Actually Tested AI

How to Use Claude to Actually Understand a Long Boring Document Last week I had to read a 47-page municipal zoning variance report for a project I'm consulting on. The kind of document where every paragraph feels like it was written specifically to make you stop reading. I'd been putting it off for three days when I remembered: this is literally what Claude is supposed to be good at. The Obvious Approach That Didn't Work My first instinct was lazy. Copy the whole thing, paste it into Claude, type "summarize this." The summary I got back was technically accurate and completely useless. It told me the document was about a zoning variance request for a mixed-use development. Yeah, I knew that from the title. The problem with "summarize this" is that Claude doesn't know what you actually need from the document. It gives you the Wikipedia version — the facts without the judgment. For a 47-page zoning report, that meant I got a neat little ...

The Perplexity AI Feature That Makes It Better Than Google for Certain Searches I was trying to figure out which USB-C hub would actually work with my specific laptop model for dual 4K monitors, and after 20 minutes of clicking through Google results, I had twelve open tabs, three contradictory Reddit threads, and a growing headache. Then I asked Perplexity the same question. What happened next made me rethink how I search for anything technical. The Feature Nobody Talks About Properly It's called Sources, but that undersells it completely. Every claim Perplexity makes comes with a clickable citation. Not buried at the bottom. Right there, inline, as it talks. Here's why that matters more than it sounds. When Google gives you an AI Overview, you get a summary. Maybe it's right. Maybe it's confidently wrong. You have no idea which until you dig. Perplexity shows its work. I can see it pulled that spec from Tom's Hardware and that compatibility note fr...

The ChatGPT Setting Most People Never Change That Makes Answers Way Better I spent three months getting mediocre ChatGPT responses before I accidentally discovered I'd been ignoring the one setting that actually matters. Not temperature. Not the model selector. The custom instructions panel that's been sitting there the whole time, quietly waiting for me to stop being lazy. Where This Setting Actually Lives Click your profile icon. Then "Customize ChatGPT." Two text boxes appear. The first asks what you want ChatGPT to know about you. The second asks how you want it to respond. Most people glance at this, think "I'll fill that out later," and never come back. I know because I did exactly that. For months. Every single conversation started from zero. I'd re-explain that I'm a writer, that I hate corporate speak, that I need practical examples not theory. Over and over. It was exhausting and I didn't even realize it was exhausti...

When Not to Use AI: The Tasks Where It Actually Makes Things Worse Last Tuesday I spent three hours using ChatGPT to write a condolence card to my friend whose father died. Three hours. The final version I sent? I wrote it myself in four minutes, after throwing away everything the AI gave me. That's when I realized I'd been using AI wrong for months. Not because the tools are bad — I've tested dozens and many are genuinely useful. But I kept defaulting to them for tasks where they actively made my work worse. Slower. More generic. Sometimes embarrassingly off. Here's what two years of obsessive testing taught me about when to close the ChatGPT tab entirely. Tasks That Require Actually Knowing Someone The condolence card was the obvious example. But it's part of a bigger pattern. Anything that requires understanding a specific human being — not humans in general, but this one person — AI fails at hard. I tried using Claude to help me write a weddi...

How to Use AI to Read and Summarize a Contract Before You Sign I was about to sign a freelance contract last month when I realized I'd spent more time picking a coffee order that morning than actually reading the eight-page document in front of me. That's when I wondered if ChatGPT could do what I clearly wasn't going to do myself — actually read the thing. What I Actually Did (And Why It Almost Went Wrong) My first attempt was embarrassingly simple. I copied the entire contract, pasted it into ChatGPT, and typed "summarize this contract." What I got back was technically accurate but completely useless. A list of what each section contained. "Section 3 addresses payment terms. Section 4 covers intellectual property." Thanks, I could see that from the headers. The problem was I asked the wrong question. AI summarizes what you give it, but it doesn't know what you're worried about. So I tried again with something more specific: ...

