Peanuts In My Pocket

I use ChatGPT to help improve my writing. I see it as a tool that offsets some of my weaknesses. One of my core challenges is that my primary education didn’t leave me with strong writing or spelling skills. I’ve had to build those skills as an adult. While I’ve never had trouble generating ideas or writing content, editing remains an underdeveloped skill. I have a large vocabulary, but I still struggle with spelling. Simply put, I need an editor to help refine my work.

I used to write another blog, where I published six or seven thousand posts over the course of a decade. I’m now reviewing those older entries. I take a post from 2004, capture it as-is, and then run it through ChatGPT. The revised version often reveals numerous spelling mistakes, grammatical errors, and structural issues. I also tend to repeat myself, did I mention that I repeat myself? Apparently, I do.

I can see my writing improving over time. Posts from 2006 show fewer errors than those from 2004. Back then, I used spelling and grammar checkers, but they weren’t enough to catch everything. Fortunately, a retired editor with strong skills offered to help. I’d run my more important posts by him, and he would give feedback that not only corrected mistakes but also improved overall quality. The key takeaway from working with a human editor was that while I make many mistakes, I have a clear voice and can write well.

ChatGPT often gets criticized on social media for its writing, but I rarely see commentary about its editing capabilities. I’ve experimented with using it for fiction. I gave it a chapter from a book I’ve been working on for years and asked it to suggest what might happen next and to write the scene in my voice. The result was dull and lifeless—but technically passable. The AI’s suggestions made sense for the character, but I’m not sure anyone would want to read the scene it produced.

Today, I mostly use AI writing for filler. For example, if I need to cite a fact about cats impacting songbird populations, I’ll ask:
“Please write a three-sentence paragraph describing the risks cats pose to songbird populations. Include a reference.”

“Free-roaming domestic and feral cats are a major threat to songbird populations, killing billions of birds each year in the United States alone. These predation pressures can significantly reduce local bird numbers, especially for species already facing habitat loss. According to Loss et al. (2013), cats are the leading human-caused source of direct bird mortality in the country, posing a serious conservation challenge.”
Reference: Loss, S. R., Will, T., & Marra, P. P. (2013). The impact of free-ranging domestic cats on wildlife of the United States. Nature Communications, 4, 1396.

I’ll take that information and include it in my Obsidian-based PKM (personal knowledge management) system. I save both the prompt and the output, then add more research. I’m in the “trust but verify” camp when it comes to using ChatGPT. These days, I find it mostly reliable.

I may use the whole paragraph, a single sentence, or rewrite it in my own words while still citing the original study. To me, this feels like having a smart research assistant on call. But is it writing my essay? No. It’s just another way to gather and synthesize information. It’s more efficient than using a search engine and certainly easier than a trip to the library.

What gives me the most value, though, is the editing process. I paste the full article into ChatGPT, but only after I work through my editing checklist.

Editing Checklist

• Misspelled words (spellchecker)
• Formatting
• Common grammar errors (grammar checker)
• Read aloud (to check readability)
• Verb tense
• Run-on sentences

Each of these checklist items addresses areas where ChatGPT has flagged sloppy errors in the past. By doing two or three rounds of self-editing, I reduce the number of issues the AI finds. Still, I’ve yet to write a post over 500 words without at least a handful of lingering mistakes.

The next step is what I call “a blow to my ego.” ChatGPT gives me detailed feedback on every mistake I’ve made, followed by a cleaned-up version of the text. This is where Obsidian’s split-screen mode comes in handy: I compare the original version to the AI-edited version. I’m looking for changes in meaning, shifts in tone, and any alterations to context.

I do have one pet peeve: I hate ChatGPT’s overuse of em dashes. I use them occasionally, but rarely. Seeing them show up repeatedly in the edited version drives me a little crazy, especially since I explicitly ask the AI not to use them. Aside from that, the only other recurring issue is when it adds ideas or concepts I didn’t ask for. It pays to carefully read and compare outputs.

Because all of my blog essays are part of a single project within ChatGPT, it sometimes pulls in related content from other posts. I almost never accept these additions. I’m not sure why this happens, especially since I’ve made my preferences clear, but it’s a good reminder to stay vigilant.

One of the most enjoyable parts of this process is asking ChatGPT what it thinks of my essay, or of my collection of essays on a topic. The analysis it provides helps me dig deeper into the subjects I care about, while also giving me new insight into my writing habits and patterns. I highly recommend using it for editing.

Notes

• I made 4 spelling mistakes in this essay.
• The AI edit removed 4 redundant sentences.
• The AI edit changed 5 sentences to improve readability. I accepted 3 of 5 and rewrote the other two.
• The AI insisted on adding EM Dashes in five spots. All but one removed.