Preserve context
Record URL, access date, full caption, stated creator, platform and least-transformed file available. Avoid spreading unnecessary personal detail.
GUIDE / PROTOCOL
Verify the context of a public image
An authentic image can still be attached to the wrong place, date or event.
Direct answer
What can a public image actually establish?
Preserve the image and publication in context, then look for earlier versions. Compare visual clues, caption, location and dates against independent sources. Metadata and provenance credentials offer clues, not automatic proof of the narrative.
01 / Review method
Review method
Find versions
Use reverse search and visual comparison to find earlier appearances, crops or frames from video.
Test place and time
Compare landmarks, relevant weather, shadows, signs and public archives. Separate depicted, publishing and claimed locations.
State the result
Distinguish file authenticity, provenance, context and caption truth. Present supporting evidence, conflicts and unknowns.
IN PRACTICE
An image needs a role, a source and a useful format
Decide what the reader must understand first. An atmosphere image may introduce a topic; an image used as evidence needs an origin and context. Decide cropping, caption and file weight afterwards.
Which check comes first?
Choose the situation you observed. Each path gives a first check, useful evidence and the limit of that check.
A photograph has no identifiable source
- First check
- Find the first traceable record and compare image, caption, date and credit. Separate the image producer from accounts republishing it.
- Evidence to retain
- Exact URL, document identifier, credit, known capture and publication dates, and the passage establishing context.
- Limit to remember
- Recognising an image does not confirm an event claimed by a new caption. An unsuccessful search does not prove fabrication.
Cropping removes necessary information
- First check
- Compare the full image with the crop actually shown on mobile and desktop. Check faces, captions, scales, landmarks and details needed for understanding.
- Evidence to retain
- Original image, crop, displayed dimensions, positioning rule and an example of the lost detail.
- Limit to remember
- An attractive crop can change meaning. Alternative text helps according to the image’s role but cannot alone make an unreadable chart usable.
An image seems to delay rendering
- First check
- Identify the file actually loaded, its size and discovery in the network timeline. Compare its width with the display and check whether it carries the main content.
- Evidence to retain
- Image URL, received variant, dimensions, transferred bytes, device, network and repeated measurements before and after the fix.
- Limit to remember
- Smaller files do not resolve late discovery. An undersized variant can harm readability; do not choose compression by byte count alone.
Further reading
| Situation | Avoid | Useful correction |
|---|---|---|
| Atmosphere image | Presenting a generated scene as a real event. | State its illustrative role and AI generation. |
| Documentary photograph | Reusing an attractive image without date or provenance. | Link the record, author or institution, capture date and reuse terms. |
| Explanatory chart | Keeping numbers only inside pixels. | Also provide values, units and limits as readable text or a table. |
Role
What meaning disappears if the image is removed?
Provenance
Which source and caption justify its use?
Format
Which variant stays sharp at the actual display size?
Final check
Review with images hidden, then on a small screen. The caption must remain understandable and essential meaning must not depend on colour alone.
Image weight
Site WebP files; an estimated scenario, not a traffic or speed measurement. Cache, screen density and other resources affect actual results.
Estimated total : 100.5 KiB
960 px reference : 283.1 KiB
Weight difference : 182.6 KiB
| File width | Estimated total |
|---|---|
| 480 px | 100.5 KiB |
| 768 px | 203.6 KiB |
| 960 px | 283.1 KiB |
- Editing desk with photo prints in different formats, a loupe and ruler. Illustrative scene.
- Measurement desk with cards, magnifier, stopwatch and prism. Illustrative scene.
- Earth photographed by the Apollo 17 crew on 7 December 1972. NASA, AS17-148-22727.
Documentary references
Consulted on .
An authentic photo can be used outside its time context
Apollo 17’s Earth photo offers a concrete case: identifying a photograph does not date the story attached to it. The date of a page republishing it and the capture date are separate facts.
- Capture date
- Document identifier
- AS17-148-22727
Check before citing
- Find AS17-148-22727 in the record, then compare the image and caption.
