Verify synthetic or manipulated media without relying on one spectacular clue.
A visual artefact, an unusual voice or an automated detector is not a verdict. Strong verification begins with provenance, context and independent confirmation.
Keep the original file, message, address, time, apparent sender and discovery channel. Messaging recompression may remove useful information.
2
Find the earliest publication
Trace reposts, search for earlier versions and identify the first known account or site. An older copy may reveal different framing or caption.
3
Check the context
Verify location, weather, chronology, people, objects and visible events using independent dated sources.
4
Examine continuity
For video and audio, inspect transitions, motion, shadows, reflections, synchronisation and cuts. For images, compare edges, perspective, repetition and compression zones.
5
Use metadata carefully
Metadata can guide the review but may be absent, changed or stripped. Consistency supports a case; absence does not prove fabrication.
6
State a graded conclusion
Distinguish consistent with available evidence, probably manipulated, probably synthetic or undetermined. State limits and the next check that could change the assessment.
Provenance outweighs a single artefact
Generation and editing tools change quickly. Checklists of supposed defects become outdated and can falsely discredit authentic media.
A coherent chain of provenance, independent contemporary publications and verifiable scene details usually matter more than an opaque detector score.
Use tools as instruments, not judges
Reverse-image search, frame review, spectrograms and metadata analysis each answer a limited question. Record the result, date, tool version and tested hypothesis.
An automated score is useful only when its scope, error rate and media type are understood. Do not publish a percentage without explanation.
Avoid amplifying a false item
Do not repost the complete item merely to ask whether it is fake. Describe the concern, share the minimum necessary and keep a separate working copy.
When a person is targeted, consider harassment, defamation, fraud and privacy risk before sharing.
Separate file, transformation and scene
A platform may recompress media, remove metadata or create thumbnails. Different bytes do not prove scene manipulation. An internal timestamp may concern the file rather than the event: preserve its value and context before interpreting it.
Assess provenance declarations
Record tool, signer and declared operations where provenance exists. Separate technical validation from trust in declarations. Independently check caption, location and date. Documented provenance may accompany a misleading caption; absence may result from copying.
Reading technical evidence
A hash compares files; it does not authenticate their story
SHA-256 produces a 256-bit digest, represented here by 64 hexadecimal characters. It is calculated from the file’s bytes.
Field or situation and its interpretation
Observed item
What it means
Renamed copy
The name is not part of the calculation; the content can retain its hash.
Recompressed image or resaved document
Bytes may change even when the document looks identical.
Hash calculated today
It proves neither creation date, authorship nor authenticity of the scene.
Record algorithm, digest, file and collection context. Compare against a reference retained separately.
Separate media provenance from the truth of its caption
Seek the earliest known publication, original context and independent views. A genuine image can carry a false caption. Examine the visual content, claimed location and claimed date separately: disproving one does not establish the others.
Retain the original link or file, circulating caption, searches and earlier versions. State when only a cropped copy is available. Detection scores and visual clues are not conclusive. Provenance or marking may document file history without proving the depicted event; their absence does not prove synthesis. Report an appropriately bounded conclusion rather than an unsupported accusation.
Origin
Seek the creator, date and earliest known publication.
Context
Compare location and event with independent evidence.
Conclusion
Distinguish unverified from demonstrated false.
Fictional example: viral video. A supposedly recent protest clip has a longer version published years earlier. The reviewer corrects the date before assessing possible visual manipulation.
Can an AI-detector score settle the question?
No. Examine provenance, context and the tool’s limitations; the score alone is an indication.
Try the reasoning
A detector labels an image synthetic with no other evidence. Can you publish an accusation?
Reasoning: No. State uncertainty and seek provenance, context and earlier versions.
For each point, record what the material actually supports. “Supported” applies to the specific observation, not to the entire case. A contradiction needs explanation; missing material remains an open question.
Define the question
Record the source
Test the observation
State the limits
Entries stay in this page until you leave or reload it. Nothing is saved or submitted by this tool. Use neutral references instead of personal data or credentials.
How to write a useful conclusion
Editorial examples, not findings about your case
Material
Useful wording
Next check
One dated source
The consulted source stated X on the recorded date.
Check its origin and whether it still applies.
Multiple copies of one claim
Several pages repeat X; independent corroboration is missing.
Find the original and a source with independent evidence.
Conflicting sources
A states X while B states Y; the difference is unresolved.
Compare dates, scope, identifiers and definitions.
No matching result
No result was found in the recorded scope.
Describe queries, coverage and access limitations.
What to check and what to record
Use this grid to turn the method into a reviewable case file. A missing item remains an open question. The interpretation limit prevents a finding from becoming an unsupported conclusion.
Topic-specific checks
Check
Useful record
Interpretation limit
Origin
Received file, earliest located publisher and reposting chain.
The earliest found occurrence may not be the original.
Context
Claimed place and date, plus independently corroborated visible details.
Real media can be used in a false context.
Automated detection
Tool, version, analysed format and stated limitations.
A detector score cannot establish fabrication on its own.
Common pitfalls
Four shortcuts that weaken the result.
Hunting for one defect
Odd details may result from compression, motion, lighting or ordinary editing.
Trusting one detector
Performance varies by model, format and transformation.
Losing the original
A screenshot often removes the file, audio and metadata needed for review.
Confusing false context with false media
Authentic media can be paired with a false date, place or caption.
Practical questions
Frequently asked questions.
Does metadata prove authenticity?
No. It may support a timeline or reveal processing, but it can be removed or changed.
Can deepfakes be spotted by eye?
Some inconsistencies may be visible, but no visible artefact proves nothing. Provenance and corroboration remain essential.
Should a detector score be published?
Only with its context, limitations and other evidence. An isolated percentage can mislead.
Public references
These primary references support the checks described here. The sequence and working sheet are an IIS editorial method; a cited organisation does not endorse this guide. Read each reference within its stated date, jurisdiction and scope.
NIST AI 100-4 — Synthetic content (2024). Technical overview of provenance, watermarking and detection approaches. These are different signals with different limits; a detector output must be interpreted within its tested conditions, rather than as a universal authenticity verdict.
Editorial scope
Published by Internet Intelligence Service on 15 September 2026. Last content update: 15 November 2026. This educational guide describes a lawful, defensive method. It is not legal advice, an emergency service or authority instruction.