1, a real clip in a group chat seemed to cancel a screening. You learned to find the original source. That is the first move here too. Provenance is information about where an image or a document came from and how it has been handled. This image has no link, no date and no named poster, so its provenance is unknown.

The caption makes a claim, a statement that something is true that you can check with evidence. The hand is a clue, not proof. The detector score is a clue, not proof. As Sami asks, “Twelve percent of what?” And a roommate’s friend is not a source. Near the library is not the courtyard, either.

Corroboration is support for a claim from an independent source. A campus alert would count. So would someone’s dated photo of the courtyard. In Lesson 9, you learned to check an official source. The alerts page is official, but it has not been updated since Monday, so it cannot prove there was no flood.

Should Jordy post the image with a question mark? No. “A question mark still spreads the picture,” Jordy says. Instead, Jordy writes a verification note, a short record of what you checked, what you found and what is still unknown. It works like the holding statement from Lesson 1: it says what is known, what is being checked and when people will hear more. If better evidence appears, the crew will post an update. In Lesson 3, you learned that a correction is a public statement that replaces wrong information with accurate information. If the note turns out to be wrong, the crew will post one where people saw it.

Real fake images

After Hurricane Helene in October 2024, an image of a crying girl holding a puppy in a rescue boat spread fast. Fact-checkers found no original photographer or news source. A digital-forensics expert pointed to details that did not hold together (UC Berkeley School of Information, October 4, 2024).

Key terms in this passage

Provenance

Provenance is information about where an image or a document came from and how it has been handled.

Corroboration

Corroboration is support for a claim from an independent source.

Verification note

A verification note is a short record of what you checked, what you found and what is still unknown.

Claim

A claim is a statement that something is true. You can check a claim with evidence.

Holding statement

A holding statement is a short update that says what is known, what is still being checked and when people will hear more.

Correction

A correction is a public statement that replaces wrong information with accurate information.

Real sources cited

• UC Berkeley School of Information, “Hany Farid breaks down the viral Helene image,” October 4, 2024. Open the source (opens in a new tab)

Checked, found, still unknown

The story: An image with no when or where

At 2:48 p.m. on a Wednesday, an image lands in the group chat with no link, no date and no name attached. It looks like a photo of the campus courtyard, with water covering the paving and the benches. The caption is urgent: “Courtyard flooded today, stay away.” By 3:10, it has been circulating for twenty minutes.

Jordy has a deadline. The crew’s recap goes out at six, and the first question in the chat is whether the courtyard picture goes in. The evidence piles up quickly, and none of it settles anything. Sami, who studies design and photographs things most people walk past, zooms in on a hand in the foreground. The fingers seem to merge at the knuckles. Someone runs the image through a free detector and gets “12% likely AI-generated.” A reply says the writer’s roommate’s friend saw water near the library. Jordy checks the campus facilities alerts page. It lists no active alerts, but it has not been updated since Monday at 9:15 a.m.

Sami redirects the conversation. Everyone is zooming in on the hand, and she did too. But she still does not know who posted the image. When she takes a photo, it has a when and a where. This one has neither.

Your role is to write the verification note Jordy posts at six. It is the final skill in the casebook, and it combines many of the others.

From clues to a verification note

Start where Lesson 1 started, with the original source. Provenance is information about where an image or a document came from and how it has been handled. This image arrived without any, so its provenance is unknown.

The caption makes a claim, the Lesson 1 term for a statement that something is true that you can check with evidence. Look at each clue honestly. The hand is a reason to look closer, not proof of anything. A natural-looking hand would not prove an image is real, either. The detector score is a number without context. Sami asks, twelve percent of what? These tools can be wrong in both directions, and none of them can say whether the courtyard flooded today. The secondhand report fails as a source twice. It is a roommate’s friend, and it is about the library, not the courtyard. Jordy would welcome it if that person had a photo with a time on it.

Corroboration is support for a claim from an independent source. The best candidates are an official alert or a dated photo from someone on site. Lesson 9 taught you to check the official source for your area, and the alerts page is that source. Its silence is weak evidence, because it has not been updated since Monday.

Jordy considers posting the image with a question mark and rejects it, because a question mark still spreads the picture. Instead, the recap will carry a verification note, a short record of what you checked, what you found and what is still unknown. It shares its logic with Lesson 1’s holding statement, which says what is known, what is still being checked and when people will hear more. Lesson 3 adds the last piece. If new evidence changes the answer, a correction goes where people saw the original note.

