There are so many AI tools right now. It’s genuinely overwhelming. New ones pop up every week, each one promising to change the way you work forever.
Most of them won’t.
The real problem isn’t that there aren’t enough good tools. It’s that there are too many, and people are spending more time bouncing between them than actually getting work done. So we put together some notes on how to actually think about picking the right ones.
Start with what you’re trying to do
This sounds obvious but almost nobody does it. People hear about a cool new AI tool, sign up, and then try to find a use for it. That’s backwards. Cloud Geometry has a good piece on this that frames it as “decision-first thinking.” Define the outcome, then pick the tool that supports it.
Before you touch anything, ask yourself: what am I actually trying to get done? Write copy? Summarize meetings? Automate something repetitive? Analyze data? The answer tells you what category of tool you need, and usually narrows things down fast.
A few questions worth sitting with:
-
What’s the specific use case? (“Marketing” is too broad. “Write LinkedIn posts in our brand voice” is useful.)
-
Do I already have something that handles this?
-
Where does this need to fit in my actual workflow?
-
What are the security and privacy requirements?
-
Who’s going to use this, and how comfortable are they?
If you can’t answer these clearly, you’re not ready to pick a tool yet. And that’s fine. Martin Waxman’s checklist on LinkedIn walks through this step by step if you want a more structured approach.
Not everything needs AI
This is something people skip over way too fast. Google Cloud Tech did a really solid breakdown on when to use generative AI vs. traditional AI vs. no AI at all. Worth watching if you want the more technical version.
Generative AI (ChatGPT, Claude, Gemini, etc.) is great for creating things. Drafting, brainstorming, coding, outlining, generating structured data for other tools. It’s flexible.
Traditional AI / Machine Learning is better when you need something very specific. Sentiment analysis, classification, prediction models. Purpose-built stuff.
No AI is the right call more often than people think. If the process is already simple, if you’re making straightforward yes/no decisions with clear variables, or if adding AI would just overcomplicate something that already works? Skip it. Not everything needs to be “smart.”
Recognizing when AI isn’t the answer is honestly its own kind of expertise.
The big three (and when to use each)
If you’re working with generative AI, the main platforms each have their own thing going on. This 12-minute guide does a great job breaking down the strengths of each one:
ChatGPT is the most consistent. It follows instructions well, doesn’t drop the ball on multi-step tasks, and handles structured work reliably. If you’re only going to pay for one, this is probably it.
Gemini is a data monster. That massive context window means you can throw meeting recordings, long transcripts, big CSVs, audio files, even full-length videos at it and it won’t choke. Really strong for summarizing and pulling key findings out of large amounts of information.
Claude tends to produce higher quality on the first try. It’s great at “one-shotting” work, especially coding, automations, and copywriting. Also weirdly good at matching your writing style if you give it examples.
General advice from what we’ve seen: start with ChatGPT for most things. If you can afford all three, use them for their strengths. Gemini for heavy data processing, Claude for polished outputs and code.
Beyond the big three
A few other tools worth knowing about:
Perplexity is the best search tool out there right now. Fast, accurate, citation-heavy. Great for verifying information or finding recent data quickly.
Grok has direct access to real-time data through Twitter/X’s APIs. Useful when you need “right now” information.
NotebookLM is interesting because it only uses sources you upload. It’s harder for it to hallucinate because it doesn’t pull from outside knowledge. Great for fact-checking in-house documents or studying a specific body of material.
One workflow example that came up in our research: Gemini to brainstorm and outline a newsletter, Claude to revise and write the final draft, Perplexity to confirm accuracy and add references. Each tool doing what it’s best at, in sequence.
Specialized tools (images, video, audio, etc.)
Quick rundown because these come up a lot. AI Master’s 2025 guide covers all of these categories in more detail if you want to go deeper.
Image generators work differently than text AI. They run on diffusion, not transformers. DALL-E is easy and beginner-friendly. Midjourney is the gold standard but has a learning curve with its prompting style. The big rule here: be specific. Don’t let the AI fill in gaps you care about.
Music generators like Suno, Riffusion, and Moobare are pretty simple. Describe what you want, use pre-created options, or feed in scripts.
Text-to-speech tools like ElevenLabs and Speecheasy convert text to natural-sounding audio.
Video tools split into two types. Creators (Sora, Runway, PiKa, Hyper) generate new video from scratch. You can be more relaxed with prompts here. Editors (InVideo, Visla, Fliki) modify existing footage and need ultra-descriptive prompts, but don’t overcomplicate it either. They sometimes “forget” what you wrote if the prompt gets too long.
Productivity AI like Superhuman (email) and Taskade (task management) are less about creating and more about organizing your work.
One tool, one task
This is the simplest rule and probably the most important: pick one tool for one task.
Don’t use three different AI tools for writing. Don’t split your summarization workflow across platforms. Find the best fit for each job and commit.
Chaining tools together is different. That’s deliberate. But running the same task through multiple tools hoping one gives you a better answer? That’s just spinning your wheels.
How to evaluate before you commit
When you’re comparing options, run them through five lenses. Purdue’s evaluation guide goes deep on this if you want the full framework.
-
Functionality. Does it actually do what you need, or are you bending your workflow to fit the tool?
-
Usability. Can your team realistically adopt this?
-
Risk. What are the data privacy, security, and accuracy implications?
-
Cost. Does the pricing make sense for how often you’ll actually use it?
-
Integration. Does it fit your existing stack, or does it create a new silo?
A tool that looks impressive in a demo but doesn’t plug into anything you already use is a liability. Avoid short-term wins that turn into long-term headaches.
Prompting matters more than the tool
This came up in basically every resource we looked at. The tool matters, but the quality of your input matters more.
A few things that consistently make a difference:
-
Add context. Who’s reading the output? What’s it for? How long should it be? What format?
-
Set limits. Models work better with constraints than with “write me something.”
-
Roleplay helps. Tell the AI to act as an editor, strategist, analyst. It shapes the output.
-
Be specific with visual tools. Vague prompts produce vague results. Especially with images and video.
-
Don’t overcomplicate it. Especially with video. Overly complex prompts can backfire.
The bottom line, and I think this is the best takeaway from everything we read: AI’s output is only as good as what we give it.
Pick the right tools. Learn to use them well. That’s really it.
Resources
Everything we pulled from while putting this together:
Picking the right tools
-
Decision-first framing for AI tools (Cloud Geometry)
-
A guide to choosing AI models and tools (Sandtech)
-
How to choose generative AI tools that fit your needs (Martin Waxman)
-
Evaluating AI tools (Purdue IT)
-
How to choose the right AI framework for your use case (Medium)
Understanding AI tools