- Keep capture, publication and consultation dates separate. Recent republication does not make a photo recent.
- For a Worldview layer, also record sensor, date and processing: this historical photograph is not a Worldview screenshot.
Real documentary case, consulted on 2026-10-14. Resized and encoded WebP version, with no crop or AI retouching.
A page date is not its photograph’s date
Earthrise lets us compare three documented dates. The photograph is real; the two official records give its context.
- Capture date
- NASA Science record published
- Republication article
Republication in 2024 does not make this a photograph taken in 2024. Retain the capture date and the reference for each consulted page.
What does a consultation date establish?
It dates your access to a source. It replaces neither capture nor publication and does not prove the page was identical earlier.
Sources consulted on . WebP version resized from the NASA-supplied file, without cropping or AI retouching.
02 / Evidence to keep
Evidence to keep
Observed post
URL, dated capture, full caption and stated author or account.
Compared versions
Earlier files or URLs, crops, edits and earliest occurrence found.
Checked clues
Visual features, independent references, available metadata and confidence.
Limited conclusion
What is supported, contradicted or unresolved and the condition for revision.
PRACTICAL CHECK
Before reaching a conclusion
A frequent misreading and the check that helps avoid it.
Premature conclusion
A similar-looking setting is not enough to locate a scene: repeated buildings, seasonal changes and camera angles can mislead.
Decisive check
Match at least two stable landmarks to a dated geographic reference, then check perspective and chronology. Keep competing locations in the record if uncertainty remains.
AFTER THE REVIEW
Decide from the evidence
Reverse search finds an older version
Compare image, crop and captions; the older appearance challenges a recent-first-publication claim without dating the original capture by itself.
Provenance credentials validate but the caption is false
Separate file history from the editorial claim. Correct the caption using independent clues.
Decision and review trail
Decision and review trail
What the check must help decide and what to retain so the choice can be reviewed.
Sufficient evidence
Find the earliest accessible version, compare crops and details, then check place and date through independent clues. Missing metadata does not prove manipulation.
Next decision
State only what image and context establish. Keep the original, compared versions and limits of location or dating.
OSINT / Official references
When locating an image
Find an independent geographic landmark and dated acquisition before interpreting the scene.
cartes.gouv.fr
Why start hereCompare roads, buildings and terrain with a dated IGN layer.
Limit to rememberCheck each layer’s origin, scale and date before comparing places.
Read source entryCopernicus Browser
Why start hereCompare satellite acquisitions of the same location.
Limit to rememberCloud cover, resolution, processing and acquisition date constrain interpretation.
Read source entryNASA Worldview
Why start hereObserve a broad phenomenon across dates and layers.
Limit to rememberA coloured visual depends on its layer and is not itself a ground measurement.
Read source entryDetailed inspection cases
Typical situations for preparing a check. They do not describe completed assignments or actual observations.
A reverse search with no result becomes proof of originality
What to inspect
Try relevant crop variants and record engines, dates and access limits. Also check the publication page and available contextual clues.
Decision and action
Describe the outcome as no match in the searches performed. It establishes neither originality, authenticity nor fabrication of the image.
Record to retain
Keep the analysed image, variants, queries and results or missing results with their conditions.
Missing metadata is treated as a decisive anomaly
What to inspect
Compare the available file with its known distribution chain. Export, compression or transfer can change technical information; inspect what is actually present.
Decision and action
Use metadata as a contextual clue and seek external corroboration for date or location. An absent field is insufficient to establish manipulation.
Record to retain
Keep the examined file, fields read, known transformations and independent evidence supporting the conclusion.
An authentic image changes context
What to inspect
Compare caption, first publication, supported place and period.
Decision and action
Separate file authenticity from caption accuracy.
Record to retain
Image, publications, context differences and bounded conclusion.
Review before deciding
These questions complement the method. They help establish what is actually supported while retaining qualifications in the record.
