Source Verification

Source Verification

What This Skill Does

Walks through structured verification of sources, claims, images, video, and documents across five verification modes — then produces a verification-trail template the reporter can file with the story or submit to legal review.

When To Use This Skill

• A source has sent you a photo, video, or document and you need to confirm it is what they say it is before publishing
• You are fact-checking a claim that rests on visual or documentary evidence
• You want to verify the origin, date, or location of user-generated content found on social media
• An anonymous source has provided information and you need to assess credibility without revealing their identity
• Someone is presented as an expert and you need to verify their credentials, affiliations, and track record before quoting them
• You received a document (memo, contract, letter, report) and need to determine whether it is authentic
• You need a paper trail showing your newsroom's verification steps for legal or editorial review

What You Need To Provide

For visual media verification (images, video)

Required: Description of the image or video; the source's stated provenance (who sent it, when, what they say it shows) Optional: The file itself or a URL where it was found; any metadata already extracted; the publication deadline; whether the material will be used in a legal or regulatory context

For document verification

Required: Description of the document (type, date, issuing body); the source's stated provenance; why you believe it may need verification (unsolicited, unexpected content, anonymous origin) Optional: The document file; known reference documents from the same issuing body; any prior correspondence with the source

For anonymous source verification

Required: The claim the source is making; what the source says about their access to the information (position, proximity, how they obtained it); any corroborating material they have provided Optional: The channel through which the source made contact; any pattern of prior contact; whether the source has asked for specific protections (legal, physical safety)

For expert source verification

Required: The person's name and stated credentials; the specific claim or opinion they are offering; the field in which they claim expertise Optional: Their institutional affiliation; publications or media appearances; who recommended them as a source

For social media content verification

Required: The post or content to verify (screenshot, URL, or description); the platform; the account that posted it; what the content claims to show or prove Optional: The date the content first appeared; any replies, quote-posts, or threads that add context; whether the account has been suspended or the content removed since you found it

How the Assistant Approaches This

The verification method depends on the mode. All five modes follow the SIFT framework (Stop, Investigate the source, Find the original, Trace the claim) as their backbone, then add mode-specific forensic steps.

Mode 1: Visual Media Verification (Images and Video)

1. Stop. Restates the claim and the source's provenance in neutral language. Identifies what specifically must be true for the material to support the claim — location, date, subject, event.

2. Investigate the source. Checks who provided the material. Prior publication history, organizational affiliation, stated expertise, any known track record of accuracy or fabrication. Flags if the source is anonymous or if provenance is secondhand ("a friend sent me this").

3. Find the original. Guides the reporter through reverse-image search (Google Images, TinEye, Yandex) or video keyframe extraction and search to locate the earliest known version. Notes whether the source's version matches the original or has been cropped, compressed, re-encoded, or altered. For video, recommends extracting keyframes at 1-second intervals and running each through reverse-image search independently.

4. Trace the claim. Maps the claim's spread: where else has it appeared, who published it first, how did it reach the source? Flags circular sourcing — multiple outlets citing each other with no independent origin.

5. Inspect metadata. Walks through EXIF data extraction for images or container metadata for video (see Metadata Fields Reference below). Flags inconsistencies between metadata and stated provenance.

6. Check for manipulation. Runs through the Visual Manipulation Indicators checklist (see reference table below). Notes C2PA Content Credentials if present — checks the credential chain, signing authority, and whether the manifest has been stripped or modified.

7. Assess geolocation. If the material claims to show a specific location, guides the reporter through geolocation cross-checks: visible landmarks, signage (language, style), vegetation and terrain, sun angle relative to claimed time and latitude, shadow direction, weather conditions compared to historical weather data for the claimed date and location.

8. Produce the verification trail. Fills in the structured template documenting every step taken, what was found, what remains unverified, and the reporter's confidence level.

Mode 2: Document Verification

1. Stop. Restates what the document purports to be, who provided it, and what the reporter plans to use it for. Identifies what must be true: the document was created by the stated author or institution, on or near the stated date, and has not been altered since creation.

2. Investigate the source. Determines how the document reached the reporter. Unsolicited documents from unknown sources carry higher risk than documents obtained through official channels. Checks whether the source has provided authentic documents before.

