Last month I had a deadline, zero energy, and a 2,000-word article due by morning. I opened Claude 3.5 Sonnet, typed a rough prompt, and had a solid first draft in about four minutes. I cleaned it up, rewrote maybe 40% of it, and hit publish on time.
That's the good version of this story.
The bad version is the time I let ChatGPT write an entire product comparison piece without double-checking anything. Two of the prices were wrong, one product had been discontinued, and a reader in the comments (politely) destroyed me. That was a fun afternoon.
So yeah — I've got opinions on this. Let's get into it.
The Tools I Actually Use
First, some context. The main players right now are ChatGPT (GPT-4o) from OpenAI, Claude 3.5 Sonnet from Anthropic, Gemini 1.5 Pro from Google DeepMind, and then a bunch of purpose-built writing apps like Jasper, Copy.ai, and Writesonic — most of which are just wrappers around GPT-4o or Claude anyway.
I use Claude most for long-form writing. GPT-4o for quick tasks and when I need code or structured data. Gemini 1.5 Pro when I'm working with long documents because its context window is massive — up to 1 million tokens as of early 2025, which is genuinely wild.
Pricing-wise: ChatGPT Plus is $20/month, Claude Pro is $20/month, and Gemini Advanced (which gives you 1.5 Pro) is $19.99/month through Google One. If you're on a budget, all three have free tiers — they're limited, but usable.
Where AI Writing Tools Are Genuinely Useful
Getting Past the Blank Page
This is where I think AI earns its keep more than anywhere else. Writer's block is real, and sometimes you just need something — anything — to react to. Typing "write me a rough draft intro for an article about X" gives you something to push against. You might rewrite every sentence, but at least you're rewriting instead of staring.
I use this constantly. I'd say 60% of my articles start with an AI-generated outline or rough intro that I then tear apart. It's not laziness — it's just using the tool for what it's actually good at.
Editing and Rewriting
Paste a paragraph you're not happy with and say "make this tighter" or "this sounds too formal, fix it." Claude is especially good at this. It doesn't just rephrase — it actually seems to understand tone, at least well enough that the results are useful most of the time.
I spent way too long figuring out the right prompts for this, but honestly the simplest ones work best. "Rewrite this like a human who's slightly tired but still knows what they're talking about" sounds ridiculous but gets surprisingly good results.
Research Summaries and Brainstorming
I use Gemini 1.5 Pro to summarize long PDFs, research papers, and documentation. Drop a 40-page report in there and ask for the three most important points. It's not perfect, but it's faster than reading the whole thing before I know if it's even worth reading.
For brainstorming angles on a topic, all three tools are solid. I'll ask for 15 possible article angles on a subject, pick the two or three that aren't generic, and go from there.
Repetitive Content at Scale
Product descriptions. Meta descriptions. Email subject line variations. FAQ sections. This kind of content is genuinely tedious to write from scratch, and AI handles it well because there's not a lot of nuance required. If you're running an e-commerce site with 200 product listings, AI tools will save you a ton of time here.
Where AI Writing Tools Actually Hurt You
Factual Accuracy — This One's a Big Deal
Here's the problem nobody talks about enough: these models don't know they're wrong. They'll state something confidently and incorrectly without any hesitation. GPT-4o's training data has a cutoff, and even with web browsing enabled, it can get things wrong. Claude 3.5 Sonnet is better about saying "I'm not sure" but it still makes stuff up sometimes.
Prices, release dates, product specs, statistics, company names — all of this needs to be verified manually. Every single time. I learned this the hard way (see: the product comparison disaster I mentioned earlier).
"The model doesn't have beliefs. It has patterns. And patterns can match the shape of a wrong answer just as easily as a right one." — something a researcher at Anthropic said in a 2024 interview that stuck with me.
If you're writing about anything technical, financial, medical, or time-sensitive, you can't skip the fact-check step. Period.
Your Voice Disappears
This is the sneaky one. You start using AI to write more and more of your content, and slowly your actual voice gets diluted. The stuff starts sounding the same. Smooth, competent, soulless.
Readers notice. They might not be able to say exactly what's off, but they feel it. The specific observations, the weird tangents, the opinions that only make sense given your particular background — that stuff is what makes writing worth reading. AI can't replicate it because it doesn't have your life experiences.
