
AI Tools That Disappointed Me
I try a lot of AI tools and honestly most of them don't make the cut, here's the specific stuff that let me down and why.
I feel like most AI content is either "here are the best tools you need right now" or "AI is overhyped and nothing works." And both of those are kind of useless if you're actually trying to figure out what to use. So I wanted to write something more honest: tools I was genuinely excited about, tried seriously, and then stopped using. And specifically why.
This isn't me dunking on anyone. These are products made by people working hard, and some of them will get better. But I think you deserve to know what actually happened when I used them, not just the best-case-scenario version.
AI meeting summary tools (in general)
Okay I have tried four of these. Four. And every single one of them has the same problem: they're excellent at summarizing what was said and terrible at capturing what actually mattered.
Here's what I mean. You have a meeting. There's a lot of talking. At the end, the AI summary tells you: "Emily discussed the project timeline. John raised concerns about the budget. Sarah suggested a follow-up meeting." All technically accurate. Completely useless. The actual important thing that happened in that meeting was the thirty-second exchange where John's tone shifted and you realized the budget concern was actually a political problem about ownership, and the AI summary has absolutely no way to capture that because it's processing words, not meaning.
I kept trying different tools thinking the next one would figure this out. None of them have. And I'm not sure they can, actually. The things that make meetings valuable are often not in the transcript. So I've gone back to writing my own notes and accepting that this is just a task that requires a human.
AI image generation for anything with text in it
This one I want to be precise about because image generation is genuinely great for some things. But I spent probably three weeks thinking I was doing something wrong because I couldn't get readable text in generated images. Turns out: it's just bad at this. Still. After years of development.
I was trying to make mockups that included product labels with actual words on them. Sounds simple. The models kept giving me something that looked like text from across the room but was actually gibberish up close. Sometimes it would get the first word right and then fall apart. Sometimes it would invent entirely different words than what I asked for.
I eventually learned to generate the image without text and add text in Canva afterward. Which works fine but it's annoying that this is still the workaround in 2026. I watched at least six demos where people generated images with perfect text and I still don't know how they did it or if those demos were cherry-picked. My honest experience is that text in AI images is unreliable enough that I don't plan around it.
A specific AI writing tool I paid for
I'm not going to name it because I don't want to be unfair to a specific product that might have improved since I used it. But I paid for a year subscription to an AI writing assistant that was specifically marketed for content creators. It had a whole personality and voice system where you could train it on your own writing.
I spent two hours feeding it my old posts. I was excited. Then I used it for a week and everything it produced sounded like... a competent stranger who'd read some of my work. Not me. Competent, but generic. Every sentence it wrote, I found myself rewriting to put actual specificity in it. It kept describing things vaguely where I would have given a specific example. It kept hedging where I would have taken a position.
Eventually I realized I was spending more time editing its output than I would have spent just writing. And the starting point of its output was actually making my writing worse because I was unconsciously trying to preserve the structure it gave me instead of going where the piece actually wanted to go.
I cancelled after three months. I think the voice-training thing is genuinely hard and most of these tools are not there yet. The output is good enough that it looks like you on the surface but the specific choices that make writing interesting, the unexpected angle, the weird example, the willingness to be direct, those don't get captured.
AI customer service chatbots (almost all of them)
I know I'm not the main audience for this one since I'm not running a business that would deploy these. But I've been on the receiving end of a lot of AI customer service bots recently and they are consistently one of the most frustrating experiences in my life right now.
Here's the specific failure mode: they're very good at understanding what you're asking and very bad at doing anything about it. I had an issue with a subscription last month. The bot understood my problem immediately. I could tell because it gave me a perfect summary of it back. And then it offered me three options, none of which solved my problem, and when I said none of those, it looped back to the beginning. Three times. Before I could get a human.
The AI made the problem feel worse because I knew it understood me. A phone tree that doesn't understand you at least has the excuse of being dumb. A system that clearly comprehends your problem and then doesn't help you anyway feels weirdly dismissive. Like someone looking you in the eye while they ignore you.
I think a lot of these deployments are just automated deflection with better UX, not actual problem-solving. And that's a choice companies are making to save on support costs, not a technical limitation. Which makes it worse, not better.
The general pattern I've noticed
Almost everything I've been disappointed by has the same shape. The tool is genuinely impressive at the obvious thing, summarizing, generating, understanding. But it falls apart at the judgment calls. At knowing what matters. At the parts of a task that require actually caring about the outcome.
I don't think that's the tools' fault exactly. They're doing what they're designed to do. I think it's more that a lot of the hype sets up expectations that include the judgment part, and the judgment part isn't there yet. Probably won't be for a while. And in the meantime, the tools are actually useful for the specific tasks where judgment isn't required. And kind of useless for the tasks where it is.
Figuring out which is which, for your specific work, in your specific situation, that's the real skill. I'm still working on it. Some days I get it right. Some days I spend an hour using an AI tool to do something I could have done better in fifteen minutes. Both happen. And I'd rather admit that than pretend I've got it all figured out.
Emily in AI
Emily in AI is a plain-English guide to AI tools, tips, and beginner guides. Every tool gets tested and written up without the hype or the jargon, so you can figure out what actually helps. New posts every week.
About Emily in AI →