3. Check document metadata. Guides the reporter through extracting document properties:

   • PDF files: Author field, creation date, modification date, producing application (Adobe Acrobat, Microsoft Word, LaTeX, etc.), PDF version, whether the file has been linearized (web-optimized — unusual for internal memos)
   • Word documents (.docx): Author, last modified by, revision count, total editing time, company field, template name
   • Scanned documents: Resolution, scanner software, whether OCR has been applied, whether the scan is a scan of a photocopy (generational loss indicates the original may have passed through multiple hands)

4. Compare against known exemplars. If the document claims to come from a specific institution, compare against authentic documents from the same source:

   • Letterhead: Logo placement, resolution, colour values, font choices. Forged letterhead often uses slightly wrong proportions, outdated logos, or consumer-grade fonts in place of custom typefaces
   • Formatting: Margin widths, line spacing, paragraph indentation, page numbering style, header/footer content. Institutional documents follow house style — deviations suggest the document was created outside the institution
   • Language and register: Vocabulary, abbreviations, internal jargon, reference-number formats. Every institution has phrasing habits. A document that uses the wrong internal terminology for the claimed issuing office is suspect
   • Signatures: Compare ink density, pen pressure patterns, character spacing. Digital signatures should be verifiable through the issuing certificate authority

5. Check internal consistency. Look for:

   • Dates that fall on weekends or holidays (for documents that should have been created on business days)
   • Reference numbers that do not match the issuing body's known numbering scheme
   • Names or titles that were not in use at the claimed date (a person who held the title "Deputy Director" in 2024 may have been "Assistant Director" in 2022)
   • Legal citations, statutory references, or regulatory codes that did not exist at the claimed date

6. Produce the verification trail.

Mode 3: Anonymous Source Verification

Anonymous source verification does not attempt to identify the source. It assesses whether the source's claims are credible, consistent, and independently corroborable — without requiring the source to reveal who they are.

1. Stop. Restates the claim the anonymous source is making. Identifies what must be true for the claim to hold, and what evidence — independent of the source — could confirm or refute it.

2. Assess claimed access. The source says they have access to the information. Evaluate whether the claimed position, role, or proximity is plausible:

   • Does the type of information match what someone in the claimed position would actually see?
   • Is the level of detail consistent with direct access (firsthand knowledge) or indirect access (heard it from someone, saw a summary)?
   • Does the source describe processes, systems, or internal procedures accurately? Someone who has actually worked inside an organization will reference specifics — file naming conventions, approval chains, internal software — that outsiders cannot easily fake

3. Test internal consistency. Ask the source the same question in different ways across multiple conversations. Fabricated accounts tend to drift in details: dates shift, numbers change, the sequence of events varies. Authentic accounts stay consistent on facts even when the phrasing changes. Note: consistent does not mean rehearsed. A source who repeats the same sentences verbatim may be reciting a script.

4. Seek independent corroboration. Identify at least one element of the claim that can be verified without the source:

   • Public records that would confirm the underlying event (filings, permits, contracts, court records)
   • Physical evidence (the building exists, the meeting happened, the document is filed where the source said it would be)
   • Other sources who can confirm elements of the claim independently (they do not need to confirm the source's identity — only that the event or condition described is real)
   • Circumstantial corroboration (the timing fits with known events, the claim explains something that was previously unexplained)

5. Evaluate motivation. Why is this source talking? Motivations do not disqualify a source, but they shape how you weight the information:

   • Whistleblower motivation (outrage at wrongdoing): often credible on facts, may overstate severity
   • Grievance motivation (fired, passed over, personal conflict): often accurate on specifics they witnessed, may omit context that casts a more balanced picture
   • Strategic motivation (wants to influence a decision, damage a rival, shape public opinion): facts may be accurate but selectively presented
   • Financial motivation (selling information): verify independently before any payment discussion; financially motivated sources are not inherently less reliable, but the incentive to fabricate is higher

6. Assess risk to the source. If publishing could expose the source to retaliation, determine:

   • How many people would have access to this information? (If only three people knew, publishing any detail may identify the source by elimination)
   • Does the source understand the risk? Have they given informed consent?
   • Can the story be told without details that narrow the pool of possible sources?

7. Produce the verification trail. The trail for anonymous sources must be stored securely and separately from the story file. It should document the verification steps taken but not contain any information that could identify the source if the trail is subpoenaed or leaked.

Mode 4: Expert Source Verification

1. Stop. Restates what the expert is claiming and in what capacity. Identifies the specific field of expertise required and the level of authority the story will attribute to this person.