How to Use AI to Learn New Software Without Watching Hours of Tutorials I stared at Blender's interface for the third time this year, closed it, and went back to pretending I didn't need to learn 3D modeling. That's been my pattern with complex software for years. Download it, get overwhelmed by the 47 toolbars, promise myself I'll "find a good tutorial series," never do it. Then about three months ago I accidentally discovered that ChatGPT could teach me software in a completely different way, and now I've actually learned four tools I'd been avoiding. The Method I Stumbled Into It started because I was too lazy to search YouTube. I had DaVinci Resolve open, wanted to do one specific thing — add a simple zoom effect to a clip — and just asked ChatGPT how to do it. Not "teach me video editing." Just "I have a clip selected in DaVinci Resolve, how do I make it slowly zoom in over 3 seconds." The answer was step-by-ste...

The AI Tool I Use Instead of Google When I Need a Fast Honest Answer Last Tuesday I needed to know if I could take ibuprofen with the antibiotic my doctor prescribed. Google gave me six sponsored links, three Reddit threads from 2019, and a WebMD article that somehow mentioned cancer in the second paragraph. I closed the tab and opened Perplexity instead. Thirty seconds later I had my answer with sources I could actually check. I've been using Perplexity AI as my default search for about eight months now. Not because I'm anti-Google or because I think AI is the future of everything. I just got tired of playing detective every time I had a simple question. What Actually Changed The first time I used Perplexity, I was researching whether a specific HDMI cable would work with my monitor's refresh rate. Google's results were mostly Amazon listings and forum posts where people argued about things that weren't my question. Perplexity just told me: yes, but...

I tested Gamma AI for presentations and this one thing genuinely surprised me I was staring at a blank Google Slides document at 11pm, knowing I had a client presentation in 14 hours, when I remembered Gamma existed. I'd dismissed it months ago as another "AI-generated content" gimmick. Turns out I was completely wrong about what it actually does. What I thought Gamma was vs. what it actually is I assumed Gamma was basically "paste your text, get ugly AI slides." That's what most AI presentation tools do. You end up with generic stock photos, weird layouts, and text that looks like it was designed by someone who's never seen a presentation before. Gamma doesn't work that way. You give it a topic or paste your content, and it generates something closer to a Notion document that happens to be presentable. The default output isn't slides — it's scrollable cards with actual visual hierarchy. This threw me off initially because I k...

AI Tools I Stopped Using After 30 Days and Exactly Why I canceled three AI subscriptions in the same week last month, and the reasons weren't what I expected when I signed up. None of them were bad tools. That's the weird part. They all did exactly what they promised. I just couldn't make them stick in my actual workflow. The Writing Assistant That Made Me a Worse Writer Jasper was first. I'd been paying $49/month for about six weeks when I noticed something uncomfortable. My drafts were starting to sound the same. Not the same as each other — the same as everything Jasper produces. That slightly enthusiastic, bullet-point-friendly, aggressively helpful tone that works great for landing pages but makes blog posts feel like they came off a conveyor belt. The tool worked fast. I'll give it that. I could generate a 1,200-word article outline in maybe 90 seconds. But I was spending more time editing out the Jasper-ness than I would have spent just writin...

How to Use AI to Prep for a Job Interview in a Single Evening I had 14 hours between getting the interview confirmation and walking into the building. Product marketing role at a SaaS company I'd only vaguely heard of. My normal prep routine takes three days minimum. So I threw everything at ChatGPT and tracked exactly what worked. The Company Research Phase (About 40 Minutes) First thing I tried: pasting the job description and asking ChatGPT to "tell me everything about this company." Useless. It gave me Wikipedia-level stuff mixed with outdated information and a few things that were just wrong. What actually worked was being specific about what I needed. I asked it to analyze the job posting for implied problems the company might be trying to solve by hiring this role. That prompt gave me three genuinely useful angles I wouldn't have thought of — like how the emphasis on "cross-functional collaboration" probably meant their current marketi...

