Emails, Proposals & Writing

The hardest part of most writing is starting. Staring at an empty reply to an awkward client email costs more time and willpower than the writing itself. An LLM removes that cost: give it the situation and a few bullet points of what you want to say, and it returns a competent draft in seconds. Reacting to and fixing a draft is far easier - and faster - than producing one from nothing. That single shift, from generate to edit, is the whole everyday win.

The key is to feed it your points. The model cannot know that the delay was the client's late decision, or that you want to sound firm but not defensive - so tell it. A good draft prompt is short but loaded: the situation, the recipient, the points to hit, the tone, and the length.

Weak:  "Write an email to my client about the delay."
Better: "Draft a short, warm-but-professional email to a residential
         client. Context: the kitchen is 2 weeks late because their
         tile selection changed twice. I want to: acknowledge it,
         explain plainly without blaming them, give the new date
         (14 March), and reassure. Keep it under 150 words."

The better prompt gives the model everything it needs to be useful and nothing to invent. You will still edit the result - but you are editing, not staring at a blank page.

There is a psychological dimension worth naming. Much dreaded writing is dreaded because it is emotionally loaded - the apology, the firm push-back, the awkward money conversation - and the blank page magnifies the dread. Getting a neutral, competent draft on the screen defuses that: you are no longer inventing the words for a hard message, only improving them, which is a far calmer task. Many designers find the biggest benefit of AI here is not the minutes saved but the friction removed - the difficult email that would have sat in drafts for two days now goes out the same afternoon, in better shape than if you had wrestled it alone.

One of the most practical things an LLM does is adjust tone. The same content can be made warmer, firmer, more formal, or plainer on request - useful when you have written something in frustration and need it to land as calm and professional, or when a proposal reads as stiff and you want it human. Make this warmer without losing the firmness. Cut this by a third. Make this suitable for a formal fee proposal. Simplify the jargon for a non-technical client. These are quick, high-value edits.

It also flexes across document types. A fee proposal needs structure and quiet confidence; a project narrative for your website needs evocative but honest prose about the design intent; an award submission needs to hit specific criteria persuasively within a word limit; a cover letter needs to connect your work to a specific opportunity. In each case, tell the model the genre, the audience and the constraints, and it will shape the register accordingly - a strong starting point you then refine.

A particularly good use is tightening. Paste your own rambling draft and ask the model to cut it by a third without losing meaning, or to make it clearer and less repetitive. Here the facts and voice are already yours - the model is just a sharp editor - which makes it both safe and genuinely improving. Many designers find AI more valuable as an editor of their own writing than as a first-drafter.

The real risk in AI-assisted writing is not error - it is blandness. Left unedited, LLM prose drifts toward a recognisable house style: smooth, over-long, faintly corporate, sprinkled with hollow phrases (we are excited to leverage, in today's fast-paced world). Clients and juries increasingly recognise this texture, and it reads as generic and impersonal - the opposite of what a design practice wants to project. A proposal that sounds like every other AI proposal actively hurts you.

So the discipline is to treat the draft as raw material and make it yours. Cut the filler. Put back the specific detail only you know. Restore your own rhythms and turns of phrase. If you have a distinctive voice, feed the model samples of your past writing and ask it to match your style - it does this surprisingly well - but always do a final pass by hand. The last edit is yours, always.

And do not outsource judgement about what to say, only help with how to say it. Whether to concede a point to a client, how to price a job, what to promise - those are yours. The model can phrase your decision well; it should not make it.

Finally, the low-stakes label has one exception: facts. An email that invents a date, a proposal that misstates a fee or a deliverable, a narrative that claims a material you did not use - these are small to write and costly to send. Skim every draft for any specific claim and confirm it is true before it goes out. The writing is fast and forgiving; the facts inside it are not.

Put together, everyday writing is one of the highest-frequency, lowest-friction AI wins in practice: it touches your week constantly, the time saved is real, and the verification is quick. If you adopt only one AI habit from this whole module, a fast draft-then-edit loop for routine correspondence is the one that pays off daily.

Two cautions keep it professional. First, confidentiality, as everywhere in this module: proposals and client emails carry names, fees and commercial detail you should not feed into a consumer tool that may train on it. Use a business tier or no-training setting, or genericise before you paste (Module 9 has the detail). Second, do not let fluency outrun honesty. AI makes it effortless to write a confident, polished promise - so it is easy to over-claim in a proposal or narrative. Keep your writing as truthful as your work; a beautifully worded commitment you cannot keep is worse than a plainer one you can.

The throughline of the whole module holds here in its gentlest form. On specs and compliance, the human-in-the-loop guards against real danger; on everyday writing, it guards against blandness, over-claiming and the occasional invented fact. Either way the shape is the same: AI drafts, you judge, you own. Beat the blank page, set the tone, tighten the prose - then make it sound like you and make sure it is true. That is the everyday writing win, kept professional.

The same messages come up again and again in practice: the project update, the payment reminder, the proposal cover note, the polite chase, the thank-you after a pitch. Rather than re-explaining the situation to the model each time, build a small personal library of prompts you reuse - a starter for each recurring message that already carries your tone, your structure and your standing rules (be concise, no hollow phrases, flag anything you would need me to confirm). You paste in the specifics of the day, and the draft comes back already close to your voice.

This is the writing equivalent of a master spec: reusable, house-styled scaffolding that makes every instance faster and more consistent. Over a few weeks it quietly compounds - your correspondence gets more uniform in quality, new team members can draw on the same starters, and you spend your editing attention on the message that actually needs care rather than reinventing routine notes. It also scales cleanly into the custom assistants of Module 8, where these prompts can become a shared, always-on writing helper tuned to your practice's voice.

Two guardrails keep the library healthy. Keep the prompts generic - hold client-specific and confidential detail out of the stored template and add it only in the moment, ideally in a business-tier tool. And revisit your starters occasionally: prune the phrasing that has started to sound like everyone else's AI, and feed in fresh samples of your best recent writing so the voice stays current and unmistakably yours. A prompt library is a living asset, not a set-and-forget macro - tend it, and everyday writing becomes both faster and better, month after month.