When Google dropped the screen-sharing feature in Gemini 2.0, I had to try it out. I thought, "What better way to push its limits than by pairing it with SpyFu, a tool I know inside and out?" My goal was to see if this AI could handle complex SEO data and actually help me uncover insights about a competitor’s rankings. Spoiler alert: It took a few false starts, but I finally cracked it with a hack that worked shockingly well.
In the video below, you’ll see the whole process—from the initial struggles to that pivotal moment when everything clicked. It’s a great example of how persistence (and a little creativity) can unlock the true potential of AI.
When something like Gemini 2.0 hits the market, I love learning first hand what it might be able to do that maybe we aren't hearing or reading in the marketing materials. The fact that I can interact conversationally with my browser opens a ton of doors, and a test drive is the best way to stumble onto some possibilities.
One of the ways to learn a tool inside and out is to also look for its weaknesses. I don't want to embrace Gemini without really knowing its limitations. Those limitations are helpful. They teach me how to get more from the tool--especially when I can test some trial and error adjustments to my approach.
That's what happens here. And this time, that adjustment was a huge "aha" moment.
It started with some basic questions. I shared my screen and asked about Wayfair’s SEO performance—simple things like their top pages and keywords. The AI did... okay. It could point me toward the obvious metrics, but when I tried to go deeper—like finding keywords Wayfair started ranking for this year—it fumbled.
The AI felt limited to what it could “see” on the screen. If I didn’t click the right dropdown or hover over a specific element, it couldn’t find what I was looking for. So I kept restarting, rephrasing, and guiding it manually. But the deeper I went, the more I felt like I was working harder than the AI.
If you know me, you know I don't just throw up my hands and quit. I tried new approaches. First, I simplified my questions, thinking maybe I was overwhelming it. Then I switched tactics and acted like a trade show attendee, asking the AI to walk me through the tool step by step. This approach was fun—it showed some promise—but it still didn’t get me the nuanced insights I was after.
Finally, I had an idea: what if the AI needed to “know” everything about the tool up front? What if I preloaded it with SpyFu’s documentation and guided it to understand all the features before we started? That’s when everything changed.
"What if the AI needed to “know” everything about the tool up front?"
I took all the help articles, feature guides, and definitions I could find for SpyFu’s SEO Keywords page and pasted them into the AI’s system instructions. On top of that, I gave it a clear directive: “You’re an SEO expert for SpyFu, and your job is to know this tool inside and out.”
Once I armed the AI with all the context it needed, I asked the same question: “What keywords has Wayfair recently started outranking Home Depot on?” The response was immediate—and spot on.
The AI pointed me to the “They Just Surpassed You” filter, which appears after you input a competitor’s domain in the “Compare to Your Site’s Ranks” field. This filter showed me exactly where Wayfair had edged out Home Depot in rankings. No guesswork, no trial and error—just results.
The breakthrough wasn’t just about finding the right filter on the page. It was realizing that AI is only as good as the context you give it. By preloading all the tool’s features and functionality into its system prompt, I turned it into an actual expert. Suddenly, the AI could guide me effortlessly, helping me discover hidden insights like:
If you’re wrestling with AI that doesn’t seem to “get” your product, try giving it a thorough onboarding process first. Fully priming the AI is what made it actually useful.
When we finally hit the right notes, I wanted to be sure that I could bypass that learning curve in the future. I asked Gemini how to efficiently load it with context--screenshots, help documentation, articles, etc.
I'll share it with you directly, but you can also check out the video from around 7:40 to 8:50. Here are the main pointers:
I even took it a big step further.
Like many SaaS websites, SpyFu has a Help Desk resource and blog articles to help learn their tools. (Well, our tools.) I found an article dedicated to the SEO Keywords feature--the very page that I was testing. I pasted that full article into the Gemini system instructions along with a note for more context that it was an SEO expert.
This isn’t just a neat trick for me—it’s a game-changer for how we think about AI and customer support. Right now, AI capabilities in our live chat platform can scan our Help Desk library and suggest articles that might give a customer the answer they are looking for. But what if we could teach AI to be more helpful to the customer? Knowing that the browser's AI can be taught how to understand some details about the tool could ease customer frustration.
In fact, I didn't even maximize the information upfront. I just pasted in a matching article that was on our blog, and it still was like having a conversation with something who was well versed in the app.
We could expand on this by offering more specific direction for the tasks at hand. If we can get it to understand where functions are loaded on the page and what a user might want to do, it can act "smarter" in every instance. If we can make some inferences about what you are trying to do with the tool at a given point, Gemini's core model will figure out how to present the right help at the right time.
We just have to teach it where the options are, and we can trust it to solve the challenge from there.
With the right setup, AI can go from “clunky assistant” to a genuinely useful guide. By capturing deeper details, it can serve them up at the right time. While I was looking at SEO keywords that our rivalflow.com site ranks for, Gemini suggested another feature to me--Ranking History--that I could view for a better sense of how long those keywords had been part of our SEO footprint.
What started as a frustrating experiment ended with a surprising revelation. This hack isn’t just about making AI work better—it’s about unlocking its potential to solve complex problems in ways we couldn’t before. And if you try it, I think you’ll see exactly what I mean.
