Useful things, and a few playful ones.
I've spent thirty years in digital media and communications, first running agencies and now in higher education, where I mostly work on research communications, content strategy and data visualisation.
I can't remember a time when I haven't used user personas in work I've created over the past thirty years (say it quickly, doesn't sound so loooong): web design; UX; print; online advertising - the lot.
The Edge Persona Prompt Suite is all about personas but takes a different approach. This collection is explicitly about public service design BUT the approach could easily be adapted for any purpose (in fact, I've been creating full suites for a wide range of higher education marketing uses).
What this version is:
Twenty-three people that may not have been considered during the service design process. Ray struggles with the letter. Jade has no fixed address. Marcus doesn't have a smartphone. Someone else may be reading Hannah's messages. Nadia's capacity changes from one day to the next.
Use them as AI roleplay prompts, layered with scenario overlays, to find where a service or a piece of comms falls over before real people do. They're a stress-test, not a list of special cases, and they sit alongside proper user research rather than replacing it.
What happens to trust when anyone can make a convincing picture of anything? This chapter looks at how AI-generated images change what people are prepared to believe, and what institutions owe their audiences when they use synthetic imagery. The floods in Valencia in 2024 saw real images being treated with suspicion and a whirlwind of climate change misinformation and conspiracy theories dominating the narrative. Sam Vimes, in Terry Pratchett's Discworld novel The Truth, says "A lie can run round the world before the truth has got its boots on". Research in 2018 found false news travelling around six times faster than the truth...
To read the full chapter - it's in Artificial Intelligence in Creative Industries: Psychological and Social Implications, edited by Francisco Tigre Moura and Caterina Moruzzi (Routledge). The whole book is open access, so dive in.
For a couple of years I tried to find out how much energy my own AI use was burning through, and kept getting nowhere. One provider publishes a per-prompt figure with a proper method, another put a number in a blog post with no method at all, and most publish nothing. So I built my own.
Receipt counts your prompts (never what you type) and gives you a range, with every factor sourced and graded. No server, no account, no network calls. It all stays on your own machine.
My week came to 65 Wh, about a kettle boil. The more useful finding was that 17% of my prompts, the ones on "thinking" modes, used roughly 93% of the energy, mostly on drafting a standard model would have handled just as well.
There's also a tab called "Not yours to optimise", for the things no amount of switching models will fix: datacentre siting, water, chip manufacture and the annotation workforce. It's early days, so feedback goes straight into the next version.
Squashies, the half-pink half-white sweets, are the exact colour of a Scottish person after twenty minutes of actual sunshine. Then Scotland qualified for the 2026 World Cup, their first since 1998, played across an American summer. The rest wrote itself. So I made the sponsorship campaign nobody asked for: posters, and a wee arcade of daft games (Squashie or Scot?, sunburn bingo, a penalty shoot-out rigged to end in heartbreak). Entirely unofficial, gloriously unsanctioned. If you know, you know.
For commissions, collaborations or a question about anything published here, email marie@springcorp.co.uk.