How real newsrooms handle it

After Hurricane Helene in October 2024, an image of a crying girl clutching a puppy in a rescue boat spread widely. Fact-checkers found no original photographer or news source. Digital-forensics researcher Hany Farid pointed to details that did not hold together, like a boat floor with impossible perspective (UC Berkeley School of Information, October 4, 2024). In 2023, fake images of an explosion near the Pentagon spread quickly and briefly moved markets before officials confirmed nothing had happened (NPR, May 22, 2023).

Looking closely is not enough. Research found that people could not reliably tell AI-generated faces from real ones (Nightingale and Farid, PNAS, February 14, 2022).

Key terms in this passage

Provenance

Provenance is information about where an image or a document came from and how it has been handled.

Corroboration

Corroboration is support for a claim from an independent source.

Verification note

A verification note is a short record of what you checked, what you found and what is still unknown.

Claim

A claim is a statement that something is true. You can check a claim with evidence.

Holding statement

A holding statement is a short update that says what is known, what is still being checked and when people will hear more.

Correction

A correction is a public statement that replaces wrong information with accurate information.

Real sources cited

• UC Berkeley School of Information, “Hany Farid breaks down the viral Helene image,” October 4, 2024. Open the source (opens in a new tab)
• NPR, “Fake viral images of an explosion at the Pentagon were probably created by AI,” May 22, 2023. Open the source (opens in a new tab)
• PNAS (Nightingale & Farid), “AI-synthesized faces are indistinguishable from real faces and more trustworthy,” February 14, 2022. Open the source (opens in a new tab)

When the honest answer is unverified

The story: A deadline and a picture

The crew’s casebook began with a seven-second clip in a group chat. It ends with a single image. At 2:48 p.m. on a Wednesday, a photo-like picture of the campus courtyard appears in the chat, showing water over the paving and benches. Its caption reads, “Courtyard flooded today, stay away.” It arrives with no link, no date on the image and no original poster named. A person in the foreground holds up a phone.

Jordy, who edits the crew’s films and checks every cut before calling it finished, is assembling the six o’clock recap. Within twenty minutes, four pieces of evidence have arrived, and each points somewhere different. Sami posts a zoomed screenshot at 3:02 showing a foreground hand whose fingers seem to merge at the knuckles. A free online detector reports “12% likely AI-generated.” A reply relays that the writer’s roommate’s friend saw water near the library. And at 3:10, Jordy checks the campus facilities alerts page: “No active alerts. Last update: Monday, 9:15 a.m.”

Sami, a design student who notices details for a living, is the first to step back from them. The hand is weird, she agrees, but the more important question is who posted the image first, and when. Jordy states the dilemma plainly. They wanted a clean yes or no, and they do not have one.

Your role is to write the verification note Jordy posts at six.

Weighing evidence without overclaiming

Provenance is information about where an image or a document came from and how it has been handled. Checking provenance first is the Lesson 1 skill, finding the original source, applied to images. This image has none, so every other clue has to be weighed without an anchor.

Each clue has a specific weakness. The caption is a claim, a statement that something is true that can be checked with evidence. The hand is an anomaly, which justifies closer scrutiny and proves nothing, since generated images can look natural and real photos can look strange. The detector score is uninterpretable without knowing the tool’s error rates. Sami’s question, twelve percent of what, gets at that. It also answers a different question from the one that matters. Even a perfect detector could not say whether the courtyard flooded today. The secondhand report fails on two counts: it is not firsthand, and it concerns the library, not the courtyard.

Corroboration is support for a claim from an independent source. Lesson 9’s skill, checking the official source for your area, points to the alerts page, but its last update on Monday limits what its silence can mean. A dated photo from someone at the courtyard would be stronger.

The new skill is confirming an image with independent sources, and publishing only what was checked. Jordy rejects the middle path of posting the image with a question mark, because the image travels regardless of the punctuation. The recap instead carries a verification note, a short record of what you checked, what you found and what is still unknown. It is a descendant of Lesson 1’s holding statement, which says what is known, what is still being checked and when people will hear more. It also sets up Lesson 3’s obligation. If new evidence arrives, a correction or update appears where the note did. “Unverified” is sometimes the most accurate conclusion available.