AI Tools That Overpromise and What to Use Instead I spent $127 on an AI writing tool last month that promised to "write like a human." The first article it generated included the phrase "in today's digital landscape" three times in 400 words. I asked for a refund. They said no. Here's the thing nobody tells you about AI tools: the ones with the best marketing are almost never the best products. The flashiest landing pages, the bold claims about "revolutionizing your workflow" — these are warning signs, not features. After two years of testing pretty much every AI tool that crossed my feed, I've developed a pretty reliable pattern detector for the ones that waste your time and money. The Overpromisers I've Actually Used Let's start with the writing tools since that's where the gap between promise and reality is widest. Jasper AI tells you it can write "high-converting copy in seconds." I tested it on 12 diff...

How to Use AI to Write Product Descriptions That Actually Convert I spent three weeks rewriting product descriptions for a friend's Etsy shop using ChatGPT, and sales went up 23%. Then I tried the exact same approach for another friend's Shopify store, and nothing happened. Same prompts. Same process. Completely different results. That's when I realized most advice about AI product descriptions misses the thing that actually matters. What I Was Wrong About I assumed the hard part would be getting ChatGPT to write well. It wasn't. The AI can string together compelling sentences about basically anything. The hard part was figuring out what to tell it in the first place. My first attempts were embarrassing. I'd paste in "Write a product description for a handmade ceramic mug" and get back this smooth, professional copy that sounded exactly like every other ceramic mug listing on the internet. Technically correct. Zero personality. The kind of ...

How to Use AI to Write Better Text Messages in a Second Language I've been texting my Portuguese-speaking in-laws for three years now, and for most of that time, I was basically a robot. Grammatically correct, sure. But stiff. The kind of messages that technically say the right thing but feel like they came from a customer service bot. Then I started using ChatGPT differently — not to translate, but to help me sound like an actual person. The difference in how my mother-in-law responds now versus six months ago is wild. She sends voice notes. She sends jokes. She stopped using formal pronouns with me. Something shifted. The Problem With Just Translating My first instinct was to write what I wanted to say in English, then paste it into a translator. Google Translate, DeepL, whatever. The output was always technically correct. But here's the thing about text messages: nobody texts formally. In English, I'd text "gonna grab groceries, need anything?...

How to Use AI to Turn Meeting Notes Into Something People Actually Read Last Tuesday I watched my project manager paste raw meeting notes into Claude and get back something that looked like it was written by a corporate communications team having a bad day. Bullet points everywhere. Action items so generic they could apply to literally any meeting in history. The standard AI summary output that everyone ignores. I've spent the past three weeks testing different approaches to this exact problem. Not because I love meetings — I actively avoid them — but because the gap between "notes exist" and "people actually know what happened" is where most projects quietly die. The Default Output Is Useless (And Why) Here's what happens when you paste meeting notes into Claude with a basic "summarize this" prompt: you get a summary. Technically accurate. Completely ignorable. The AI extracts what was said without understanding what mattered. I ...

How to Use AI to Improve Your Writing Without It Sounding Rewritten I submitted an essay to a client last month that I'd run through ChatGPT for "light editing" and they emailed back asking why I suddenly sounded like a corporate training manual. That stung. I'd been using AI to edit my work for almost two years and somehow still managed to let it strip out everything that made my writing mine. Here's the thing I got wrong: I thought the problem was which AI tool I used. So I tested Claude, Gemini, different GPT models, even some dedicated writing assistants. Same issue every time. The problem wasn't the tool. It was how I was asking for help. The Default Settings Will Ruin Your Voice Every AI has a house style. ChatGPT loves transition words. Claude tends toward measured, academic phrasing. Gemini gets weirdly enthusiastic. When you paste your writing and say "make this better," you're asking the AI to rewrite it in its voice, no...