Most articles about AI productivity read like testimonials for tools you're not sure the author has actually used. They describe workflows that sound slick and produce outputs that don't look like real work.
This isn't that. These are workflows I run regularly, with real numbers, real tools named, and honest notes on where they break.
I also included the three that wasted my time — because that's the part no one writes about.
The 5 That Actually Work
1. The Research-to-Outline Pipeline (Saves ~2 hours per article)
The problem it solves: Starting a piece of writing from a blank document. Staring. Getting up to make tea. Staring some more.
The workflow:
Step 1: I type my topic into Perplexity AI. Not ChatGPT — Perplexity. The reason is that Perplexity searches the web in real time and cites sources. For anything remotely time-sensitive or data-dependent, I want to know where the information came from.
Step 2: I ask Perplexity: "What are the most credible and counterintuitive things currently known about [topic]? Focus on specifics — stats, examples, cases — not general claims."
Step 3: I take the most interesting 5-7 facts or angles from the Perplexity output and paste them into Claude with the prompt: "I'm writing an article about [topic] for an audience of [describe]. Here are 7 research points I want to incorporate. Build me a detailed outline — not just headers, but a sentence or two describing what each section should actually say."
Step 4: I edit the outline to match how I actually think about the topic, then write from the outline.
What this replaces: The 2 hours I used to spend tabbing between browser windows, taking notes, and never quite feeling ready to start writing.
Where it breaks: When you're writing about something where your own direct experience is the value. For personal essays, skipping this step entirely is often better. The research pipeline is for informational writing, not narrative writing.
2. The Newsletter in 45 Minutes System
The problem it solves: Newsletters take forever. The good ones, anyway.
The workflow:
Monday: I send myself a voice memo describing what I want the newsletter to cover. Usually 3-5 minutes of rambling. This is where my actual ideas live — unfiltered and specific to my experience.
Tuesday: I transcribe the voice memo (MacWhisper does this in under a minute for free). I paste the transcript into Claude with: "Clean this transcript up into a structured newsletter draft. Don't invent anything — stay close to what I said. Organize it with: a short personal opener, the main insight section, 3 bullet points of the key takeaway, and a 'what I'm thinking about' closer."
Tuesday (continued): I edit the draft. Usually takes 20-25 minutes because the structure is already there. I'm improving, not starting.
Total time: Voice memo (5 min) + transcription (1 min) + AI draft (2 min) + editing (25 min) = about 33 minutes for a newsletter that reads well.
The key insight: The voice memo is the most important step. When I skip it and go straight to writing, I spend 45 minutes figuring out what I want to say. The voice memo externalizes that process.
3. The Reply-to-Comment Generator
The problem it solves: Responding thoughtfully to comments is time-consuming. Responding with "Thanks!" is worse than not responding at all.
The workflow:
Once per day, I copy the 3-5 comments I most want to reply to (across LinkedIn, YouTube, and email) into a single prompt: "Here are [number] comments I've received on my content about [general topic]. For each, write a genuine, specific reply that adds something to the conversation rather than just acknowledging the comment. Match the tone of each comment — some are analytical, some are casual. Keep replies under 3 sentences."
I review each reply, edit for accuracy (AI sometimes misremembers context), and publish.
What this replaces: The guilt of not replying to comments because I couldn't find the mental energy for 15 individual thoughtful responses.
Important caveat: I always edit before posting. AI replies occasionally miss the specific point being made or sound slightly off in tone. The edit catches this. Never publish AI-generated replies unread.
4. The Content Repurposing Engine
The problem it solves: Writing one piece of content that only lives in one place.
The workflow:
After publishing any long-form piece (article, newsletter, video transcript), I run it through this prompt: "Here is a [type of content] I published. Generate the following from it: (1) a Twitter/X thread — 5-7 tweets, each under 240 characters, each adding value on its own, (2) a LinkedIn post — opening hook, 4 bullet points of the key insights, closing question for comments, (3) a 3-sentence Instagram caption with a hook opener, and (4) a TikTok script using the most interesting 90 seconds of information from this piece."
