An SOP for Clearer Public Policy Writing

Skip to content Aaron McLoughlin The Systems Brussels Lobbyist

• About
• The Trusted Adviser Diagnostic
• Good Practice
• Publications

Skills · 16th August 2026

An SOP for Clearer Public Policy Writing

An SOP for Clearer Public Policy Writing Most advice about writing well is advice about sentences. Almost none of the failures I see in Brussels are sentence failures.…

An SOP for Clearer Public Policy Writing

Most advice about writing well is advice about sentences. Almost none of the failures I see in Brussels are sentence failures. A paper lands badly because it answered a question nobody asked, because the evidence underneath it was never there, or because eleven people held the pen. You cannot edit your way out of any of those.

So what follows is a process rather than a style guide: eight stages and five checkpoints, running from working out what good looks like to asking, three weeks later, whether it worked. It uses Barbara Minto’s Pyramid Principle for structure, Steven Kotler’s work on flow for the drafting discipline, and my own practice notes for the rest.

Two things prompted it. The first is that after years of reading position papers, consultation submissions and reports in this town, my honest conclusion is that clear public policy writing is close to forbidden in Brussels. The second is volume: the written repertoire of a public affairs practice runs to about forty separate products — the full list is at the end of this post — and any given week pulls several of them off the shelf. Without a process, each one gets invented from scratch.

Eight stages, five checkpoints

Stage What you produce Effort Checkpoint 1 What does good look like? The standard you will be judged against 5% — 2 Reader and brief Who it is for, their clock, the goal, the ask 5% — 3 Intelligence and research Sourced material, every claim traceable 25% — 4 Structure on the pyramid A one-page skeleton — no prose 15% 1, after this stage 5 Draft A complete, ugly draft 20% — 6 Edit in passes A clean draft that passes the checklist 25% 2, 3 and 4, after this stage 7 Send The deliverable — — 8 Did it work? One line in the log 5% 5 — this stage is the checkpoint

The proportions matter more than they look. Most weak papers are weak because stages 3 and 4 were skipped in the rush to have words on a page — and no amount of stage 6 rescues a paper built on the wrong so-what.

Three health warnings

Check for these before stage 3. All three are cheap to deal with at the brief and ruinous to discover in the draft — and none of them is fixable by writing better.

1. No objective data or evidence

What it looks like. The position rests on assertion, on what the client believes, or on one industry figure everyone has already seen. The opposing dossier’s numbers go unchallenged because you have nothing of your own to put against them.

Why it is fatal. Officials and expert committees discount unevidenced advocacy almost automatically. The paper then either overstates and gets found out, or hedges so heavily it says nothing. No amount of editing rescues an empty evidentiary position.

What to do. Say it in writing at stage 3. Name the evidence needed, who could produce it, and by when. Then commission it, narrow the ask to what the evidence actually supports, or advise against filing at all. Never let a gap in the evidence become a flourish in the prose.

2. Drafting by committee

What it looks like. The text goes round a working group. Each member adds a qualifier, nobody has authority to cut, and the ask degrades into an invitation to consider. Association files are the natural habitat.

Why it is fatal. Every round adds words and removes edge. A committee converges on the sentence nobody objects to, which is almost always the sentence that says nothing.

What to do. Split the two jobs the group is actually for. Take the skeleton to them at Checkpoint 1 — governing message, the one to three points, the ask — where disagreement is cheap and genuinely useful. Then one named person holds the pen. Comments come back as objections and questions, not as rewritten sentences. Technical experts list the points and the evidence; they do not draft and they do not wordsmith.

3. Bonkers asks

What it looks like. The client wants something outside the procedure, outside the timetable, or outside what any official could actually deliver — reopening a closed consultation, an exemption with no legal base, a meeting to relitigate a settled point.

Why it is fatal. It burns the meeting, it burns credibility, and it is disclosable. Officials remember who wastes their time, and the next reasonable ask is heard differently because of it.

What to do. Separate the objective from the ask. The objective is usually legitimate; it is the requested action that is not. Offer the achievable version — right procedural moment, right interlocutor, proportionate request — and state what the original would cost. If the client insists, record the risk in writing before it goes.

1. What does good look like?

Start from the standard, never from a blank page. Establish what a good one of these looks like before you think about this one at all.

