The verification chain

01Why order matters more than tools

Verification failures rarely come from a lack of tools. They come from starting in the wrong place — analysing what an image appears to show before establishing where it came from — and from stopping as soon as the material appears to confirm an expectation.

A chain is therefore a sequence, and each link is a question that can fail. Provenance, then location, then time, then independent corroboration, then documentation, then an explicit statement of what remains unestablished. Material that fails a link is not published as though it had passed.

Before any of it: research must stay inside the law and the data-protection framework. Circumventing access controls, using credentials that are not yours or acquiring data of unclear origin are outside the scope of this method, and nothing here is legal advice.

02Link one: provenance

The first question is who published this first, and when. Find the earliest available copy rather than the most widely shared one: reverse image search for prior appearances, archived versions of pages, earlier uploads of a video that later circulated in a shorter cut.

Then examine the account or body that posted it. How long has it existed, what has it published before, does it claim first-hand knowledge or is it aggregating, and has it circulated recycled material previously? An account created a week ago that posts exclusively about one conflict is not a source; it is an actor whose material may still be genuine and must be treated accordingly.

Where possible, reach the person who made the material. A direct account of how it was recorded — where they stood, what happened before and after, what device — is worth more than any technical analysis, and inconsistencies in that account are diagnostic.

03Link two: location

Locate the scene from features that cannot move: skyline shape, hill profiles, junction layout, window patterns, road markings, kerb and pavement design, street furniture, cable arrangements, signage conventions, vegetation. Language, alphabet, plate formats and the side of the road in use narrow the country quickly.

Test a candidate location against independent imagery feature by feature, looking actively for disconfirmation: something that should be visible and is not, a road curving the wrong way, a roofline that does not match. Record the capture date of every reference used, because a mismatch may reflect construction between the reference and the material rather than a false claim.

A location is established when several independent fixed features agree and no strong feature contradicts. One matching building is a coincidence; four matching features in the correct spatial relationship is evidence.

04Link three: time

Dating uses independent physical records. Shadow direction and length, given a location and date, constrain the time of day through published solar geometry. Official weather observations for a nearby station record cloud, precipitation and visibility by hour, so bright sunlight contradicts a claim of heavy rain at that time.

Seasonal and event markers add further constraint: foliage state, snow, election posters, seasonal displays, sports fixtures, the stage of a construction site, roadworks visible in dated imagery. Each narrows the window, and together they can often fix a date to within days.

Embedded file metadata may carry a timestamp and coordinates. Treat it as a lead: platforms strip it, and it can be edited. A conclusion should rest on features that agree with independent records, not on a field anyone could have altered.

05Link four: independent corroboration

A finding is reportable when sources that do not depend on each other agree. Three accounts reposting one original are one source. Genuine independence means separate origins: another item recorded from a different angle, an official statement, a court or planning record, a business register, published statistics, a physical document, a named witness.

Look for contradiction as deliberately as for agreement. The question is not whether some evidence supports the claim but whether any strong evidence undermines it, and the second question is the one that protects a publication.

06Link five: documentation

Capture everything as you use it: full-page copies, the address, the date and time of capture, and where possible an independent archive reference. Public material is deleted and edited, and a finding that cannot be shown to an editor or a lawyer is a recollection.

The written record should list the searches performed and when, the features matched and against which references with their capture dates, the solar and weather sources consulted, the corroboration obtained, the contradictions found, the judgements made and the questions left open. Written this way, the verification can be checked by someone who was not present for it.

07Link six: the stated limit

Every verification ends with a boundary, and naming it is part of the method rather than a weakness in it. A typical honest conclusion: several fixed features match a specific junction; shadow direction is consistent with mid-morning; the date is consistent with the claimed week but not confirmed; the identity of the person filming is unknown.

That formulation is more useful to a reader than a confident assertion, and it is the only one that remains true if new material emerges. It also disciplines the headline, which is where over-claiming usually happens after a careful body has been written.

08When a link fails: hold, discard or publish with a limit

A chain that does not complete is the normal case rather than the exception, and the useful question is which link failed. Failures are not equivalent, and treating them as a single category is how material is either wrongly discarded or wrongly published.

A provenance failure is close to fatal. If the earliest copy cannot be found and the account that posted it has no history, there is nothing to build on and the material should be set aside rather than published with a hedge. A location failure means the material may be genuine and its setting unknown: it can sometimes be used for what it shows without any claim about where, but only if the text says the location is unestablished and the absence of a location does not itself change the meaning. A time failure is the most survivable, because a scene consistent with a place and inconsistent with nothing can be reported with the date stated as unconfirmed.

A corroboration failure demands the sharpest language. A single-source item that has survived provenance, location and time is still one source, and the honest form names it as such: one recording, its origin and date established, with no independent record of the event found. Holding is often the right decision — most material that cannot be corroborated today can be corroborated in a week — and the record should say what would resolve it, so that the decision to hold does not quietly become a decision to abandon.

09Two people, one chain

Verification degrades when one person performs every link, because each conclusion becomes the premise of the next. Someone who has settled on a junction will read the shadows in a way that fits it, and will notice the features that agree while passing over the one that does not. Splitting the work interrupts that.

The practical arrangement is a blind second pass on the two links most vulnerable to it. A second person is given the material and the claim to be tested — this was recorded at a stated place, on a stated morning — but not the candidate location or the reference images used to reach it. If they arrive independently at the same junction, the location is strong. If they arrive somewhere else, the disagreement locates the weak feature immediately.

Where only one person is available, the substitute is time and an explicit search for disconfirmation: write down what would have to be true for the conclusion to be wrong, then look specifically for that. It is weaker than a second pair of eyes, and it is considerably better than reading one's own conclusion back.

10Data protection and harm

Assembling information about identifiable people is processing personal data even when each item is public. Journalistic purposes are recognised in the framework, but proportionality, security of what is held and a retention decision continue to apply. Keep research files encrypted and access-limited, and collect only what the story needs.

Harm assessment precedes publication. Consider the subject, bystanders visible in material, the person who originally posted it and the researcher. A street-level location can endanger people, and precision that serves no public interest should not be published merely because it was established.

11A worked order of operations

The order is deliberate. Location and time are established before corroboration because they frequently eliminate material outright, and eliminating early saves the effort of corroborating something that is two years old and from another country.

• Step 1Find the earliest available copy; record where and when it first appeared.
• Step 2Assess the publishing account's history and claim to first-hand knowledge.
• Step 3Locate from fixed features; test against dated reference imagery.
• Step 4Constrain the time using shadows, weather records and seasonal markers.
• Step 5Seek corroboration from sources with separate origins.
• Step 6Actively look for evidence that would contradict the conclusion.
• Step 7Archive every source with capture dates; write the reasoning.
• Step 8State the limit; decide the level of detail to publish.

12Manipulation, synthesis and proportionate suspicion

Most misleading material is genuine footage with a false caption. Sophisticated manipulation exists, but a chain that establishes location and time defeats the common cases without any pixel-level analysis, which is why the chain is ordered as it is.

Where manipulation or synthesis is suspected, detection tools give probabilities and misfire on compressed files. They are one input. The reliable route remains corroboration through independent records of the event, and a claim that material is fabricated needs the same standard of evidence as a claim that it is real.