What I get: One piece of research and writing turned into 4 platform-specific assets in under 5 minutes.
What I still do manually: The editing pass on each one. LinkedIn especially needs the opening line rewritten — AI versions tend to start with "I want to share..." which nobody should ever write on LinkedIn.
5. The Cold Email First-Draft Machine
The problem it solves: Cold outreach is painful to write. Every email feels like it sounds like every other cold email.
The workflow:
I research the target (5-10 minutes on their website, LinkedIn, recent content). I then prompt: "I'm reaching out to [name], who [describe who they are and what they do]. I want to [describe what I'm asking for — intro, collaboration, feedback, etc.]. The reason I'm contacting them specifically is [genuine, specific reason]. Write a cold email that: mentions their specific work in the first sentence (not in a fawning way — specifically), states my ask in the second paragraph clearly, keeps the total email under 150 words, and ends with a single easy yes/no question."
What makes this work: The specificity I bring. If I give generic inputs, I get generic outputs. The research I do manually becomes the intelligence that makes the email sound personal.
The 3 That Wasted My Time
1. AI Image Generation for Professional Use
I spent about 6 weeks trying to use DALL-E and Midjourney for featured images on articles and social posts. The results ranged from "almost usable" to "genuinely cursed." The time I spent writing prompts, iterating, downloading, and cropping was more than I would have spent browsing stock photo sites.
For highly specific, illustrative images — workflow diagrams, product mockups, specific scenarios — AI image generation isn't reliable enough yet for professional use without significant manual fixing.
What I do instead: Unsplash for editorial images. Canva templates for branded graphics. AI-generated images only when I have 10+ minutes to iterate and don't need the result to be precise.
2. AI-Generated Social Media Scheduling "Strategies"
I ran experiments with asking AI to generate my social media strategy — what to post, when, on which platforms, how often.
The strategies were plausible. They were also generic. They didn't know my audience, my time constraints, my existing backlog, or which content types my specific audience responds to. Following AI-generated posting schedules resulted in a period where I was producing more content with lower engagement, because the schedule optimized for frequency rather than for my specific channel dynamics.
What works instead: Using AI to generate content quickly and letting your analytics tell you what to post more of and when. AI strategy, no. AI production of a strategy you've already defined, yes.
3. Long-Form AI Writing I Tried to Edit Into My Voice
Early in my AI experimentation, I made the mistake of asking AI to write complete articles and then trying to rewrite them to sound like me. This is harder than starting from scratch. The AI establishes a structure, a tone, and a set of claims. Editing them out while preserving what's useful is a weird editorial puzzle that takes more time than just writing.
The correct use of AI for long-form writing is generating an outline and then you write the body. Or writing individual sections with you directing what each section needs to say. Not generating the whole thing and hoping you can salvage it.
The only exception: lower-stakes content where the voice doesn't matter as much (product descriptions, FAQ pages, listicles). For anything where your perspective is the value, AI shouldn't hold the pen.
The Pattern
Looking at what works and what doesn't, the pattern is consistent:
AI excels at processing, structuring, and reformatting. You give it information or ideas, and it makes them more organized, more readable, or more platform-appropriate.
AI struggles at originating — coming up with the initial idea, the personal insight, the specific experience that gives the content its reason to exist.
The failures in my list (image generation, strategies, full article drafts) all involve asking AI to originate something. The successes (research pipeline, newsletter drafting, repurposing) all involve giving AI good inputs and asking it to process them.
This isn't a criticism of the technology. It's just a clarification of what it's actually for.
Once you internalize that distinction, the workflows that save time become obvious. You're always the source. AI is the accelerant.
I write about AI tools and building small businesses with them. If this was useful, share it with someone spending too much time doing things AI could help with.
Top comments (0)