• The template for this document type. Start from a house skeleton if one exists. Copying the closest and adapting beats inventing a shape.
• Your organisation’s own written standards. Read the bar before you write, not after. If a checklist is going to be applied to this document, you should know its contents before the first sentence goes down.
• The best previous example. The last one of these that actually landed. Stage 8’s log is where you find out which ones those were — which is the whole reason for keeping it.

If nothing exists for this type, say so now rather than discovering it at the review. Draft it anyway, and the finished document becomes the template. That is the cheapest moment there will ever be to create one.

If the template exists but is wrong for this reader, deviate deliberately and record why. Unrecorded deviation is how a house standard quietly decays into a folder of unrelated documents.

Done when you can describe what a good version of this document looks like — its shape, its length and the bar it will be judged against — before you have written a word of it.

2. Know the reader before anything else

Nothing downstream can be decided until this is settled. Four questions, answered in writing, before you open any material.

• Who, by name. Not “the client” — the individual who will open it. A trade association secretariat fluent in the file and a company general counsel who has never heard of the committee in question are not the same reader and cannot take the same document.
• What do they already know? Their expertise on this specific file sets how much you explain and how much jargon survives.
• What is their clock? Two minutes is the realistic default. Five is generous. Ten essentially never happens. Length follows from their clock, not from how much you know.
• What is the goal — inform, alert, call to action, or persuade? And what, in plain words, do you want them to do or decide?

The honest question. If there is nothing for them to do and nothing for them to decide, ask whether the document should exist at all. One that exists because something happened, rather than because someone must act, is junk mail with a letterhead.

Done when you can complete this sentence: “This is for [name], who has [n] minutes, and when they finish it they will [do the thing].”

3. Intelligence and research — before you sit down

The writing block is for writing. Finding out is a different activity, and doing it inside the drafting block is what turns a two-hour draft into a five-hour one. Separate them in the diary.

• Get everything in front of you first. The recording, the proposal, the amendments, the opinion, the dossier, the register entry. If you have to go looking mid-draft, you have left flow and you will not get back in cheaply.
• Traceability as you gather. Against each key point, note where it came from — the debate timestamp, the paragraph number, the register entry, who said it and when. This protects accuracy, survives challenge, and makes the document far faster to update when the file moves.
• Intelligence, not just documents. What was said informally, who has moved, who is wobbling, what the competing organisation is briefing. This is usually the part the client cannot get anywhere else — and the part that justifies the fee.
• Flag the unknowns explicitly. Resolve them before drafting where you can. Where time does not allow, say in the document that the point is being checked and will be updated. Never paper over a gap; it will be the one thing the client asks about.

Done when you can answer “how do I know that?” for every claim you intend to make.

4. Structure on the pyramid

Build the skeleton before writing a sentence of prose. This is the stage that most often gets skipped and most often causes the rewrite.

Get to the governing message

Work the four moves in order. They take ten minutes and they decide whether the document has a reason to exist.

• Situation. What the reader already accepts as true. Uncontroversial, brief.
• Complication. What has changed. The development, the opinion, the vote, the deadline.
• Question. The question that change puts in the reader’s head. If the complication produces no question for this reader, stop — you have an event, not a document.
• Answer. Your response to that question. This is the governing message. It goes at the top, in one or two lines, and often doubles as the subject line.

Then build down

• Answer first, always. A reader who stops after the first paragraph should still have the point. This is the one rule that does not flex for client, content or document type.
• Each level answers the question the level above provokes. If a section answers a question nobody asked, cut it.
• Group, don’t list. Supporting points should be mutually exclusive and collectively exhaustive — no overlaps, no gaps. Three is the natural ceiling.
• Order each group deliberately. Chronological for procedural steps, structural by institution or actor, or by importance with the biggest risk first. Pick one logic per group and hold it — mixed ordering reads as muddle.
• Prefer grouping to argument chains. A busy reader who takes only the top line of each group should still end up in the right place. Long deductive chains fail the scanner.

Only then slot in the substantive elements: what happened and its source, relevance to this reader, recommendation and actions, next steps, background or annex at the bottom.

Done when the skeleton fits on one page, contains no prose, and you could hand it to a colleague who would draft roughly the document you intend.

Checkpoint 1 — Is the structure right?