2. Verify credentials. Check each claimed credential independently:

   • Academic degrees: Confirm the degree, institution, and year through the institution's alumni directory or registrar (not the expert's own CV alone). Diploma mills are common — check the institution against accreditation databases
   • Professional licenses: Verify through the relevant licensing board (state medical boards, bar associations, engineering boards). A license can be current, expired, suspended, or revoked — each status matters
   • Institutional affiliation: Confirm current employment or affiliation through the institution's directory, not just the expert's LinkedIn profile. People who left an institution years ago sometimes still list it
   • Job title: Confirm through the institution. "Senior Research Fellow" and "Adjunct Lecturer" carry different weight. Some people inflate titles

3. Assess publication record. Search for the expert's published work in the claimed field:

   • Peer-reviewed papers in indexed journals (Google Scholar, PubMed, Scopus) — not self-published books or blog posts
   • Conference presentations at recognized professional conferences
   • Citations by other researchers in the field (a high paper count with near-zero citations suggests the work has not been adopted by peers)
   • Whether the expert publishes in the specific subfield relevant to the story, or in a related but different area. A cardiologist is not an epidemiologist. A constitutional law professor is not a criminal defense expert

4. Check for conflicts of interest.

   • Financial ties: consulting fees, board memberships, patent holdings, industry funding. Check disclosure statements on recent papers, financial disclosure databases, and corporate filings
   • Advocacy positions: does the expert serve on the board of an advocacy organization related to the topic? This does not disqualify them but must be disclosed
   • Litigation history: has the expert served as a paid expert witness? In what capacity? Expert witnesses are not neutral — they were hired to support a position. Check court records (PACER for federal cases)
   • Prior public statements: has the expert publicly committed to a position on this topic before? Someone who has staked their reputation on a particular conclusion may not offer a balanced assessment

5. Cross-check the specific claim. Does the expert's opinion on this specific topic align with the established consensus in their field?

   • If yes, the expert is functioning as a representative voice. Standard sourcing applies
   • If the expert is offering a contrarian view, that is not automatically disqualifying — but the story must note it as a minority position and explain the basis for the mainstream view
   • If the expert is speaking outside their specific field (a physicist commenting on biology, an economist commenting on public health), flag this explicitly. Cross-disciplinary commentary is not expertise

6. Produce the verification trail.

Mode 5: Social Media Content Verification

1. Stop. Restates what the social media content claims to show and what the reporter wants to use it for. Identifies the platform, the account, and the date of posting.

2. Verify the account. Assess whether the account is what it claims to be:

   • Account age: When was it created? Accounts created days before a major event and posting "eyewitness" content are suspect
   • Posting history: Does the account have a consistent posting pattern over months or years, or was it dormant until this one post? Check for gaps that suggest a purchased or hijacked account
   • Follower/following ratio: Authentic personal accounts typically have a follower count roughly proportional to their social circle. An account with 12 followers that suddenly posts viral content is not necessarily fake — but it warrants closer scrutiny
   • Bio and self-description: Does it match the account's posting history? An account claiming to be a nurse in Portland whose prior posts are all about cryptocurrency trading in Lagos raises questions
   • Verification status: Platform verification badges confirm identity, not credibility. A verified account can still post false information

3. Archive the content. Before any further verification, archive the post and surrounding context:

   • Screenshot the post with visible timestamp, account name, and engagement metrics
   • Use the Wayback Machine (web.archive.org/save) or Archive.today to create a permanent copy if the content is on a public web page
   • Save the URL, even if the content may be deleted — cached versions and archival services may retain it
   • If the content includes a video, download the original file before the platform compresses or removes it

4. Check for prior circulation. Run reverse-image search on images and keyframes. Search the text of the post (in quotes) across platforms. Check whether the same content appeared elsewhere earlier — recycled content from old events is one of the most common forms of misinformation.

5. Verify claimed location and time.

   • Cross-reference visible details (weather, light conditions, signage, vehicles, clothing) with the claimed location and date
   • Check weather archives for the claimed location and date — if the post shows sunshine but historical records show rain, investigate
   • Look for platform-specific metadata: some platforms embed location data in the original upload (often stripped from screenshots or re-uploads)

6. Assess the context around the post. Who is amplifying it? Early replies and quote-posts often reveal whether the content is being shared organically or pushed by coordinated accounts. Signs of coordination:

   • Multiple accounts posting the same content within minutes of each other
   • Accounts using identical or near-identical wording
   • Amplification by accounts with no prior interest in the topic
   • Sudden spikes in engagement that do not match the account's normal reach

7. Produce the verification trail.

Output Format

For all modes

Two parts. First: a narrative verification report (400–1,000 words depending on complexity) walking through findings in the order described above, written in direct factual language. Second: a filled verification-trail template (structured form) the reporter can attach to the story file.