Detection, provenance and professional standards

NIST’s overview of synthetic content describes provenance data, labeling, watermarking and detection, and it explains the technical limits of each (NIST, November 20, 2024). Real incidents show why no single signal suffices. The Hurricane Helene image of a crying girl with a puppy had no traceable photographer or outlet, and forensics researcher Hany Farid identified internal inconsistencies (UC Berkeley School of Information, October 4, 2024). The News Literacy Project’s RumorGuard check on that image models a verification note, setting out what is known, what is not and why (RumorGuard, no date listed). In 2023, fake images of a Pentagon explosion were shared by verified accounts, showing that a trusted-looking sharer is not provenance (CNN, May 22, 2023).

Human perception is not a reliable safeguard. In a PNAS study, people could not reliably distinguish AI-synthesized faces from real ones (Nightingale and Farid, February 14, 2022).

Journalism codes frame Jordy’s choice as a duty. The RTDNA code says journalism “verifies, provides relevant context, tells the rest of the story and acknowledges the absence of important additional information” (RTDNA Code of Ethics, June 11, 2015). The SPJ code asks journalists to “identify sources clearly” (SPJ Code of Ethics, September 6, 2014).

Key terms in this passage

Provenance

Provenance is information about where an image or a document came from and how it has been handled.

Corroboration

Corroboration is support for a claim from an independent source.

Verification note

A verification note is a short record of what you checked, what you found and what is still unknown.

Claim

A claim is a statement that something is true. You can check a claim with evidence.

Holding statement

A holding statement is a short update that says what is known, what is still being checked and when people will hear more.

Correction

A correction is a public statement that replaces wrong information with accurate information.

Real sources cited

• National Institute of Standards and Technology (NIST), “Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency,” November 20, 2024. Open the source (opens in a new tab)
• UC Berkeley School of Information, “Hany Farid breaks down the viral Helene image,” October 4, 2024. Open the source (opens in a new tab)
• RumorGuard (News Literacy Project), “Widely shared image of crying girl and puppy after Helene is AI generated,” No date listed. Open the source (opens in a new tab)
• CNN, “‘Verified’ Twitter accounts share fake image of ‘explosion’ near Pentagon, causing confusion,” May 22, 2023. Open the source (opens in a new tab)
• PNAS (Nightingale & Farid), “AI-synthesized faces are indistinguishable from real faces and more trustworthy,” February 14, 2022. Open the source (opens in a new tab)
• Radio Television Digital News Association (RTDNA), “RTDNA Code of Ethics,” 2015. Open the source (opens in a new tab)
• Society of Professional Journalists (SPJ), “SPJ Code of Ethics,” 2014. Open the source (opens in a new tab)

Verification under deadline

The story: Four clues and no source

The casebook closes where it opened, with a piece of media in a group chat and a group deciding what to tell people. At 2:48 p.m. on a Wednesday, an image of the campus courtyard under water appears with the caption “Courtyard flooded today, stay away.” It carries no link, no date and no original poster. Jordy has until six to decide what the crew’s recap will say.

By 3:10, four signals have accumulated. Sami’s zoom shows a hand whose fingers seem to merge at the knuckles. A free detector reports “12% likely AI-generated.” A secondhand message says someone’s roommate’s friend saw water near the library. The campus facilities alerts page shows no active alerts, with a last update on Monday at 9:15 a.m.

The case is designed so that none of these signals resolves the question, and several tempt overinterpretation. Sami models the necessary shift. She acknowledges the anomaly she found and then sets it aside in favor of provenance. Who posted the image first, and when? Jordy models the editorial decision. Wanting a clean answer is not a reason to publish one. The recap will say what was checked and leave the image out.

Your role is to write the verification note Jordy posts at six.

An evidentiary standard for images

Provenance is information about where an image or a document came from and how it has been handled. For images that claim to show a news event, provenance is the primary question, and it applies the Lesson 1 skill of finding the original source. Its absence here shifts the burden to corroboration.

The caption asserts a claim, a statement that something is true that can be checked with evidence. The available signals differ in kind. The hand anomaly is diagnostic of nothing on its own, since generated images can look natural and genuine photos can look odd. The detector’s output lacks the error rates that would make it interpretable, and it addresses whether the image is synthetic rather than whether the event occurred. The secondhand report is attenuated in source and mismatched in location.

