How to Use AI to Be More Productive

I write for a living, which mostly means I sit in a chair and try to turn a fog of half-formed opinions into sentences that hold together. Three or four pieces a week, plus client calls, plus the endless admin that comes with working for yourself. For a long time my answer to “how do you get it all done” was a shrug and a lot of coffee. That answer changed over the past year, not because AI suddenly became magic, but because I stopped treating it as a novelty and started treating it as plumbing.

The honest version of this story is that AI tools gave me back maybe five to seven hours a week. Not thirty. Not a complete transformation of how I work. Five to seven hours, which is roughly a free afternoon, which is enough to matter. This piece is about which tools earned that time and which ones I quietly stopped opening. I’ll walk through what I actually use, how it fits into a normal day, and every place it has let me down.

The tools I actually keep open

My daily stack is smaller than you’d guess. I have a browser pinned with three tabs I use constantly: ChatGPT, Notion for drafting, and Otter for calls. That’s the core. Around the edges I run Zapier in the background to move things between apps, and I keep Fireflies around for the one client account where call notes need to land directly in a CRM. Everything else I’ve tried, and I’ve tried a lot, sits in a bookmarks folder I haven’t opened in months.

The reason the stack stays small is that context switching is expensive. Every extra tool is another place to check, another subscription, another set of keyboard shortcuts to remember. The productivity gain from AI collapses the moment you’re spending your afternoon managing the tools instead of using them. So I optimize for fewer things used more deeply, and I’m suspicious of any recommendation list that starts with “you need all twelve of these.”

Here’s what each one does in my actual workflow.

Starting the day with ChatGPT

For quick lookups, drafting emails, and brainstorming, ChatGPT is the thing I reach for first. My morning routine is unglamorous: I open it, paste in whatever mess is on my mind, and ask it to organize the mess. A client email I’ve been avoiding gets drafted in about ninety seconds, then I rewrite it in my own voice. A vague article idea gets turned into a rough outline with six section headings, four of which I throw away.

What makes it stick for me is voice mode. When I’m doing dishes or walking to get coffee, I can talk through a problem out loud and get a response back instead of typing. Voice mode is useful while you’re actually doing other things, which sounds like a small thing until you notice you’ve worked through a structural problem in an article while waiting for water to boil. In practice I use voice mode maybe twice a day, and almost always for thinking rather than producing.

On the substance side, I lean on GPT-4o for complex reasoning. When I need something scrutinized rather than generated, a pros-and-cons breakdown, a pressure test of an argument, a “what am I missing here,” that’s the mode I switch into. Image generation built-in is a nice bonus. I’ve used it for rough blog header concepts and for mocking up what a visual might look like before I hire someone, though I’ve never shipped an AI-generated image as final art. A free tier available means I can recommend it to readers without feeling like I’m shilling a paywall, and Plus runs $20/mo, which is the subscription I’d cancel last if I had to cut costs.

The caveat is real, though. It is confident about things it does not know. I’ve had it invent a plausible-sounding citation, get a date wrong by two years, and describe a feature that does not exist. My rule is simple: anything factual goes through a second source before it goes into a draft. Anything stylistic or structural I take happily. The tool is a drafting partner, not an authority, and the days I forget that are the days I waste an hour cleaning up confident nonsense.

Notion AI inside my writing workflow

Notion AI is where drafting actually happens, because Notion is where my notes already live. This is the argument for it that I underestimated for months: the best AI tool is usually the one already sitting in the place your work resides, because there’s no friction between thinking and using.

My setup is one master database of ideas. Half-formed thoughts land there in fragments, some of them a single sentence. When an idea graduates, I open the page, ask Notion AI to expand it into a rough structure, and start writing into the gaps. The summaries it generates are genuinely useful for something I wrote three months ago and now can’t remember: paste a long note, ask what the argument was, and get a compressed version.

Where it earns its keep is the boring middle of writing. Rephrasing a clunky paragraph. Tightening a section that ran long. Turning a messy bullet dump from a call into something readable. It’s a competent editor for small moves and a mediocre one for big ones. I would not trust it to write a whole piece in my voice, and I’ve tried, and the result read like a press release written by someone who has never met me.

The drawback is that it’s only as good as the workspace around it, and my workspace is not tidy. It will summarize my own messy notes back to me with unjustified confidence, occasionally blending two projects together. Subscription cost also stacks: Notion itself, plus the AI add-on, on top of everything else on this list. For someone who doesn’t already live in Notion, the value proposition is much weaker, and I’d tell them to skip it rather than switch their whole system for one feature.

Meetings with Otter.ai

Otter.ai is the one where the time savings were immediate and undeniable. I have a lot of calls, and before Otter I was the person half-listening while typing, which is the worst of both worlds. Now I join, say hello, and pay attention to the human being on the other side.

It transcribes in real-time and generates summaries. My call routine is: Otter captures the conversation, auto-generated summaries and action items land in my inbox within minutes of hanging up, and I do a two-minute pass to fix names and clean up the three things I actually committed to. What used to be a fifteen-minute post-call write-up is now closer to three. Connects to Zoom, Google Meet, and Teams, so I don’t have to think about which platform a client prefers.

The real win is not the transcript. It’s that a transcript exists at all. When a client says in week six that we agreed to something different, I can search and find the exact moment instead of arguing from memory. That alone changed how I handle scope conversations. The tool made me better at my job in a way that had nothing to do with writing.