The verification trail template adjusts its fields to the mode used:

VERIFICATION TRAIL
Mode: [visual media / document / anonymous source / expert source / social media content]
Material type: [image / video / document / claim / social media post]
Source: [who provided it, when, through what channel]
Claimed provenance: [what the source says it shows or means]
---
SIFT steps completed:
  Stop: [claim restated, what must be true]
  Investigate source: [findings]
  Find original: [search methods used, earliest version found, match status]
  Trace claim: [spread map, original publication, circular sourcing flags]
Mode-specific checks:
  [For visual media] Metadata inspection:
    Method: [EXIF / container metadata]
    Key findings: [timestamps, GPS, device, edit history]
    Inconsistencies: [list, or "none detected"]
  [For visual media] Manipulation check:
    Deepfake indicators: [findings, or "none detected"]
    Tampering indicators: [findings, or "none detected"]
    C2PA credentials: [present/absent, chain status, signing authority]
  [For visual media] Geolocation: [cross-checks performed, result]
  [For documents] Metadata: [author, dates, producing application, revision count]
  [For documents] Exemplar comparison: [letterhead, formatting, language — match/mismatch]
  [For documents] Internal consistency: [dates, references, names, titles — findings]
  [For anonymous sources] Access assessment: [claimed access, plausibility]
  [For anonymous sources] Consistency: [tested across conversations — stable/drifting]
  [For anonymous sources] Independent corroboration: [what was corroborated, how]
  [For anonymous sources] Motivation: [assessed type, implications]
  [For expert sources] Credentials verified: [degrees, licenses, affiliation — confirmed/unconfirmed]
  [For expert sources] Publication record: [field match, citation count, peer recognition]
  [For expert sources] Conflicts of interest: [financial ties, advocacy, litigation, prior positions]
  [For expert sources] Claim vs. consensus: [aligned / contrarian / outside field]
  [For social media] Account assessment: [age, history, patterns, verification status]
  [For social media] Archival: [methods used, permanent URLs]
  [For social media] Prior circulation: [found / not found, earliest version]
  [For social media] Location/time verification: [weather, signage, metadata — consistent/inconsistent]
  [For social media] Amplification pattern: [organic / coordinated / inconclusive]
Unverified elements: [what could not be confirmed and why]
Confidence level: [HIGH / MEDIUM / LOW — with one-line rationale]
Verified by: [reporter name]
Date: [verification date]

Verification Tools Reference

The assistant cannot run these tools. It tells the reporter which tools to use, what to look for in the results, and how to interpret what they find.

Red Flags Reference

Quick-reference table of warning signs organized by verification mode. A single red flag does not prove fabrication — but each one demands a specific follow-up action.

Visual Media Red Flags

Document Red Flags

Anonymous Source Red Flags

Expert Source Red Flags

Credibility Indicators — Positive Signals

Not every verification is about finding problems. These indicators support a source's credibility:

Example 1: Visual Media Verification

Input

Material: A photograph showing what appears to be a chemical spill along a riverbank, with dead fish visible on exposed mud. Source says the image was taken yesterday morning at approximately 7:15 a.m. along the Millford River, downstream from an industrial park. Source is a local angler who contacted the newsroom by email, attaching the image and a brief statement. No prior contact with this source. Publication deadline: tomorrow evening edition.

Output

Verification report

The claim: a photograph shows a chemical spill on the Millford River downstream from an industrial park, taken yesterday morning around 7:15 a.m. by a local angler. If true, this is evidence of an unreported pollution event at a specific, identifiable location.

Source check. The sender has no prior relationship with the newsroom. A name search returns a local fishing-club membership list from three years ago — consistent with the angler claim but not proof. No social media accounts found under that name in a quick search. The email address uses a consumer provider, not a burner domain. Nothing flags this source as unreliable, but nothing independently confirms their identity either. Treat as single-source until corroborated.

Finding the original. A reverse-image search on Google Images and TinEye returned no matches. Yandex returned no matches. This is consistent with a newly taken photograph rather than a recycled image. Search terms "Millford River spill" and "Millford River dead fish" returned no news coverage, social media posts, or forum threads as of this morning. The image does not appear to be in circulation.

Tracing the claim. No other outlet or social account has published this image or a similar claim. The source's email is the only known origin point. There is no circular sourcing to flag. The regional environmental agency's incident log (publicly available online) shows no reported spill for the Millford River in the past seven days — which could mean the event is unreported, or that it did not happen.