Version and context
Was the used version compared with earlier traces, framing and the recovered original caption? Reverse search provides a lead; it does not alone establish date, place or authenticity.
View a worked decision
Fictional example: the state applies only to the criterion and scope described.
- Fictional observation
- Reverse search finds a thumbnail; the original and first-publication caption remain unavailable.
- Declared state
- Not checked
- Decision and next check
- Keep the trace as a lead without certifying date or place. Seek a fuller version and earlier contexts.
Corroborated clues
Were location and period clues compared with independent sources? Retain contradictions and distinguish present metadata, verified metadata and a visual hypothesis.
View a worked decision
Fictional example: the state applies only to the criterion and scope described.
- Fictional observation
- The caption names location X; two visible features contradict the map used to support X.
- Declared state
- Observed discrepancy
- Decision and next check
- Retain the contradiction and suspend the proposed location. Do not discard conflicting clues to preserve the caption.
Provenance and truth
Is provenance evidence distinguished from the truth of a caption? Its absence alone is also insufficient to declare an image false.
Documentation consulted : C2PA · Content Credentials Explainer 2.2 ·
View a worked decision
Fictional example: the state applies only to the criterion and scope described.
- Fictional observation
- The record separates validated file provenance from an uncorroborated location caption.
- Declared state
- Verified within this scope
- Decision and next check
- The separation is correct; location remains open. Provenance evidence does not close scene verification.
TOOLS / REGISTERS
Information to record
Keep the received image and discovered versions separately. Reverse search finds occurrences; the earliest found does not prove the first publication or authorship.
- Access date and time zone
- Exact URL
- Capture / acquisition date
- Place / coordinates
- Producer / publisher
- Limit / uncertainty
- Decision / next action
- Review owner
- Next review
03 / FAQ
Frequently asked questions
Do metadata prove authenticity?
They can help reconstruct file history but need corroboration.
Does reverse search find the original?
Not always. It finds known occurrences, not necessarily the first publication.
Does one clue prove manipulation?
Avoid premature certainty; examine compression, cropping and routine processing.
Publish precise location?
Only when necessary, supported and proportionate to risks for people.
04 / Official references
Official references
Describe a context
DOCUMENTED REFERENCE
A real satellite image with two dates to retain
This observation distinguishes acquisition from publication of its record. Treat the image as an identified document, then consider the metadata needed before taking measurements.
Islands, inlets and coastline of Kimberley, Western Australia, observed by Landsat 9 on 31 October 2021.
Real satellite image · NASA / USGS · Source · Reuse
NASA/USGS observation image. Resized and converted to WebP without cropping. Documentary use.
| Reference | Value or form | What it identifies |
|---|---|---|
| Landsat 9 | 2021-10-31 | Image acquisition. |
| NASA SVS · 31172 | 2022-01-13 | SVS record publication. |
| NASA / USGS | Kimberley, Australia | Credit and observed region. |
Implication for the record
Retain sensor, bands, projection, resolution and processing when available. This editorial view alone does not support area measurement or a conclusion about change. Web resizing is disclosed in the credits.
Primary documentation
Documentation consulted on .
DOCUMENTED REFERENCE
EXIF: assemble a timestamp from the right components
Table 12 of CIPA DC-X010-2017 documents these fields and XMP mappings. It is an explicit tag-identifier reference, not a summary of every Exif version. One software-displayed date can hide present or missing components.
