Is AI Failing You or Are You Failing AI?

A few months ago, I had a tedious task to pull cost info from hundreds of web pages. Perfect, I figured AI could knock it out in minutes. What followed was hours of bouncing between tools, writing scripts, tweaking prompts, and hitting limitations I didn’t expect.

I was treating AI like a shortcut. AI was failing… but the problem wasn’t AI. The problem was me using it like a point solution, without understanding what different models are actually built for or how to match the right tool to the job or that this wasn’t all or nothin’. 

That experience changed how I work with AI. Not just what I ask it to do, but how I think about its role in my workflow: creative partner or reliable system. Dreamer or doer.

Understanding AI: When You Want Consistency vs. When You Want Creativity

One of the most misunderstood qualities of AI is also one of its most powerful: it doesn’t always give the same answer twice. It’s a dream machine. And that’s a feature! Note a bug. 

• Creative writing: Ask AI to “write a story about a dog,” and one day you’ll get a goofy puppy tale. The next day? A tearjerker about a loyal companion
• Paraphrasing: When you ask AI to “reword this sentence,” it might give you a slightly different version each time, without losing meaning
  AI art tools: Ask for a “painting of a sunset,” and you’ll get wildly different visuals — vibrant colors, muted palettes, abstract forms, or photorealism
• Logo generation: Tools that use AI to design logos don’t settle on one perfect version. Each click gives you new fonts, layouts, and icons

Ironically, one of AI’s best traits is its ability to code so it can create code/scripts/formulas where you need a consistent output.

While all of this is great for some tasks… it’s not so great for others. What do you do? Start thinking of AI as both a creative partner and a technical assistant. Whether you need an imaginative spark or a reliable system, it’s often worth asking AI first.

4 Key Ideas for Using AI Consistently and Effectively 

1. Keep Your Eye on the Prize: Don’t forget the job to be done

AI makes it easy to keep tweaking—but endless refinement can become a time trap. Stay focused on the job to be done. The endless possibilities with AI can be distracting frmo why your there in the first place. 

If you’re generating an image for a slide, ask: is that extra 15 minutes of fine-tuning really worth it? Or was version two already good enough to get the point across? A good rule is to timebox your time for value, like setting a target to get that image “good enough” in 15m or less. 

2. Loop AI in as a thought partner if you aren’t sure of the best solution

It’s easy to assume AI only useful once you’ve figured out what you need it to do. But sometimes, that’s exactly when you should be looping it in.

I once spent over five hours trying to automate a fairly specific task: identifying cost-related info across 300+ web pages. My goal was to pull the clean on-page copy for each page (no ads, nav, or footer), drop it into a spreadsheet, and then extract only the sentences that mentioned costs. Perfect for AI, right? 

I started with Make.com. I knew it was powerful, so I went deep trying to make it work. But after a couple of hours wrangling JavaScript and running into roadblocks…  I knew I needed to pivot to something I understood better.

Next, I moved to GPT for Sheets. I used ImportXML to grab the page content, then ran a prompt in four columns using different models—something like:

=GPT("Identify each sentence that has cost information.",$D13,0,"gpt-4o-mini")

It was clear here that  GPT-4o was the best option, but still capped at around 8 sentences. nowhere near enough for longer pages. I’d still have to manually comb through the rest.

Finally, after 5+ hours, I asked ChatGPT a simple question: “Is there a Google Sheets formula that can extract sentences about cost?”

And, yes. 

=IFERROR(TEXTJOIN(CHAR(10), TRUE, FILTER(SPLIT(B5, "."), REGEXMATCH(SPLIT(B5, "."), "(?i)(cost|price|one-third|two-third|1/3|2/3|afford|expense|\\$[0-9])"))), "None")

The lesson: I love playing in Sheets and still fell into the trap of thinking I needed a more advanced solution. If I had used AI as a thinking partner up front instead of just a doing tool, I could’ve saved hours. I’ll chalk up those 5+ hours as important learning and focus on saving time with future problems 🙂) 

3. Most things should be attempted with multiple AIs until you confidently know the best use cases for each one of them

The skill that separates good from great AI users? Knowing when to switch models.

• Need a fast answer or quick code? Use a lighter, lower-reasoning model
• Already know what you’re doing? Avoid a deep reasoning model if you don’t want a full explanation

Try the same prompt across models. You’ll be surprised how much time you can save

🤖 Tool Tip: Poe.com allows you to run the same prompt across multiple models at the same time. 

This is how I feel waiting for a deep reasoning model when all I wanted was a quick answer.

On the flip side, reasoning models come in clutch when you actually do want that full context and understanding. Do you need help in dreaming up a solution or is a lower reasoning model getting stuck?

For whatever reason, some models are better at some things. Learn when to go to different AI tools. For example, Gemini does better with google sheet formulas. Claude is known for being the go-to for content writing. 

4. Know why AI didn’t nail it (and When it Will) 

If AI didn’t get you 100% of the way, identify the blocker.

• Too much data? Watch for models with bigger context windows or better CSV handling
• Weak reasoning? Revisit the task monthly or test it with newer models

Knowing the gap helps you spot when AI will be ready—and saves you from giving up too early.

The real power of AI is how you think with it. 

It’s a thought partner, not just a tool. The people getting ahead aren’t wasting time perfecting prompts, they’re asking better questions, picking the right model for the job, and knowing when to move on.

Use it to think faster, not do more.

Curious about how we’re using AI to unlock critical marketing tasks? I’d love to swap note.