Corroboration is support for a claim from an independent source. The official source, Lesson 9’s habit, is the alerts page, and its staleness means its silence is weak evidence. An independent, dated image or a same-day official notice would be probative.

The case’s new skill is confirming an image with independent sources and publishing only what was verified. The chosen output is a verification note, a short record of what you checked, what you found and what is still unknown. It is the structural descendant of Lesson 1’s holding statement, which says what is known, what is still being checked and when people will hear more. It commits the crew to Lesson 3’s duty of correction, a public statement that replaces wrong information with accurate information, if later evidence changes the picture. Jordy’s refusal to post the image “with a question mark” reflects a practical insight. Hedging language does not travel with an image, and the image is what spreads.

Technical limits and professional norms

NIST’s overview of synthetic-content transparency surveys provenance metadata, labeling, watermarking and detection, and it emphasizes that each has technical limits (NIST, November 20, 2024). Perception offers no backstop. Nightingale and Farid found that people could not reliably distinguish AI-synthesized faces from real ones and rated synthetic faces as more trustworthy (PNAS, February 14, 2022).

Recent cases show the costs of skipping provenance. The Hurricane Helene “girl and puppy” image had no identifiable photographer or outlet, and forensic analysis found inconsistencies such as a boat floor with impossible perspective (UC Berkeley School of Information, October 4, 2024). Local reporting documented how emotion outran verification (KPRC Click2Houston, October 9, 2024). In 2023, a fabricated image of an explosion near the Pentagon spread through verified accounts and briefly moved markets (NPR, May 22, 2023; CNN, May 22, 2023). Critics now warn that AI content is spreading confusion across feeds (PBS NewsHour, October 22, 2025).

Professional codes treat verification and transparency about uncertainty as obligations. RTDNA says journalism “acknowledges the absence of important additional information” (RTDNA Code of Ethics, June 11, 2015). SPJ holds that the public “is entitled to as much information as possible to judge the reliability and motivations of sources” (SPJ Code of Ethics, September 6, 2014). PRSA asks practitioners to “investigate the truthfulness and accuracy of information released on behalf of those represented” (PRSA Code of Ethics, October 2000). A verification note meets all three standards, even when its conclusion is “we cannot confirm this yet.”

Key terms in this passage

Provenance

Provenance is information about where an image or a document came from and how it has been handled.

Corroboration

Corroboration is support for a claim from an independent source.

Verification note

A verification note is a short record of what you checked, what you found and what is still unknown.

Claim

A claim is a statement that something is true. You can check a claim with evidence.

Holding statement

A holding statement is a short update that says what is known, what is still being checked and when people will hear more.

Correction

A correction is a public statement that replaces wrong information with accurate information.

Real sources cited

• National Institute of Standards and Technology (NIST), “Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency,” November 20, 2024. Open the source (opens in a new tab)
• PNAS (Nightingale & Farid), “AI-synthesized faces are indistinguishable from real faces and more trustworthy,” February 14, 2022. Open the source (opens in a new tab)
• UC Berkeley School of Information, “Hany Farid breaks down the viral Helene image,” October 4, 2024. Open the source (opens in a new tab)
• KPRC Click2Houston, “Houstonians learn the cons of a viral AI-generated photo after Hurricane Helene,” October 9, 2024. Open the source (opens in a new tab)
• NPR, “Fake viral images of an explosion at the Pentagon were probably created by AI,” May 22, 2023. Open the source (opens in a new tab)
• CNN, “‘Verified’ Twitter accounts share fake image of ‘explosion’ near Pentagon, causing confusion,” May 22, 2023. Open the source (opens in a new tab)
• PBS NewsHour, “AI content spreads confusion and misinformation, critics say,” October 22, 2025. Open the source (opens in a new tab)
• Radio Television Digital News Association (RTDNA), “RTDNA Code of Ethics,” 2015. Open the source (opens in a new tab)
• Society of Professional Journalists (SPJ), “SPJ Code of Ethics,” 2014. Open the source (opens in a new tab)
• Public Relations Society of America (PRSA), “PRSA Code of Ethics,” 2000. Open the source (opens in a new tab)