• Why here. This is the cheapest possible moment to be wrong. A wrong so-what costs ten minutes now and a full rewrite later.
• Who to ask depends on who owns the position. Where one person’s view decides the document — the partner, the client lead, the client — take it to them. Where a working group owns the position, take it to the group: this is the moment for collective input, and, per health warning 2, the only moment for it.
• Ask precisely. “Is this the right so-what for them, and is this ask the one they would act on? What am I missing, and who will push back?” Do not ask “does this look OK”.
• Where a client expects a finished product, run this checkpoint internally instead — but run it.
• The guardrail. Never draft more than a page of prose against an untested structure.

5. Draft

The place — this is not a detail

Complete concentration is the first and most important condition for flow, and everything else in this stage depends on getting it. The single biggest determinant is where you sit.

• A quiet room with the door shut. Not the open-plan floor, not a corner of it with headphones on. Open-plan offices are, as a rule, terrible for this kind of work — the cost is not the noise itself but the constant low-grade expectation that someone may turn to you.
• Unreachable, by arrangement. Devices disabled the night before, not in the moment. Email closed, phone out of the room, notifications off. Willpower loses to biology at six in the morning.
• The same room, every time. Same time, same place, every day. The consistency is doing real work: the room becomes a cue, and you stop spending the first twenty minutes arriving.
• Tidy it before you leave it. Mess is cognitive load you pay for tomorrow morning.
• Silence is not the requirement — being unreachable is. A café can work where an open-plan office cannot, because nobody there can put a question to you. Find the place that works and then stop changing it.

Then write

The goal for the block was written the previous evening: which section, and what the reader should feel at the end of it.

• Vomit first. Get the bones out — who, what, when, why, what it means for them. Ugly is the point. Editing while drafting is what makes a paper take three days.
• Hold the skeleton. The draft serves the tested structure. If the writing wants to go somewhere else, that is information — but it means returning to stage 4, not quietly abandoning it.
• Open each session by editing yesterday’s words. It warms you up and builds the first quality pass into the rhythm.
• Stop while you are still fired up, mid-sentence if need be. You bank the momentum and return to an unfinished thought rather than a blank page.

Done when the draft is complete, ugly, and structurally faithful to the skeleton.

6. Edit in passes

Five passes, in this order. Each does one job; trying to do them at once is why editing sprawls.

1. Structure. Does it still match the skeleton? Is the answer genuinely first? Does anything below contradict what is above?
2. Relevance. Would this paragraph read identically if it went to a different client? If yes, it is generic — make it specific or cut it. This is the single most common failing in an otherwise competent document.
3. Language. Jargon out. Short paragraphs, short sentences. Emphasis sparing. Length matched to their clock, not to your research.
4. Aloud. Read the governing message and the ask out loud. What stumbles gets rewritten. The sentence that trips you is the one that gets quoted back out of context.
5. Honesty. Does it read as advocacy dressed as analysis? Would the reader think you have selected the facts that suit your line? Say what cuts against you, briefly, in your own words — before someone else does.

Then grade yourself, A to F, on five things: clarity of the essence, accuracy, intellectual honesty, relevance to this reader, actionability. For anything below an A, name what would lift it. Then role-play the sceptical reader and ask what makes them stop reading, and what would make them act.

Done when the pre-send checklist passes without argument — then put it through the three checkpoints below, in order.

Checkpoint 2 — Does it follow the guidelines?

This is a compliance pass, not a matter of taste: does the document do what your own written guidance says a document of this kind must do? I run it through an AI assistant loaded with our house standards. A colleague working from the same checklist does the same job more slowly.

• Give the reviewer the brief, not just the draft. Hand over the stage-2 brief — reader, expertise, clock, goal, ask — alongside the text. Without the brief, any reviewer can only judge the writing generically, and generic feedback is the one thing you do not need.
• What should come back. A grade against the house bar with what would lift each score; the sceptical-reader role-play; the honesty test; and the junk-mail test — restate the one to three reasons this reader should open it and the specific actions you want. If those are thin, so is the document.
• What this pass does not do. It does not verify facts, and it must never invent them. Where the source material does not support a point, the reviewer flags the gap rather than filling it. This is a second brain to think against, not a ghostwriter, and it does not get the last word.
• Run it before Checkpoint 3. It is quick and it costs nobody else’s time. Spend a colleague’s attention on a draft that has already cleared the house rules.

Checkpoint 3 — Does it sound right?

