Beyond the label: values live in the image edit

The copy is signed off. The link is ready. The post is almost done. Then someone asks: what image are we using?

I used to treat that as the easy part. Pick something that looks good. Something “human”. Something that stops the scroll. Now I think it is often the most values-heavy decision in the whole chain. Images do not just sit beside our words; they steer the conversation, sometimes before the first sentence has had a chance.

Bond’s recent piece on artificial authenticity puts evidence behind what many teams are already seeing: when people suspect an image is AI-generated, attention shifts from the issue to the image itself. In the study Bond summarises, researchers looked at 171 AI-generated images used by 17 major organisations, alongside more than 400 public comments. Only 80 comments (under 20%) engaged with the humanitarian issue. Most argued about whether the image was real, or whether the organisation should be using it at all. The cause was still there, but it had been pushed to the edge of the frame.

The University of East Anglia has called this an “AI shortcut to empathy” that can backfire. And The New Humanitarian makes a point that feels uncomfortably familiar: AI does not arrive in a clean room. It learns from the visual history we have already normalised, stereotypes included.

So yes, we need policy and training. But we also need something you can use when it is 4.47pm, you need a header image, and you can sense your comments section drifting towards “is this real?” before you’ve even hit publish.

Start with the job, not the asset

When I’m under time pressure, I try to stop myself from starting with “what do we have?” and start with “what is this image for?”

That question sounds obvious until you actually answer it.

If the honest answer is “make people feel something quickly” then pause. Emotion is not the problem. Shortcuts are. Shortcuts pull us back towards familiar tropes, whether the picture comes from an AI tool or a stock library: the same faces, the same framing, the same implied helplessness.

If the job is “help people understand”, your options open up. A place. A process. A meeting. A tool. A map. A simple chart. A clearly stylised illustration. Images that show systems and choices often age better than images that ask for sympathy at speed.

This is not a new idea. The Dóchas Code of Conduct on Images and Messages has long argued for dignity, equality and truthful context, and warns against stereotyping or sensationalising. The point is not to hide reality. The point is to stop reducing people to a single, predictable note.

Captions and alt text carry more weight than we admit

A label alone rarely does the work we hope it will. Even when teams disclose AI use properly, audiences can still drift into suspicion if the image is doing too much work by itself. One line that says “AI-generated” can be transparent and still leave a vacuum where context should be. Vacuums fill fast, usually with scepticism.

So I try to treat captions as part of the story, not a tag at the end. A good caption keeps the reader with the issue. Mine aim to do three things:

1. Describe what the reader is looking at
2. Explain why I chose this approach (privacy, safeguarding, illustrating a concept)
3. Anchor back to the point and what happens next

Alt text is the other unglamorous place where values show up. It is easy to treat it as a compliance chore. It is also where dignity lives for readers using screen readers. If you want a practical guide, W3C’s alt decision tree is useful. It helps you decide whether an image is decorative, informative or complex and what the alt text should actually do.

A prompt I use when I’m tempted to write something vague or sentimental: describe the image as if you were describing it to someone you respect. Because you are.

A two-minute pause that saves you later

If you want something you can actually use on deadline, here is the pause I run before publishing an image-heavy post. It is short enough to survive a busy day, but long enough to interrupt autopilot.

• Who benefits from this image being used, and who carries the risk?
• Would the person pictured recognise themselves with dignity?
• Are we giving context, or relying on stereotypes to do emotional work?
• If this travelled widely, would we stand by it without defensiveness?
• Is the medium undermining the message?

That last question is the one the artificial authenticity debate keeps circling. If the medium undermines the message, audiences stop engaging with the cause and start auditing your choices. Once that happens, it’s very hard to steer the conversation back without sounding defensive.

A final practical fix: build your future self a better option. A small “no face” library goes a long way. Places and environments with clear context. Hands doing a task. A meeting table. Documents and data. Maps. Commissioned illustrations that explain systems. These visuals age well, and they keep attention on the work rather than the legitimacy of the image.

And if you do use AI, treat it like a high-risk tool. Avoid photorealism. Do not use it to depict suffering or mimic a specific community. Be explicit about why you are using it, not just that you used it. The closer an image looks like documentary photography, the more it will be read as a claim.

An African proverb says: when you pray for rain, you have to deal with the mud too. Ethical storytelling has mud. It takes more thought at the start. It sometimes means choosing a less dramatic image, or no image at all. But the payoff is real.

When we choose visuals for the job they need to do, make captions carry context, and run a two-minute trust check, we give our messages a better chance of landing where they should: on the issue, on the people, on the change we’re trying to make. Values live in the edit. In our sector, the image is often the edit that matters most.