I've read tech blogs that clearly went all-in on AI content and the comment sections dried up. People stopped engaging because there was nothing human to engage with.
The SEO Trap
There's a whole industry of people using AI to spin out massive amounts of mediocre content and hoping Google ranks it. And look — it worked for a while. Then Google's Helpful Content updates (particularly the big one in September 2023 and the follow-ups throughout 2024) absolutely wrecked sites doing this at scale.
HouseFresh, a small independent review site, published a detailed breakdown of how they were being outranked by AI-generated garbage on major publishing sites — and then watched those same sites get hammered by algorithm updates. It's a cycle that punishes the people trying to do it cheap and fast.
Ranking in 2025 and into 2026 requires actual expertise and original information. AI can help you write that content, but it can't generate the expertise itself.
Over-Reliance Atrophies Your Own Skills
I'll be honest — there were a few months where I leaned on AI way too hard, and I noticed my own first-draft writing got worse. It's like using a GPS for every drive and then struggling to navigate somewhere without it. The muscle weakens when you stop using it.
If you're a professional writer, or trying to become one, this is worth taking seriously. Use the tools, but also make sure you're still writing things yourself, from scratch, regularly.
How to Actually Use These Tools Well
Think of AI as a First Draft Engine, Not a Finished Product Machine
The best workflow I've found: I write my own notes and rough thoughts first — sometimes just bullet points, sometimes a messy paragraph dump. Then I give that to the AI and say "turn this into a structured draft, keeping my voice and these specific points." Then I rewrite heavily.
The output is maybe 30-40% AI-originated words by the time I'm done. But I got there faster than starting cold, and it still sounds like me.
Prompt Specificity Makes a Huge Difference
"Write an article about wireless earbuds" gets you garbage. "Write a 300-word intro for a hands-on review of the Sony WH-1000XM5, written in a casual but informed tone, for readers who already know what noise cancellation is and don't need it explained" — that gets you something useful.
The more context you give, the better. Mention the audience, the tone, the format, what to avoid, what to emphasize. Treat it like briefing a freelance writer who's smart but knows nothing about your specific situation.
Use Different Tools for Different Jobs
I'm not loyal to one tool. Here's roughly how I split it:
- Claude 3.5 Sonnet — long-form drafts, editing, anything where tone matters
- GPT-4o — structured tasks, lists, code, anything where I need reliable formatting
- Gemini 1.5 Pro — summarizing long documents, research with files attached
- Perplexity AI — when I want AI-assisted research with actual citations I can check
Each one has different strengths. Using only one is like only ever using a hammer.
Always Fact-Check Anything Specific
I can't say this enough. Numbers, dates, names, prices, product specs — verify all of it. I keep a browser tab open to the relevant primary sources while I'm editing any AI-assisted piece. It adds maybe 15-20 minutes to the process and has saved me from several embarrassing corrections.
Add the Stuff AI Can't Know
Your personal experience. Your actual opinion. Specific details from products you've held in your hands. Conversations you've had. This is the stuff that makes a piece worth reading and that no AI can generate for you.
Think of the AI draft as the skeleton. Your job is to put the actual flesh on it — the specific, lived, opinionated details that make someone want to read your site instead of anyone else's.
A Quick Note for Writers in Developing Markets
Something I think gets overlooked in these conversations: AI writing tools are particularly valuable if you're writing in a second language, or if you're in a market where good editorial support is hard to find or expensive. Using Claude or ChatGPT to check your grammar, improve clarity, or help you find more natural phrasing in English is completely legitimate and genuinely useful.
The $20/month for Claude Pro or ChatGPT Plus is a real cost if you're in a country where that's a significant amount of money — but the free tiers are actually decent for this kind of editorial assistance. GPT-4o mini is free on ChatGPT and handles grammar and style checks well.
The same rules apply though. The AI doesn't give you the expertise — that still has to come from you.
My Actual Take
AI writing tools aren't going away, and I don't think they should. They're useful. I use them regularly and I'm not embarrassed about it.
But I've watched enough blogs crash and burn — and caught myself getting lazy — to know that these tools can quietly undermine the thing that makes your writing worth anything if you let them. The voice, the expertise, the accountability for what you publish. That stuff is still on you.
Use them to write faster. Don't use them to replace actually knowing what you're talking about. That distinction is basically everything.