I had a tool problem. Past tense.
For about a year, every week brought a new “game-changing” AI tool that was going to solve whatever I was stuck on that day. Jasper for copy. Hootsuite AI for social. Superhuman for email. I bought most of them on the wow factor, then kept paying for them long after the wow wore off.
When I finally added it up, I’d burned $7,337.47 on AI subscriptions I no longer use. Here’s the painful breakdown, so you can skip the part where I learned it the expensive way.
The pattern underneath all of it is the one I keep coming back to in which AI tool for which job: the marketers pulling ahead don’t buy a point tool for every problem. They route the work to a tool they already own. Most of the graveyard below is me learning that lesson one $49-a-month subscription at a time.
The wake-up call
I uploaded a year of Amex statements to ChatGPT and told it to act as a financial analyst and categorize every tool charge. That was the shock therapy. Seeing the line items grouped together, with annual totals next to each one, landed differently than approving them one card-swipe at a time.
Turns out I’m not special here. Industry surveys put the share of paid software licenses that sit unused somewhere around half. The exact figure moves depending on who’s counting, but the point holds: a real chunk of every software budget is paying for tools nobody opens. AI made that worse, because the FOMO is louder and the trials are everywhere.
The tools that failed me
Jasper, Copy.ai, Rytr: expensive Mad Libs
I tested all three early, and the first session always felt like magic. Then the copy started to feel formulaic and a little soulless, like filling in the blanks on a pricey set of Mad Libs. I also forgot to cancel a couple of them, which is its own lesson.
Here’s where I landed. A well-built prompt in ChatGPT produces copy as good as any of the copy-specific tools, without yet another $49-to-$99 monthly subscription. I keep a small set of custom instructions running so it already knows my voice and my rules, then I hand the draft to Claude to edit and tighten. Two tools I was paying for anyway, doing the job three tools used to.
Social slop: Hootsuite AI, Blaze.ai, TweetHunter
Roughly $2,208 of automation without a strategy. These promised to auto-generate and schedule “engaging” posts across every channel. In practice, auto-posting isn’t a strategy. It’s activity. My engagement started sliding, and when I looked closer, I was shipping slop.
My audience could tell. People are good at spotting machine-written posts now, and they do exactly what you’d expect: scroll right past. The mostly-AI posts pulled lower engagement than the ones I actually wrote, and the platforms didn’t reward the volume either. A bot impersonating a person, with no point of view, wins no fans.
So I went back to a hybrid. AI brainstorms and researches the topics. A human (me, or someone on my team) reviews, adds the context only a person has, and schedules it on purpose. We also post natively to each platform instead of pushing everything through a scheduler, which keeps the algorithms friendlier.
Email tools: right intention, wrong problem
I hate managing email. Six accounts means six streams of cold outreach and spam. I tried throwing tools at it, and the real lesson was that I was solving the wrong problem.
Superhuman, around $480 a year. To get the value you have to live inside their app. I replaced it with ChatGPT, a few Zapier automations, and plain Gmail across four of my accounts, and pocketed the savings. The other two accounts belong to client orgs whose IT teams won’t let me connect anything, which is its own reminder: the tool you can actually deploy beats the one with the best demo.
Other casualties
Leonardo, about $288. I picked it up to test a Midjourney alternative. It wasn’t the one for me. And the bigger shift since then is that the base assistants got good at images. When I wrote the first version of this, I guessed that ChatGPT’s image generation would make a standalone image tool hard to justify. That mostly played out. ChatGPT 5.x and Gemini 2.x now handle the everyday image and ad-creative work I used to route to a separate app, with text inside the image that actually holds together.
The tools that survived the purge
The keepers all share a trait: each one owns a job nothing else does as well, and I reach for it on reflex.
- ChatGPT (Plus and Pro): the $20 Swiss Army knife. Drafting, brainstorming, the financial-analyst trick above, and now most of my quick image work.
- Claude: my editor, my code and automation seat, and the second brain I check everything against. With Claude Code and the current 4.x models (Opus 4.6 for the heavy lifting), it does work that used to need a developer.
- NotebookLM: turns a stack of PDFs into a podcast I can listen to on a walk. Nothing else replaces it cleanly.
- Notion: my second brain and the datastore everything else points back to.
- Perplexity: still my fastest path to a sourced answer when the question is “what’s true right now.”
A couple of honorable mentions still earn their keep: Descript for video editing, and Midjourney when I want a specific look the base models don’t nail. Even those I re-check every few months, because the gap keeps closing.
How I keep the graveyard from filling back up
The audit isn’t a one-time event. Every month I upload the QuickBooks and Amex PDFs to ChatGPT and have it flag any tool charge against actual usage. When I test something new for the newsletter, I run it for a month or two and then cancel it on purpose. I also keep a dedicated Gmail label per tool so the renewal notices never get lost in the inbox.
If you want to run your own audit, it’s three steps:
- Upload a few months of card statements or screenshots to your AI of choice and have it categorize every software charge.
- Put actual usage next to each cost. Be ruthless about “I might use it later.”
- Cancel anything you don’t use consistently, then rebuild that job as a workflow inside a tool you already own (OpenAI, Google, or Microsoft).
The bottom line
AI tool FOMO is expensive, and the bill arrives quietly. You’re far better off learning to master three tools than to dabble in thirty. A good process, clean inputs, and one or two assistants you actually know how to drive will get you most of the way on almost any job, without the context-switching tax of a dozen logins.
That’s the same move I lay out across the AI marketing hub: pick the job, route it to the tool that’s best at it, supervise the result. The graveyard is just what happens when you skip the routing step and buy a tool instead.
Send me your worst one
I keep a running list of AI subscriptions I regret, and reading yours makes mine feel a little less lonely. Twice a week I send the teardown: what I’m testing this week, what I just cancelled, and the one tool that earned its spot. If you’ve got a $7,000 graveyard of your own, this is your people.