That is precisely what I’m doing, or at least, attempting to do. 
I believe you are correct on both counts. This is why my project instructions are explicit: “Claude does not generate, rewrite, or expand prose. Claude polishes, corrects, formats, inserts requested material at breadcrumb locations, and structures paragraphs. The voice, argument, and substance remain the author’s throughout. When in doubt about a breadcrumb’s intent, ask rather than assume.”
I am striving to ensure that AI functions like an editor, not my ghostwriter. My intent is that it function no differently than when I pass my reports or articles to my EA or a book or chapter draft to the editor for review and markup for changes. I may not adopt all the suggestions, but in many instances they are helpful and improve my writing. Neither my EA nor an editor writes for me. That is precisely what I want from the AI.
The struggle to write, combined with significant amounts of reading across a variety of genres, makes one a better writer. I do not put my name to something I did not write.
I appreciate your suggestions above, particularly about writing to “Andrew.” One approach I use is telling Claude that I am writing as a mentor to a mentee.
That said, I would not feel comfortable intentionally adding mistakes “so it looks like I wrote it.” If I wrote it, I do not need to add intentional mistakes. However, on the flip side, I now hesitate to use em dashes where I naturally would, for the opposite fear—that readers will assume AI wrote the text and I did not.
I was in the middle of writing a long post in very short sentences to make exactly this point! Gary Provost makes it much better than I would have done…
The problem with advice like this (for humans, never mind for statistical text generation machines) is that it’s not really intended to be taken literally. No good human writer has ever followed this rule precisely, except in parody.
It’s meant instead as training wheels for young and inexperienced writers to help them the worst common errors, not to prevent good writers from expressing their thoughts in the most appropriate way for the text. That’s because by concentrating on individual sentences and paragraphs it completely ignores one of the most important tools that any author has to convey their meaning and convince their readers: the overall rhythm and flow of the entire text.
Also: it’s a good idea to prefer the active voice, but there are many times when the passive conveys your meaning more precisely, or fits in better with the overall rhythm.
1 Like
Absolutely: writing is a tool for thinking. But writing is also a tool for communicating thinking, and for that end, a good editor can make all the difference. I say this as someone who has had the great gift of working with an editor who, in the process of sharpening my writing, also sharpened my thinking.
I think @Bmosbacker sums it up nicely. Used with care and discretion, AI can do some of what an able human editor can do, especially if you’re willing to put effort into your prompts, and, what I think is most important, engage in an iterative back-and-forth about the text itself.
I will add that there are times when using AI as a ghost writer is absolutely appropriate. It’s a useful tool for those situations where you need to turn a list of disorganized, group-generated bullet points into a something like a job description or process document—i.e., when you have to turn a bunch of idiosyncratic voices into one clear and neutral voice.
In lieu of a massive prompt, if you are using Claude desktop or other local AI, point the AI at a local folder containing what you think are your best compositions, and tell the AI to read and learn from the material in that folder. If this is part of a project, then that local learning for the AI can be more useful than trying to tell the AI what your personal elements of style are. [/quote]
That is a great suggestion, thanks!
FWIW, this very easy if you’re doing all this in DEVONthink 4.
Question, I use DT extensively but I’m not sure how I’d use it as you described, can you elaborate?
The Gary Provost text is a good one! I’ll add:
Short and long sentences do different things. Short sentences make key points or recap them, and trade in swift action, jokes and little swerves in thought. Long ones take readers on a mental tour, offer a rambling inventory or knead and stretch out a thought like a pizza chef working dough. Short sentences give your brain a rest; long ones give it an aerobic workout. Short sentences imply that the world is cut and dried; long ones restore its ragged edges. Short sentences are declarative and sure; long ones are conditional and conjectural. Vary your sentence length and you mirror the way the mind works, veering between seductive certainty and hard-won nuance.