| Reference | Value or form | What it identifies |
|---|---|---|
| DateTimeOriginal | 0x9003 · 36867 | Date and time when original data was generated. It contains no UTC offset by itself; do not arbitrarily append Z. |
| OffsetTimeOriginal | 0x9011 · 36881 | UTC offset associated with DateTimeOriginal. Link it to that date, not the receiving computer’s time. |
| SubSecTimeOriginal | 0x9291 · 37521 | Fractional seconds for DateTimeOriginal. Preserve the string and precision; decimals do not establish clock accuracy. |
| DateTimeDigitized | 0x9004 · 36868 | Date and time the image was stored digitally. Digitizing an old physical item and the historical scene depicted can have different dates. |
| OffsetTimeDigitized | 0x9012 · 36882 | UTC offset for DateTimeDigitized. Do not substitute OffsetTimeOriginal unless the operations are documented as identical. |
| SubSecTimeDigitized | 0x9292 · 37522 | Fractional seconds for DateTimeDigitized. Keep the group separate from capture fields to avoid combining different operations into one instant. |
Implication for the record
Retain raw fields, metadata group, reader tool and original file. Software may rewrite or strip metadata; presence authenticates neither scene nor author. Download and file-copy times are separate observations.
Is it possible to fill a missing offset from the country visible in the photo?
Not without more evidence. A presumed scene location establishes neither camera settings nor the timezone rule at that date. Retain an unresolved local time and explain uncertainty.
Primary documentation
Documentation consulted on .
DOCUMENTED REFERENCE
C2PA: validate provenance without certifying the scene
C2PA 2.4 defines signed manifests associated with digital content. It separates structural and binding validation from judging assertions. A validation indicator does not replace editorial image examination.
| Reference | Value or form | What it identifies |
|---|---|---|
| c2pa.claim.v2 | CBOR | A claim gathers assertion references about content at a given time. Its label and encoding do not establish that every real production event was recorded. |
| claim signature | COSE_Sign1_Tagged | Protects the claim. Check integrity, identity and trust using the validator’s configuration; a technical signer is not automatically the photographer. |
| hard binding | hash | Cryptographic binding under hashing and exclusion rules. The specification notes that covered EXIF GPS does not make the signer responsible for location accuracy. |
| soft binding | fingerprint / watermark | Content fingerprint or watermark matching versions with different bytes. Can help recover a separate manifest but cannot replace hard binding. |
| c2pa.ingredient.v3 | parentOf / componentOf | Describes a source or component. Read the relation and available references; a displayed chain guarantees neither original access nor a full history. |
| c2pa.actions.v2 | actions | Describes declared operations such as creation or editing. The specification’s news-photo crop and vignette example is explanatory, not an investigation of a real newsroom. |
Implication for the record
Retain file, available manifest, stated signer, tool and version, result and limits. Missing Content Credentials can result from unsupported tools or stripped metadata; it does not establish falsity. Check context, place, date and rights separately.
Does C2PA validation exclude AI generation?
No. It can also document generated or transformed content. Read available assertions and sources; technical validation is not an automatic real-photo or AI-free classification.
Primary documentation
Documentation consulted on .
DOCUMENTED REFERENCE
Exif orientation: eight arrangements, not just four rotations
The Orientation tag, identifier 274 (0x0112), specifies where the first row and first column of encoded data appear on the display. Some values require mirroring. The diagrams reuse one asymmetric shape and four corners marked A, B, C and D; they do not depict an edited photograph.
| Reference | Value or form | What it identifies |
|---|---|---|
| 1 | A B C D | Reference arrangement: first row at the top, first column on the left. |
| 2 | B A D C | Horizontal mirror of arrangement 1. |
| 3 | C D A B | 180° rotation of arrangement 1. |
| 4 | D C B A | Vertical mirror of arrangement 1. |
| 5 | A D C B | Transpose: horizontal mirror of arrangement 6. Width and height axes are exchanged. |
| 6 | D A B C | 90° clockwise rotation from 1. The axes are exchanged. |
| 7 | C B A D | Horizontal mirror of arrangement 8. The axes are exchanged. |
| 8 | B C D A | 90° counterclockwise rotation from 1. The axes are exchanged. |
Compare the eight arrangements
Every diagram starts from the same 160 × 100 data: A top-left, B top-right, C bottom-right, D bottom-left. Corners after arranging, in top-left, top-right, bottom-right, bottom-left order.
1 · Reference arrangement: first row at the top, first column on the left.