It is not free of annoyances. Real-time transcription struggles with accents, crosstalk, and people who talk over each other, and it will confidently render a name as something nobody has ever been called. Auto-summaries occasionally flatten a nuanced discussion into a bullet that reads more decisive than we were. And the privacy question is not trivial: you’re uploading recordings of client conversations to a third party, which needs to be disclosed and, for some clients, isn’t acceptable at all. Otter runs $10/mo for Pro, which is cheap for what it does, but I’ve had two clients ask me to turn it off and I’ve respected that.

Fireflies.ai when search and CRM matter more

Fireflies.ai is a second meeting transcription tool, and I initially picked it up expecting to replace Otter. I didn’t. What it offers is better search and CRM integration, which is a different job, and it’s good if you’re in sales or client-facing roles.

The feature that decided it for me is auto CRM logging. Notes from a call flow into Salesforce or HubSpot without me copying anything, which matters when the point of the call was the pipeline, not the prose. Topic tracking across calls is genuinely useful in a way single-call summaries can’t be: I can see that pricing came up in four consecutive conversations with one account, which is a signal I’d have missed reading each transcript separately. Shareable soundbites let me send a thirty-second clip of a client saying exactly what they wanted, which ends more internal debates than any summary I’ve ever written. A free tier available means the experiment costs nothing.

The honest tradeoff is that Fireflies is more tool than a solo writer needs. CRM logging is worthless if you don’t run a CRM. Topic tracking needs volume to say anything interesting, and with a handful of calls a week the patterns are mostly noise. Search is strong but the interface takes real time to learn, and I’d be lying if I said I’d fully mastered it after a couple of months. If your calls are about relationships and deliverables, Otter is the lighter, better-fitting choice. If your calls are about deals and follow-through, Fireflies earns its place. Running both is possible, and I did for a while, but two meeting bots on one call is one too many.

Zapier as the glue

Zapier isn’t an AI assistant in the conversational sense, but it’s the piece that makes the rest of the stack feel like a system instead of a pile of tabs. It moves information between apps on triggers I define, and it handles the transitions I used to perform by hand.

My two most valuable automations are boring. First, when a new call summary lands from my transcription tool, a copy is filed into the right client folder in my notes and a task is created with the action items attached. Second, when a post I’ve written goes live, a draft social post and a newsletter blurb are created for me to approve. Neither is clever. Together they remove about twenty minutes a day of copy-paste, which is an hour and forty minutes a week of pure administrative drag.

The cost is that automations are brittle. When an app changes its API, or a field gets renamed, the zap breaks silently and you find out three weeks later when things are missing. I’ve learned to check the task history weekly, and I’ve had a client folder quietly stop filling up for a month. There’s also a learning curve: the simple zaps take ten minutes, the ones that would genuinely save you time take an afternoon of trial and error.

Where these tools fall short

I want to be precise about the failures, because recommendation lists tend to skip them and they’re the part that determines whether this works for you.

Hallucination is the big one, and it’s not solved. Every tool here will occasionally state a wrong fact with total composure, and the failure mode is worst exactly where you’re most tempted to trust it: names, numbers, dates, quotes. My mitigation is procedural rather than technological. AI output is a draft until a human verifies the specifics. I never let a generated sentence containing a statistic reach a client without checking the source myself.

Privacy is the second one, and it’s the reason two of my clients asked me to stop recording. Transcription is the sharp edge: you are sending recordings of other people’s conversations to a company they haven’t vetted. For regulated work in particular, that’s not a preference, it’s a constraint, and no amount of convenience overrides it. I now disclose recording at the top of every call as a matter of course.

Subscription cost is the third and the most mundane. Individually these look small, and stacked they add up fast across a year. I do a quarterly audit and I’ve cancelled more tools than I’ve kept, which is the correct ratio. The learning curve is the fourth: none of these are hard individually, but adopting five at once means you adopt none of them well. I added one at a time, a couple of weeks apart, and only kept the ones that survived contact with a real deadline.

How I’d build this stack from scratch

If I were starting today with no tools and no habits, I would not buy five subscriptions. I’d do this instead.

Start with one general assistant and use it daily for a month. That means ChatGPT, and the free tier available is genuinely enough to know whether this fits how you think. Only upgrade to Plus at $20/mo once you catch yourself hitting limits, which for me took about three weeks. The point of the first month isn’t productivity, it’s calibration: learning where it’s reliable and where it’s a confident liar, in low-stakes conditions.

Then solve your single biggest time sink. For me that was meetings, and Otter at $10/mo for Pro paid for itself in the first week. If your time sink is drafting, look at Notion AI instead, but only if your notes already live in Notion. If it’s sales follow-through, Fireflies is the better fit, and the free tier available is the right way to find out. Then, and only then, add Zapier to connect things, because automations only make sense once there are two or three things worth connecting.

The order matters more than the specific tools. Each addition should be a response to a frustration you’ve already felt for a couple of weeks, not a tool you’re excited to try. That discipline is why my stack is small. It’s also why it works, and why I’ve gone back to being just a person in a chair who gets more done than they used to, with a free afternoon to show for it.

Where to look

Disclosure: this post contains an affiliate link. If you buy through it we may earn a small commission, at no extra cost to you.