Orwell's five rules of writing are your best AI prompt guide for copy

AI content often lacks emotional pull and resonance. In fact, it sounds like what it is:  a confident machine that’s read everything and feels nothing.

The antidote to this slop was written in 1946 by George Orwell, just four years before Alan Turing asked, “Can machines think,” which arguably laid the foundation for the pursuit of AI. 

George Orwell's essay Politics and the English Language is one of the most cited pieces of writing advice in the English-speaking world, and for good reason. Orwell was obsessed with a single idea: that lazy, vague, or dishonest writing is not just an aesthetic failure — it is a moral one. Unclear prose, he argued, is how politicians dodge accountability, how institutions obscure bad news, and how ideas that cannot survive plain language get smuggled past the reader.

His five rules were designed to fight that tendency in human writers. They turn out to be equally powerful when used to fight it in AI ones.

The five rules, and how to use them as prompts

Rule 1: Never use a long word where a short one will do

Orwell's target was the kind of writer who reaches for "utilise" when they mean "use," or "facilitate" when they mean "help." It serves the writer's ego at the expense of the reader's time.

AI models do this. Left to their own devices, they will "leverage synergies" and "demonstrate proficiency" and "provide transformative solutions." None of this means anything.

The prompt fix: Tell the model to use plain English. A phrase like "write this as if explaining it to a smart friend over coffee; no jargon, no corporate language, no long words where short ones will do" produces dramatically different output. You can even add: "if you use a word with more than three syllables, ask yourself if there's an alternative."

The difference between AI prose and human prose often comes down to vocabulary anxiety where the model reaches for big words to signal authority. Let the model be simple, and it will be.

Rule 2: Never use a phrase, scientific word, or jargon if you can think of an everyday simple equivalent

Orwell had in mind writers who use vocabulary to signal membership of a particular tribe — academic, professional, political.

AI models are trained on the internet, which means they've subsumed tribal language. Ask an AI to write about marketing and it will reach for "omnichannel strategies" and "customer journey touchpoints." Ask it about technology and you'll get "scalable infrastructure" and "robust ecosystems."

None of this is wrong, exactly. But it isn't writing. It's signalling.

The prompt fix: Name the jargon you want to avoid. "Do not use marketing buzzwords. Write about this as a person would, not as a consultant's deck would." Or better: "If you find yourself using industry jargon, stop and replace it with the plain English version of what you actually mean." You can also ask the model to flag where it has used technical terms and explain why they're used.

Rule 3: If it is possible to cut a word, cut it

This is where Orwell gets strict. Every word should do work. Anything that’s merely decorating the sentence, adding length without meaning, should go.

AI models are challenged here as they tend toward completeness over concision. They will write five sentences where two would do, add a recap paragraph that restates everything you just read, and round off every section with a summary that begins "In conclusion..." or "Overall, it is clear that..."

The prompt fix: Be ruthless in your brief. "Write this in [X] words maximum — do not go over. Every sentence must earn its place. No summary paragraphs, no recaps, no throat-clearing introductions." You can also ask the model to review its own output: "Now read what you've written and cut 20% of the words without losing any meaning." This second-pass instruction often produces noticeably tighter prose.

Rule 4: Never use the passive where you can use the active

"Mistakes were made" is the passive voice at its most politically useful — and its most dishonest. The passive allows the writer to describe events without assigning responsibility. Things happen. Outcomes occur. Nobody does anything.

AI content is littered with it. "It has been suggested that..." (by whom?). "Research has shown..." (what research?). "It is widely believed..." (widely by who?). The passive voice creates an illusion of authority while actually draining the writing of accountability and life.

The prompt fix: "Use the active voice throughout. Something or someone must be doing something in every sentence. Do not write 'it has been found'. Write who found it and what they found." This forces the model to be specific.

Active writing also simply sounds more human. People in conversation don't speak in a passive way. They say "we tried this and it didn't work". Not "attempts were made which proved unsuccessful."

Rule 5: Never use a metaphor, simile or other figure of speech which you are used to seeing in print

Orwell called these "dead metaphors", phrases so overused that they have lost all meaning and imagery. "Level playing field." "Move the needle." "Low-hanging fruit." "At the end of the day." They were once vivid comparisons. Now our eyes wash over them and brains disengage.

This is perhaps the rule where AI fails most. Models are trained on a catalogue of cliché. They reach for familiar phrases because familiar phrases are statistically common. When an AI wants to convey difficulty, it talks about "navigating challenges." When it wants to convey opportunity, it reaches for "unlocking potential." When it summarises, everything is "key takeaways."

This is the texture of AI slop. It's not wrong. It’s just inert.

The prompt fix: "Avoid all clichés and overused business phrases. If you find yourself about to write 'at the end of the day,' 'move the needle,' 'unlock potential,' or any similar phrase, stop and say the thing directly instead." You can also prompt the model to be concrete rather than metaphorical: "Don't tell me something is 'challenging' — tell me specifically what is hard about it and why." Specificity is the natural enemy of cliché.

Putting it all together, the Orwell prompt

To use these rules every time you prompt an AI for written content, here’s a compact version for any brief:

Write in plain English. Use short words over long ones. Avoid jargon and buzzwords. If you use a technical term, it must be necessary. Be concise: cut any word that isn't doing work, and do not add summary or recap paragraphs. Use the active voice throughout, someone must be doing something in every sentence. Avoid all clichés and dead metaphors; say the thing directly instead of reaching for a familiar phrase. The writing should sound like a knowledgeable person talking, not a consultant presenting, not a press release, and not a machine.

That's it. Paste that into any content brief and the output will be noticeably more human, more readable, and more honest.

Why this matters more than it sounds

Orwell's view was that clear writing encourages clear thinking.

That holds for AI too. When a model waffles and uses abstractions it’s often hiding or promoting unsupported claims. The jargon obscures whether the model actually knows what it's talking about. The clichés fill space where insight should be.

The Orwell rules don't just make AI content more pleasant to read. They make it more honest. And in a world increasingly drowning in AI generated text, honesty is the only differentiator left.

"Good prose is like a window pane," Orwell wrote. The goal is for the reader to see through the writing to the idea, not the writing itself.

Most teams read these rules and nod. Then they go back to getting the same flat, generic output. The difference is knowing how to build these principles into the way your whole team works with AI — not just once, but every time. That's exactly what we teach at Fantasia.