An AI editor,
available right now.

Most writing that does not happen fails at the start rather than in the middle. Editing something bad is a known process; producing the first bad version is where the hours disappear.

The Writing Copilot is most valuable there - not as a ghostwriter, but as a way to get something on the page fast enough that the real work, which is editing, can start.

Generated prose has a recognisable register: fluent, balanced, and completely free of anything a specific person would say. Readers detect it even when they cannot name it.

The fix is not better prompting. It is editing - putting back the specific number, the awkward admission, the sentence that only makes sense if you know the subject. That part cannot be delegated, and it is the part that makes writing worth reading.

A piece that is not working is usually not working at the outline level, and no amount of sentence polishing reaches that. The paragraphs are fine; they are in the wrong order, or one of them is a different article.

Asking what the argument is - in one sentence, out loud - fixes more drafts than any line edit. It is also the question writers avoid, because sometimes the honest answer is that there is not one yet.

When a draft is not working, the instinct is to rewrite sentences. That is almost always the wrong layer, and it is why some pieces get polished for days without improving.

The first diagnostic is whether there is one argument. Read the draft and write down, in a single sentence, what it is claiming. If that sentence is hard to produce, the problem is not prose. If it produces two sentences, the piece is two pieces and the fix is to choose.

The second is order. A piece often contains all the right material arranged so that each paragraph requires context the reader does not have yet. Listing the paragraphs by what they do - claim, evidence, objection, implication - exposes that quickly, and the fix is rearrangement rather than rewriting.

The third is the opening. A great many drafts contain two or three paragraphs of throat-clearing before the piece begins. Deleting everything before the first genuinely interesting sentence improves a majority of drafts and costs nothing.

Only after those three is line editing worth doing. Compression, rhythm and word choice are real work and they cannot rescue a structural problem - they make a piece that does not work read more smoothly, which is not the same as reading well.

The most common failure in professional writing is a piece written from inside the writer knowledge, and it is invisible to the person who wrote it by definition.

It shows up as unexplained jargon, as assumed context, and as an argument that skips the step the writer finds obvious. The reader hits that step, does not follow, and stops - and no amount of good writing after that point matters.

The repair is to name the reader specifically. Not a demographic but a person: what they already know, what they care about, what they are likely to object to, and what they are going to do after reading. A piece written to a specific reader is narrower and lands harder than one written to everybody.

The objection-handling point deserves emphasis. Writing that anticipates the reader most obvious counter-argument and addresses it is far more persuasive than writing that ignores it, because the unaddressed objection sits in the reader head for the rest of the piece drowning out everything else.

The cheapest test remains giving it to someone in the intended audience and watching where they slow down. They will not always be able to say what is wrong. Where they pause is the information.

Editing is not one activity, and doing it as one is why it feels endless. Separating it into passes with a single job each is faster and produces better results.

The structural pass asks only whether the argument holds and the order is right. Nothing is polished; whole sections move or leave. This is the pass with the largest effect and the one most often skipped because it involves discarding work.

The clarity pass goes paragraph by paragraph asking what each one is doing. Paragraphs that restate, that hedge, or that exist because they were interesting to write get cut. Most drafts lose between ten and thirty per cent here and improve.

The line pass works at sentence level: compression, unnecessary qualifiers, passive constructions that hide who did what, and the phrases that take nine words to do four words of work.

The read-aloud pass catches what the others miss - rhythm, accidental repetition, and sentences that are grammatical and unsayable. It remains the single most effective editing technique available and it requires no expertise at all.

Then leave it overnight if there is time. Distance does something no technique replaces, and a piece that seemed finished at midnight rarely seems finished at nine.

The first paragraph decides whether the rest is read, and a large share of drafts spend it warming up.

The warm-up has a recognisable shape. A broad claim about the world. A definition nobody asked for. A sentence establishing that the topic is important, which the reader had already assumed by clicking. None of it is wrong and none of it is the piece.

The reliable repair is mechanical: delete paragraphs from the top until the first sentence is one that could not be the opening of any other article. That sentence is almost always already in the draft, sitting two or three paragraphs down, waiting for permission.

What works in an opening is a specific thing. A number that is surprising, a situation the reader recognises, a claim they might disagree with, or the conclusion stated flatly before it is argued. Each gives a reason to continue that a general statement of importance does not.

The academic instinct works against this and it is worth naming, because it is trained rather than natural. Building context before a claim is correct in a thesis and costs readers everywhere else, where the reader has not committed to finishing and will not.

The same applies to endings, where the failure is the opposite: summarising what was just read rather than saying what follows from it. A conclusion that only restates is a paragraph the reader has already had.