Verify the claim, not the picture

The reframe

The instinct on seeing a doubtful image is to study it. That is the least productive available action, for every reason the earlier lessons gave.

The productive question is not "is this generated" but "what is being claimed, and what supports it". Almost always the claim is bundled with the image — that this shows a specific event, at a specific place, involving specific people. Those are verifiable independently of the pixels.

This is not new technique. It is what newsrooms have done with user-submitted photographs for fifteen years, and it survived every previous improvement in image manipulation.

The routine

Find the earliest appearance. Reverse image search across several services, because they index differently — Google Lens, TinEye, Yandex and Bing all return different results, and all are free. TinEye in particular sorts by oldest, which is exactly what you want. If the image appears three years before the event it supposedly shows, you are finished.

Identify the source account. How old is it, what has it posted before, does it have a name attached, does anybody with a reputation stand behind the claim. An image with no source is unverified regardless of how it looks.

Check the place. If the image claims a location, compare against satellite and street-level imagery. Building shapes, road layouts, signage, the side of the road traffic drives on, the language on shopfronts, vegetation. This is slower and it is decisive, because a generated or misattributed image rarely matches a real place in detail.

Check the time. Shadows give a rough time of day and, with a known location, a rough date. Weather records are public. If the image shows heavy rain and the archive says clear skies, that is a fact rather than an impression.

Look for corroboration. Real events of consequence generate multiple independent recordings from different angles. A single image with no other coverage is a reason for caution, whether or not it is generated.

Read the metadata, knowing it proves little. Its presence can be informative; its absence is normal.

Check provenance for credentials, knowing most files have none.

Why this beats inspection every time

Every step above depends on facts outside the file. The generator's quality does not affect a satellite photograph, an archived weather record, or the date an image was first indexed.

That is what makes the approach durable. Inspection-based methods expire with each model release. This one does not.

The discipline of not concluding

The hardest part is being willing to end at "unverified".

There are three possible outcomes, and people reliably collapse them into two. Verified: origin and claim check out. Debunked: found earlier elsewhere, or the place is wrong. Unverified: nothing established either way, which is the most common result and a legitimate place to stop.

An unverified image should not be shared as true and should not be denounced as fake. Both mistakes are common, and the second one is now doing serious damage to people whose genuine work gets dismissed.

Practical habits

Slow down at the moment of highest emotion. The images engineered to spread are the ones that make you angry. That reaction is the trigger to check, not to share.

Screenshot before checking, keeping the original file and its address. Content disappears while you are looking at it.

Search the claim in words, not just the image. If something significant happened, credible outlets will have it; if only one account has it, that is informative.

Write down what you established and what you did not. For anything you publish, this record is both your defence and your reason for confidence.

A note on what to do with an old image being reused. This is the single most common form of misleading media and it involves no generation at all: a real photograph of a real event, presented as a different event. Reverse search catches it in seconds, and it accounts for a large share of what circulates during any crisis. Anyone who learns only the first three steps of this routine will catch far more misinformation than someone who becomes expert at spotting generated pixels, because the recycled photograph is both more common and more persuasive — it survives every kind of inspection, since nothing about it is fake except the caption.

The limitation of this method, stated plainly: it works on images making a factual claim about the world, which is where the harm concentrates. It does nothing for an image where the claim is about a private person and there is no public record to check against — non-consensual imagery, harassment, fabricated evidence in a personal dispute. Those cases need law and platform action, not verification technique, and they are where the current gap is widest.