Putting our impact report on video without letting the pictures overclaim

Every January I write the shop’s impact report. It runs to a page and a bit: refills against new packaging, which suppliers changed and why, how the cargo bike coped with deliveries, and a short section headed “What didn’t work”. It goes on our website as a PDF and gets a line in the newsletter.

Hardly anyone opens it.

Meanwhile a wobbly clip of me topping up the oat bin, filmed on my phone on a quiet Tuesday, has been watched by more people than the report has had readers in its whole life. I am not proud of that comparison, but I stopped arguing with it. So this year the report is also going out as short clips, one claim at a time, on the shop’s social accounts.

Doing that properly turned out to be less about video and more about honesty. Moving pictures say things. Some of them say things you never meant to say.

Start with the sentence

Before anything is filmed or made, I write the sentence the clip exists to carry. One sentence, one claim, in words a customer would use at the counter. “Most of what we sold last year left the shop in a container someone brought from home.” “We changed our washing-up liquid supplier because the old one couldn’t take its drums back.”

Next to each sentence I note where the proof lives. The till report. The supplier’s email. The paragraph in the report that explains how we counted.

If I can’t point to anything, the sentence doesn’t get a clip, and it probably shouldn’t be in the report either.

It sounds fussy. It saves time. Half my ideas for clips fell away at this step, and they were exactly the ones that would have got us into trouble.

Proof gets filmed, mood can be made

This is the rule I’ve settled on, and I’d pass it to any small business with an environmental story to tell.

Anything that shows our operations has to be real footage: our shelves, the cargo bike, sacks of lentils coming off the wholesaler’s van, my own hand on the scoop. Phone footage is fine. Better than fine, really, because a slightly shaky clip of the actual shop is evidence, and a polished clip of somewhere else is not.

Generated footage is for everything that isn’t evidence. The backdrop behind a sentence, the weather, a bit of atmosphere to open the series. It must never be mistaken for us.

Regulators read pictures as well as words. The Competition and Markets Authority’s Green Claims Code guidance says all aspects of a claim may be relevant, including “the colours, pictures and logos used”. In the US, the FTC’s Green Guides describe an advert showing a laser printer in a bird’s nest, surrounded by dense forest, with “Buy our printer. Make a change.” in green type. The advert never says the printer is good for the planet. The Guides call it deceptive anyway, because the pictures and words together suggest far-reaching environmental benefits the company would be very unlikely to prove.

So a generated meadow behind the words “our packaging” is a claim. So is a sunlit forest drifting behind a product you sell.

Things I won’t generate, whatever tool I’m using:

• our shop, our suppliers’ farms or warehouses, or anyone who works with us
• wildlife or plants presented as living anywhere near our business
• the untouched-nature set pieces: a seedling cupped in two hands, a dewdrop on a leaf, the planet resting in grass
• anything carrying a certification mark, a logo or a label that looks official

Stock libraries don’t get a free pass here. The cupped-hands seedling turns up on so many company pages that it has stopped meaning anything except “we found this in a stock library”.

Where made-up footage earns its place

Most of a report series isn’t evidence anyway. It’s explanation. Why we stopped selling anything in compostable plastic. How to keep flour fresh without a sealed bag. What to bring along the first time you come to refill.

For posts like those, a short loop of movement under a line of text holds the eye better than a still photo, and nothing in the frame says anything about us. My backdrops so far are deliberately plain: rain streaking a dark pane for the post about what didn’t work, and a tree in full autumn colour seen through an old window for the note that the new report is out.

For the made-up kind I use a free AI video tool on rewind.ai. You describe the scene in plain words, choose square, landscape or portrait to suit where the clip will sit, and download the file. It asked me to set up a free account before the first clip, and it never wanted a card. The FAQ on that page says what you make is yours to use for commercial work as well as personal, which matters on a business account, and my clips came without a watermark.

A few seconds of movement behind one sentence is plenty. Hold it, or let it loop.

Writing the scene

The prompt is where the honesty holds or slips, so I write it the way I’d brief a photographer I trusted.

Name one subject, one kind of light and one slow movement. Something like “frost on a hedge in low winter sun, slow push-in” gives the model a single job, and gives you a single thing to check before the clip goes anywhere.

Leave lettering out of the prompt. Put your sentence on afterwards in whatever editor you use, where you choose the spelling and the font.

Keep people out as well. Hands, faces, a figure walking past: in a generated clip each one invites the question “who is that?”, and the true answer is nobody at all.

Then read the scene back as a stranger would. If a viewer could reasonably think “that’s their shop” or “that’s where their beans are grown”, change it.

Choose the shape before you press the button. Square for the feed, portrait for stories, landscape for the banner above the report on the website.

Posting it so the proof is one tap away

Each clip carries its one sentence on screen, large enough to read on a phone with the sound off. The caption says where the figure comes from and points to the full report.

When a clip uses generated footage, the caption says so in a few plain words: “Background made with AI.” It costs nothing, and it means nobody has to guess.

I space them out, roughly a week apart, so each claim gets its own day and its own questions. The questions are the point. A report people ask about is a report someone has read.

What stayed in the PDF

Not everything made the cut. Our estimate of the carbon saved by the cargo bike came out of a spreadsheet with wide error bars, and I couldn’t find a way to put it on screen without it looking surer than it is. It stays in the report, beside its caveats, for the reader who wants the working.

That was the most useful thing the exercise taught me. A clip behaves like a headline. A figure that can’t survive being a headline belongs on the page, with its footnotes.