Google AI Pro: Image Editing, Deep Research and Fantasy Premier League

I recently switched to the EE mobile phone network, mostly because they reportedly have the best 5G network in the UK. While I’ve had to reroute my favourite park walk to avoid a pesky signal dropout spot, the service has mostly lived up to the hype. The real bonus though, was getting free access to Google AI Pro. That’s excellent timing, as Google’s AI products have really taken off this past few months. Here are some of the ways I’ve been using my new subscription.

Image editing

One of the things that made me nervous about cancelling my ChatGPT subscription was the lack of image editing functionality in the Google ecosystem. Here is Google’s hilarious attempt to put me in a Where’s Wally image. Kind of misses the point of the cartoon…

As you can see I had every reason to be afraid of Google’s image related AI capabilities. But in September, something big happened. Nano Banana launched. And wow.

Nano Banana is Google’s image editing AI model. What impressed me most was its seeming ability to only edit the specific part of the image you want it to, while leaving the rest untouched.

In 6 ways I like to use ChatGPT, my girlfriend and I wanted to see what our future kids might look like. Notice how it added two realistic-looking kids. But it changed us into different people. And most other parts of the photo.

See below the response from Nano Banana to the same prompt.

It saddens me that the kids clothes don’t match ours anymore. But it has done a much better job of not changing us into different people. It has even kept the background the same. This is a big improvement from the ChatGPT version. Though if you run the prompt a few times, we do eventually change into different people again. And I am still left wondering what our kids will actually look like, as ChatGPT and Gemini seem to be miles apart here. I might need to bring in Claude as the judge!

Google suite integration

As a Google Drive user, the ability to use Gemini (Google’s leading generative AI model) in Google Sheets is a massive timesaver. I no longer Google for formulas. I ask Gemini to write them from within Google Sheets. This is a big time saver, as copying and pasting spreadsheet formulas is never a simple copy and paste.

Email drafting is another valuable feature. Gemini’s suggested responses are often right on the mark, and even when they need editing (which they often do), they provide a great foundation to work from.

I also value the ability to easily export results from Gemini chat directly into the Google suite. However, this isn’t always perfect. Lately, I’ve run into an annoying issue where complex tables won’t export or copy/paste properly into Google Sheets. The workaround involves converting them to CSVs, downloading, uploading to Drive, and then converting them again. Poor me, right? Get out the world’s smallest violin!

Deep research

When you use deep research mode, instead of starting to give you a response within seconds, the model will take time to think. It will first break down your question into a step-by-step research plan. You can edit the research plan to get it right before continuing. It will then work through the research plan step-by-step, conducting google searches along the way to gather information from the web that might be relevant to the question.

I have used deep research mode thoroughly at work to help me with jobs such as:

• Competitor analysis
• Product comparisons
• Critiquing strategy decisions
• Briefing me on product domains as part of discovery research
• Sourcing hard to find documents for a database (and generating metadata)

Three things I find great about Gemini deep research compared to ChatGPT:

• Speed. Gemini usually responds in minutes. ChatGPT often takes 30 minutes or more to answer questions
• Understanding user intent: ChatGPT seemed to ask followup questions every time, even if the answer is obvious from my prompt. Gemini’s research plan helps me validate that my intent has been understood, and quickly correct it if it has not.
• Google suite integration: quickly getting results into Docs or Sheets saves a lot of copying, pasting and formatting losses

One thing that could be improved: it can be verbose. I often give specific output instructions to LLMs to put results in a table. Gemini often cannot resist giving me a few pages of insight before my lovely tables. Like ChatGPT with the ‘always asking followup questions’, it feels as if deep research mode is trying hard to follow a template in the way it sets out it’s responses, even if that isn’t what I am looking for. Or perhaps I am not reading it’s research plans thoroughly enough!

Gemini vs Fantasy Premier League

When I play Fantasy Premier League with my mates, I usually win. Those victories come at a cost though – hours of life lost to obsessively researching the best transfers, questioning who to make captain and deliberating on when to play wildcards. So this year I decided to try something different. I let Gemini play for me.

Two days before the start of the season, I started a deep research conversation. I prompted Gemini to first explore the best strategies for winning the game. And then to build me a team to win the season. Every week I then pasted in my current squad and available transfer budget, and asked it to tell me what to do.

Things started off well. Up until the end of September I was in the top three in all of my leagues. But since then things have gone downhill rapidly. I’m now near the bottom of most of my leagues (though above my brothers thankfully).

Jokes aside, despite my poor performance, I think this is a task that Deep Research is generally well suited to. The tool’s access to Google search gave it live access to the latest team selection news on the internet. It frequently recommended changes based on which players were on/off form, and which teams had easy/hard transfers coming up. That is the same process I go through when I am playing the game.

It made a few noticable mistakes. It occasionally recommended unviable transfers, suggesting I bring six midfielders into my squad rather than the maximum of five. It also could not read screenshots of my team. Though the fact that I even think I can screenshot an image on my mobile phone and have it identify the names and positions of the 15 players in my squad is a sign of just how high expectations have risen.

The biggest issue remains though: it’s losing! Where have things gone wrong? Sticking with Salah for too long? Not captaining Haaland? Transferring Evanlison in the week he gets injured? Not having enough Bournemouth players in the team (something I have never been guilty of as a Bournemouth fan)? Some speculate that AGI will be here in a couple of years, but AI in 2025 can’t seem to manage the simple task of beating my friends at this game – a feat I myself have managed many a time.

Note on ChatGPT comparisons

It’s probably a tad unfair for me to compare Gemini now to ChatGPT of 3 months ago. A few months is a long time in the world of AI!

Other AI blog posts written by me

AI prompt engineering as a product manager

6 ways I like to use ChatGPT

AI product management in high stakes domains

8 ways I am using AI to help me be a better product manager (and 4 ways I am not)