- Corners
- A B C D
- Dimensions after arranging
- 160 × 100
Axes preserved
Axes preserved
Axes preserved
Axes preserved
Axes exchanged
Axes exchanged
Axes exchanged
Axes exchanged
Implication for the record
The tag certifies neither the scene nor the date nor edit history. Software may already have transformed the pixels, retained or removed the tag, or rendered it differently. Before publishing variants, compare decoded appearance and metadata to avoid rotating twice. This reader imports or corrects no file.
Primary documentation
Documentation consulted on .
DEEPER READING / PRACTICE
Examine an image with its previous uses
A photo is broadcast as an illustration of a recent event. The verification searches its origin, the previous versions and the visible elements that allow the scene to be located.
Updated on .
Priority checks
Release URL, legend, displayed date and features of the available file. Compare the occurrences found and distinguish first observed publication and real image creation.
Interpretation pitfalls
An earlier occurrence can invalidate a recent assignment without establishing the exact location. A lack of search results does not prove that the image is original.
Working record to retain
The dossier retains the image consulted, its occurrences, context and a limited conclusion. Record identified transformations or crops and their possible effect on interpretation.
PRACTICAL CASE / WORKED EXAMPLE
An old picture accompanies a recent story.
The example below is pedagogical. It shows how to move from a document to a qualified proposal, preserving the context, the contradictory elements and the information that the research does not yet allow to establish.
Example context
A photo is broadcast to illustrate a situation presented as present. An older occurrence shows the same visual, with another publication context.
Suggested approach
Compare images, recordings, legends and observed dates. Keep the events and look for the elements that locate the scene. Differentiate invalidity and exact unknown origin.
Conclusion and next step
The file may establish that the image is preceding the story without determining all the details of its creation. The conclusion retains this scope difference.
Questions for deeper practice
Does a search without results prove the originality of the image?
Not to. The tool coverage and file transformations limit search. The absence of found event describes the outcome of the consultation, not a general proof of first publication.
DETAILED WORKSHOP / APPLICATION
Compare image occurrences while separating creation and distribution
01 / Prepare the necessary information
Assemble the viewed image, URL, legend and displayed date. Identify visual characteristics and possible transformations to compare the found versions.
02 / Carry out the approach
Examine the events, their context and their dates. Check the signs and visible elements. Distinguish first observed publication, represented scene and file creation.
03 / Examine an alternative situation
If an earlier occurrence invalidates a recent attribution, limit the conclusion to this point. The exact location or author may remain unknown despite this useful verification.
04 / Decide on the next step
The case is concluded for the actual elements. Negative research and missing resources remain described with their coverage and limits.
05 / Hand over the verification records
Transmit images, occurrences and comparison. The conclusions refer to observable details rather than a general similarity that is difficult to verify.
ANNOTATED WORKING DOCUMENT / REVIEW GRID
A report of events with limited conclusion
The folder collects the file consulted, versions, legends and observed dates. It distinguishes recent attribution, antiquity of dissemination and origin of creation when these elements are not all established.
On a small screen, scroll the grid horizontally to read all three columns.
| Observation | Supported interpretation | Next step |
|---|---|---|
| An older occurrence corresponds to the same visual. | The image precede the recent context in the publications found. | Keep the correspondence details and qualify the award invalid. |
| The recharge makes certain elements difficult to compare. | The similarity must be examined before associating the files. | Search for a more comprehensive version and keep the transformations. |
| Research does not find any other occurrence. | The result describes the coverage of the consultation, not a proven originality. | Present this limit and supplement with other relevant indications. |
Reviewing the working document
The final verification separates the conclusions made from the unknown details. A proven antiquity can be useful without automatically identifying the exact author, place or date of creation of the image.
REPORTING / EXAMPLE WORDING
Presentation of results: Examine an image with its previous uses
Point to clarify
An invalid attribution does not mean that all the creation details are known.
Before handing over
Separate proven conclusions from indications and absence of research results.
Example of a qualified conclusion
The occurrences found establish the described element. The author, the exact place or date that is not documented remains open questions.