Joe Moran, from First You Write a Sentence
1 Like
That bit about adding 2 spelling mistakes was mostly tongue-in-cheek, but - to be fair - that would actually look more like my “voice” and have few spelling mistakes than my average.
1 Like
Indeed, I’d have a number of typos and other errors if I did not have my work checked carefully. Aside from my EA and AI, I have found having the text read back to me helps me catch those pesky errors!
This is in regard to Haiku 4.5, which is what I use. I point the AI at a group and its subgroups that contain many writing samples for my own articles. In DEVONthink, the AI has no persistent memory between chat sessions, but it can be asked to read the writing samples, form an understanding of how my style works, and write its own aide memoir which I can save and point the AI at the next time I want it to reflect my voice in a draft.
Haiku’s somewhat answer to this suggestion was:
Absolutely-and that’s actually a clever workaround to a key limitation!
Here’s the full workflow:
- Analyze your writing samples from the specified group
- Create a document (using
tool_create_item) that summarizes your writing style-capturing patterns like tone, vocabulary, sentence structure, punctuation habits, etc.- Save it to your database with a clear name like “My Writing Style Guide” or similar
- In future conversations, you can ask me to reference that saved document when writing something for you
This solves an important problem: while my internal learning from your samples only lasts within a single conversation, the document becomes a persistent reference in your database. So in a new conversation months from now, you could say “Here’s my style guide [link/reference]-please write this email in my style” and I’d have the guidance right there.
The document becomes your aide memoir, but it’s one I can actually access and apply in future conversations, rather than relying on memories I won’t have.
So yes, this is absolutely doable and would be an effective system.
And, of course, you can just tell the AI to consult the relevant group containing your writings, every time, and / or you can edit the AI’s reference document if you want.
1 Like
Loosing my writing voice with AI?
AI is super cool, but it won’t take my words away! It’s like when I ask Mommy to help me build a tower with blocks. She helps, but I still pick the colors and where to put them! AI can help me write stuff, like maybe tell a story about a fluffy bunny, but I still get to say how the bunny hops and what yummy carrots it eats. My words are my words, and AI just helps me make them even more fun!
Gemma-3-12b
First, my apologies for the delayed response; it has been a busy two days, and I will be at the office today until 8:30 or so. Second, thank you for taking time to share this process. I had no idea that something like this could be done in DEVONthink. Very impressive, and it gives me some ideas. Thank you again for the thoughtful and helpful response. Much appreciated!
It was still good advice [to write short sentences] Cindy. If you’d told him to vary sentence length, I imagine it would have flummoxed him.
I think I have something in my customer instructions that says to also vary sentence length, but every so often I shout at it WRITE LIKE A HUMAN NOT AN EFFING COPYWRITER and we get back on track.
Rewriting till my friend Andrew would like it … that’s made a huge difference. Andrew prefers full sentences, for instance.
Style should match purpose. In technical writing, where it’s important to be clear and concise so the user can easily follow what might be complicated instructions, short sentences are often helpful.
But in creating a persuasive argument, too many short sentences can be overly assertive (short sentences tend to be declarative) or even seem condescending.
2 Likes
I think the key principle is not to absolutize a principle.
There is a place for complex sentence structure, and there is a place for short sentences. Generally, I find mixing them helpful. Because of how I was taught, I struggled with the idea of a single-sentence paragraph. I was taught that a paragraph should always comprise two or more sentences. I am over it now.
So much of the human existence is nuanced, I guess I’m not surprised the best conclusion is “it depends”.
Definitely.
I hope you don’t think I’m disagreeing with you Lisa, but my angle is that most people aren’t good at writing, and writing shorter sentences is an easy way to get better and faster at writing.
That’s based on my experience - I spent too many years trying to write beyond my skill level, and my “style” was 
.
You’re a good writer, with a lot of skill, so you don’t need my advice!
Disagreeing with me is fine. It is in fact a practice I encourage.
I think most people are better at writing than they think they are.