Metadata inspection. EXIF data extraction (using ExifTool or Jeffrey's EXIF Viewer) should yield:

• Camera model and lens: Check whether the device is consistent with what a casual angler would carry (phone camera vs. professional DSLR). The source did not mention photography equipment.
• Timestamp: Look for the DateTimeOriginal field. If it reads yesterday's date at approximately 07:15, that supports the claimed timeline. If it reads a different date, or if the field is absent (common with screenshots or images saved from messaging apps), flag that.
• GPS coordinates: If present, compare to the Millford River's course downstream from the industrial park. A mismatch of more than a few hundred metres is significant. If GPS data is absent, the location cannot be confirmed through metadata alone.
• Software field: If this field shows editing software (Photoshop, Snapseed, GIMP), the image was processed after capture. Cropping or exposure adjustment is normal; content-aware fill or layer work warrants follow-up.

For this image: EXIF shows an iPhone 14, DateTimeOriginal yesterday at 07:12, GPS coordinates placing the camera 340 metres downstream from the Acme Chemical complex on the Millford River. No software editing field. All metadata is consistent with the source's account.

Geolocation cross-check. SunCalc for the GPS coordinates on yesterday's date at 07:12 shows the sun at 14 degrees elevation, bearing 72 degrees (east-northeast). Shadows in the photograph fall toward the west-southwest, consistent with the calculated sun position. Weather archives show clear skies at the nearest station yesterday morning — consistent with the even lighting in the photograph.

Manipulation check. Visual inspection: lighting and shadows are consistent across the frame. No visible clone-stamp repetition in the mud or water. The dead fish show natural variation in size and position — not copy-pasted. Error-level analysis (via FotoForensics) shows uniform compression across the image with no regions of anomalous quality, consistent with a single-capture JPEG. No C2PA Content Credentials are embedded. This is expected — consumer iPhones do not currently write C2PA manifests by default.

What remains unverified. The source's identity is unconfirmed. The substance in the water has not been identified — the brown discolouration could be a chemical discharge, but it could also be sediment runoff after heavy rain. The cause of the fish deaths is not established by the photograph alone.

VERIFICATION TRAIL
Mode: visual media
Material type: image
Source: local angler, email to newsroom, no prior contact
Claimed provenance: chemical spill, Millford River, yesterday ~07:15
---
SIFT steps completed:
  Stop: photo must show Millford River, yesterday morning, with visible pollution
  Investigate source: name consistent with local fishing club; no red flags, no independent confirmation
  Find original: no matches on Google Images, TinEye, Yandex — image not in prior circulation
  Trace claim: no other outlet or social account has published this; no agency report filed
Mode-specific checks:
  Metadata inspection:
    Method: EXIF extraction (ExifTool)
    Key findings: iPhone 14, 07:12 yesterday, GPS 340m downstream of Acme Chemical complex
    Inconsistencies: none detected
  Manipulation check:
    Deepfake indicators: not applicable (still image)
    Tampering indicators: none — consistent lighting, no clone artefacts, uniform ELA
    C2PA credentials: absent (expected for consumer iPhone)
  Geolocation: sun angle and shadow direction consistent with GPS coordinates and timestamp; weather archives confirm clear skies
Unverified elements: source identity; substance identification; cause of fish deaths
Confidence level: MEDIUM — metadata, geolocation, and image analysis support the location and timing claim; substance and source identity unconfirmed
Verified by: [reporter name]
Date: [today's date]

Next Step: Contact the regional environmental agency directly to ask whether a spill has been reported. Send a photographer or reporter to the GPS coordinates to confirm the scene independently. If the scene is confirmed, file a public records request for inspection reports on the industrial complex upstream (see the FOIA Request Writer skill). Do not publish the image as showing a "chemical spill" — publish it as showing "dead fish and discolouration on the Millford River" until the substance is identified by a lab or agency.

Example 2: Anonymous Source Verification

Input

A person contacted the newsroom through an encrypted messaging app claiming to be a mid-level procurement officer at a metropolitan transit authority. The source says the transit authority awarded a $14 million bus maintenance contract to a company whose owner is the sibling of a senior transit official — and that the winning bid was $3.2 million higher than the next-lowest bid. The source has provided a photograph of what they say is the bid evaluation scoresheet, taken with a phone camera. The source will not meet in person and asks to be identified only as "a current employee with direct knowledge of the procurement process."

Output

Verification report

The claim: a transit authority awarded a $14 million contract through a rigged or improperly influenced procurement process, favouring a company with a family connection to a senior official. The key assertions to verify: (1) the contract exists and is worth approximately $14 million, (2) the winning company's owner is related to a transit official, (3) the winning bid was significantly

…(truncated)