[ recap ]

Last week we looked at Orwell's 1946 warning that bad language produces bad thinking, and a Cornell study from March showing workers who rated corporate nonsense as "business savvy" scored worse on analytical tasks.

[ thoughts ]

Marshall McLuhan wrote one line in 1964: the medium is the message.

McLuhan was a Canadian professor of English literature who spent the 1950s analysing advertising, mass culture and early television age.

Understanding Media came out in 1964 and his single argument underpinned everything: the tools we use to communicate quietly restructure how we think, independent of what we say through them.

His specific claim was that the content of any medium is always another medium. Speech is the content of writing, written word is the content of print, print is the content of the telegraph.

In late February 2026, OpenAI announced that ChatGPT had reached 900 million weekly active users. It remains the fastest-growing consumer application in history, having hit 100 million monthly actives within two months of launch.

Most of those 900 million people are doing the same thing: writing prompts.

A prompt is a sentence or paragraph of natural language that tells generative AI what to make:
"Write a cover letter for this job."
"Summarise this document."
"Generate an image of a cat in the style of a Japanese woodblock print."

The instruction goes into a text box, the system produces output, and the exchange continues as long as the user keeps typing.

Apply McLuhan's test: what is the content of the prompt?

A designer types:
"Write a tagline for a minimalist running shoe, in the voice of a 1960s Volkswagen ad, then generate three product photos on a white background, and draft the Instagram caption."

One sentence. Inside it: copywriting, art direction, mid-century print advertising, studio photography, social media convention. Each is a medium with its own history and its own trained practitioners. The prompt absorbs them all and asks for output in seconds.

This is the character of this medium. It rewards articulation -- the ability to describe what you want in words -- and removes the friction that used to force people to think before they could execute.

[ creativity ]

Eight years later, John Berger was working the same question from a different direction.

John Berger was a British art critic, painter and novelist. In 1972 the BBC broadcast his four-part series Ways of Seeing, a direct response to Kenneth Clark's Civilisation (the establishment history of Western art). Berger drew from Walter Benjamin's 1935 essay on mechanical reproduction and argued that the way we look at images is shaped by what we have been taught to see.

Berger opens:

"Seeing comes before words. The child looks and recognises before it can speak."

Berger's claim was that seeing is not neutral. It is shaped by what we notice, what we believe, what surrounds us. The relation between what we see and what we know, he wrote, is never settled. We explain the world with words, but we are surrounded by it first.

By the time you type, you have already seen. You have noticed some things and not others. You have decided, without deciding, what kind of answer would count. The AI meets you there, at the end of that seeing, and works with what you brought.

This is why two people asking the same AI the same question get different work from it, and why the difference is not usually about prompt engineering. It is about what each person brought into the room before they started typing. The tool rewards articulation, but articulation is downstream of seeing. The prompt can only reach what the person typing it has already learned to notice.

Which leaves the question: did you bring a way of seeing to the box, or is the box deciding what is worth asking?

[ making ]

An experiment to try for three days:

Every time you send an instruction to an AI -- typed, dictated, pasted from a template, or drafted with help from another tool -- paste the final prompt into a note before you press enter.

On day two, add a step. Before your next prompt, write three sentences:
What do I actually see here?
What do I already know about it?
What is the gap between the two?

Then write the prompt. Log both the three sentences and the prompt they produced. Keep doing this for the rest of the three days, alongside your normal prompts.

At the end of the three days, read the list. For each entry, ask a single question: where did the thinking happen?

[ sign off ]