• Someone else reads it aloud to you. Not you reading it to yourself — a colleague reads, you listen, and you do not follow along on the page. Hearing your own words in another voice is the fastest way to find the sentence that does not work.
• What you are listening for. Where they stumble. Where they have to re-read a line to get the sense of it. Where the emphasis lands somewhere you did not intend. Where they pause because they expected the point sooner.
• Two questions afterwards, in this order. “What do you think I am asking them to do?” and “When did you first know what this was about?” If the first answer is not your ask, or the second is not the first paragraph, the document has failed at the top and no amount of polishing lower down repairs it.
• Anyone can do this. They do not need to know the file — arguably better if they do not. This is a test of the writing, not of the substance.
• Under real time pressure, read the governing message and the ask aloud to yourself. Never skip it entirely: it is the cheapest quality check in the whole process.

Checkpoint 4 — Is it verified?

• Never send on autopilot. Whatever produced the draft — you at 6am, a junior, an AI — someone senior verifies before it goes.
• Personally check every date, every name and political group, every procedural step, every number, every citation, every hyperlink.
• State the gaps. Hand over the clean draft plus a short note of what is uncertain and what is still being confirmed. Explicit gaps build trust; discovered ones destroy it.
• If someone else signs it, they get the same package — draft, checklist, uncertainties — not just the file.

7. Send

• Confidentiality marking on client deliverables unless told otherwise.
• Hyperlink sources inline — the development, the dossier, the recording, the register entry — rather than pasting raw URLs into the body.
• Short paragraphs, short sentences, and a clearly labelled clarifications or ask block.
• Make the action unmissable. A scanner should see what to do without reading every line.

8. Did it work? — Checkpoint 5

The stage almost every writing process omits, and the only one that improves the next document. Run it two to three weeks after sending, or straight after the event it was written for.

• Did they read it? A reply, a forward, a line quoted back at you in a meeting. Silence is data too.
• Did they do the thing? If not, was the ask wrong, unclear, or badly timed? Those are three different fixes.
• Did it hold up? Anything overtaken, contradicted, or challenged since? Note it — accuracy failures compound reputationally.
• What did they ask you afterwards? The highest-value signal in the whole process. A question asked after reading is a question the document failed to answer. A question that recurs across readers becomes a fixed section in the template.

Capture it in one line per document in a running log — date, reader, type, did it land, what to change. Review the log quarterly and look for the pattern, not the incident. An ask that never lands needs reframing, not repeating.

Pre-send checklist

• Built on the house template for its type — or the deviation is deliberate and recorded.
• Essence and solution summarised at the start.
• One to three key things, no more.
• Free of jargon this reader will not know.
• Right length for their clock.
• Headers and sparing emphasis guide a scanner through.
• Clear, accurate, honest — in that order.
• Every claim traceable to a source.
• Checked against the house guidelines, and the grade addressed.
• Read aloud by someone else, and it survived.
• Actions highlighted and unmissable.
• Gaps and uncertainties stated, not hidden.
• If there is nothing relevant and nothing to do — do not send it.

The register of written products

Thirty-nine separate products, in six families. The families matter more than the count: they tell you which standards apply, and — in family C especially — that you should assume every word becomes public.

A. Intelligence and advice, to the client Issue update · Advisory note · Debate report or read-out · Regulatory update · Proposal summary · Amendment summary · Policy memo · Monitoring report · Meeting summary · Evaluation

B.Strategy and planning Strategy proposal · Lobby or campaign plan · Stakeholder map · Red lines brief · Workshop and action register · Elevator pitch

C.Advocacy documents — assume every one becomes public Position paper · Persuasive memo · Letter to the Commission · Letter of support · Amendment text · Consultation feedback, in the Commission’s public feedback window · Shadow impact assessment · Regulatory agency submission, to an expert committee · Regulatory consultation response, on an agency’s own dossier · Op-ed or bylined article

D.Meeting machinery Meeting request · Pre-meeting briefing · Agenda · Leave-behind · Post-meeting summary · Internal read-out

E.Presentation, teaching and correspondence Policy presentation · Training material · Guide or SOP · Email

F.Practice and compliance New business proposal · Scope of work · Transparency Register entry

← The 7 Ps Model for European Public Policy Lobbying

Leave a Reply

Your email address will not be published. Required fields are marked *

Comment *

Name *

Email *

Website

Notify me of follow-up comments by email.

Notify me of new posts by email.

A note from Brussels on EU trends and what seems to work.

An infrequent letter. No noise.

© Aaron McLoughlin · Brussels RSS