Build Your AI Team With Google Gemini Gems

Every week, real estate agents spend hours on the same repetitive tasks: writing blog posts from scratch, reviewing contracts under pressure, drafting social captions, and trying to remember how they framed that listing description last month. None of it is high-value work. It is the administrative drag that eats into the time agents should spend building relationships and closing transactions.

Tony Ray Baker found a way around it. A REMAX Realtor in Tucson, Arizona for 32 years, he built a full AI team inside Google Gemini using a feature called Gemini Gems: a blog writer, a legal reviewer, a social media manager, and a listing marketer. Each one is trained in his voice, knows his market and compliance requirements, and is ready the moment he opens his laptop. The entire setup costs him $20 a month. On Episode 10 of AI Agent Advantage, he walked JMan through exactly how he built it.

What a Gemini Gem Actually Is

A Gemini Gem is a specialized AI assistant you build and train once inside Google's Gemini platform. Unlike a standard chat session where you re-enter your preferences every time, a gem retains your instructions permanently. You set your writing style, your target audience, your formatting rules, and any language it should always apply or always avoid. Every future session starts exactly where you left it.

The practical result is significant. Opening your blog writer gem means going straight to content without re-explaining your brand. Opening your legal reviewer gem means getting a contract answer in seconds without re-uploading your references. And Gemini is not the only platform with this kind of persistent assistant: ChatGPT calls them GPTs, Claude calls them Projects. One setup, then specialized output on demand every time you return.

Meta-Prompting: Stop Guessing at the Training Prompt

Most agents who try to build a gem hit the same wall: they are unsure what to put in the training prompt, so they guess, and the output disappoints them. Tony Ray's solution is to stop guessing entirely.

His approach, which he calls meta-prompting, works like this. Instead of writing the training prompt yourself, describe your goal to Gemini in plain language and ask it to generate the optimal training prompt for you. Tell it what the gem will focus on, who your clients are, what standards you expect, and what it should always or never do. Then ask: now write the prompt that will make this gem perform at its best. Gemini builds the technical instruction set, you paste it into your gem configuration, and you are done — no prompt engineering background required.

The AI Team Tony Ray Built

His blog writer gem produces posts at a minimum of 500 words, structured with NLP principles, optimized for both traditional search and AI-powered search engines, written in his voice, with sources cited and his banned filler language excluded.

His legal reviewer gem is the one that gets the most attention. He uploaded Arizona's real estate contract forms, fair housing law documentation, and the NAR code of ethics directly into the gem, and instructed it to draw only on those sources. When a contract question comes up mid-transaction, he pastes the clause in and gets a response citing the exact lines and documents that apply. During one negotiation, when the agent on the other side was not performing to the contract timeline, Tony Ray had his legal gem help draft a note that was firm without escalating. The documents came through the same day.

"I love my lawyer. Have you ever heard that before? You have an advisor that is never going to talk to anybody else. So you can really be honest."

— Tony Ray Baker

He also built a social media gem, a website analyzer, and a listing marketer. Separately, on ChatGPT, he created a free public tool called Tony Ray's Little Marketing Helper: load an address, add property photos, a floor plan, and a handwritten upgrades list, and it generates MLS descriptions, social captions, fliers, geo-tag information, and lifestyle paragraphs in seconds. Any agent can use it at no cost.

One Gem, One Job

The instinct when building AI tools is to create one assistant that covers everything. Tony Ray tried it and found it consistently produced lower-quality output across every task. A gem trained to write blogs and review legal documents and create social posts ends up doing none of them particularly well. Keeping each gem focused on a single recurring task produces better output and makes each gem easier to improve, because when something needs adjusting, you know exactly which instruction to change.

His recommendation for getting started: write down the ten tasks in your business that cost you the most time each week. Prioritize the list. Build your first gem around the top item. Once it is running well, build the next one.

Key Takeaways

• A Gemini Gem is trained once with your preferences, voice, and rules, and retains everything permanently.
• Use meta-prompting: describe your goal to Gemini and ask it to write the optimal training prompt for you.
• Build single-purpose gems. One gem, one job. Focus beats a do-everything assistant every time.
• Load documents into a gem's knowledge base — Tony Ray's legal reviewer runs on Arizona contract forms, fair housing documentation, and the NAR code of ethics.
• Keep only one AI tab open per session. Multiple tabs on the same account cause inconsistent results and hallucination errors.
• Tony Ray's Little Marketing Helper is free on ChatGPT for any agent's listing content.

About Tony Ray Baker

Tony Ray Baker is a REMAX Realtor based in Tucson, Arizona with 32 years of experience in residential real estate. A social entrepreneur and committed AI practitioner, he owns five companies and studied prompt engineering under educator Devin McFaul. He created Tony Ray's Little Marketing Helper, a free GPT tool for real estate listing content, and runs a nonprofit art gallery in Tucson supporting children and local artists. Find him on Facebook at Tony Ray Baker or reach him at (520) 631-8669.