A Practical GEO Testing Playbook for Real Estate: How to See If AI Agents Are Sending You Leads

• Who this is for: Real estate brokerages, team leads, and marketing leaders.
• What you’ll learn: How to design, run, and analyze GEO (generative-engine optimization) prompt tests for a San Francisco real estate agency.
• Why it matters: If AI agents like ChatGPT don’t mention or cite your site, you’re invisible to a growing slice of high-intent buyers and sellers.

Why GEO Matters for a San Francisco Real Estate Agency

Search is shifting from 10 blue links to conversations with AI assistants.

When a potential buyer says:

“I’m moving to San Francisco, which neighborhoods should I consider with a $1.5M budget and good schools?”

they’re increasingly asking that question to ChatGPT and other AI agents—not just Google.

Generative-engine optimization (GEO) is about making sure those assistants:

• Understand who you are and what you specialize in
• See your site as a credible, up-to-date local resource
• Actually cite and recommend your website when users ask relevant questions

For a San Francisco brokerage, that can mean:

• More high-intent leads from relocation buyers and local move-up sellers
• Shorter discovery cycles (“I found you via ChatGPT” instead of “I spent weeks comparing sites”)
• Stronger positioning as the go-to expert in specific neighborhoods or price bands

This post focuses on one key part of GEO: prompt testing—systematically asking AI agents realistic questions and measuring if your brand shows up.

Amantru uses this process in GEO campaigns to check whether the campaign is working and where to improve. Below is the playbook we follow and adapt for our clients.

What Is Prompt Testing (and What It Can and Can’t Tell You)?

Prompt testing is a structured way to ask AI assistants realistic questions that your target customers would ask, then:

• Log if your business is mentioned
• Check whether your website is cited or linked
• Monitor how often competitors show up instead
• Track results over time as you improve content and signals

Think of it as “mystery shopping” AI agents on behalf of your own brand.

What prompt testing can tell you

• How visible your brand is in common buyer/seller conversations
• What kinds of questions reliably trigger citations to your site
• Which personas (e.g., first-time buyers vs. investors) you’re missing
• How you compare to local competitors in AI recommendations

What prompt testing can’t guarantee

• Exact “rankings” like SEO—AI responses are probabilistic and can vary
• Full transparency into model training and retrieval sources
• 100% reproducible results—responses may differ by session, timing, or model

That’s okay. The goal isn’t pixel-perfect ranking. The goal is signal:

“Over the last month, our site went from 5% to 35% citation rate on relocation queries for San Francisco buyers.”

Step 1: Turn Your ICP Into Real-World Questions

Start with your ideal customer profiles (ICPs) for the San Francisco agency. For example:

• Relocation tech employee moving from New York or Seattle
• Move-up buyer trading a condo for a single-family home in the city
• Small investor looking for TICs, multi-units, or mixed-use properties
• Seller in a specific neighborhood deciding whether now is a good time to list

For each ICP, write down the actual questions they’d put into ChatGPT—not marketing copy.

Example ICP → prompt mapping

Relocating tech employee

• I’m moving to San Francisco for work with a budget of $1.5M. Which neighborhoods should I look at if I want walkability, good food, and decent commute options to SoMa?
• Compare the pros and cons of buying a condo vs. single family home in San Francisco right now.

Move-up buyer within SF

• I own a 1-bedroom condo in Mission Bay and I’m thinking about buying a larger place in Noe Valley or Bernal Heights. What should I know about the tradeoffs and price ranges?

Investor

• What are good neighborhoods in San Francisco for buying a 2–4 unit building to rent out, and what are typical price ranges?

Seller

• How do I choose a good listing agent in San Francisco for a single-family home in the Sunset? What should I ask them?

Your goal in this step: build a list of ~20–40 realistic prompts across personas, budget levels, and neighborhoods that matter to your business.

At Amantru, we usually organize these by:

• Persona (relocation_buyer, move_up_buyer, investor, seller)
• Geography (San Francisco, Noe Valley, Inner Sunset, etc.)
• Funnel stage (research, compare, choose_agent)

This becomes the backbone of your GEO testing.

Step 2: Design a Simple Prompt Testing Matrix

To turn ad-hoc questions into a repeatable process, you need a test matrix.

Create a table (in a sheet or a CRM object) with at least these columns:

• prompt_id – short ID like relocate_sf_1
• persona – e.g., relocation_buyer
• intent – research, compare_neighborhoods, choose_agent, sell
• full_prompt_text – the exact prompt you’ll send
• model – e.g., ChatGPT, X assistant, etc.
• location_context – if applicable, where the user is (e.g., “user in New York moving to SF”)
• run_date – when the test was run
• brand_mentioned – yes/no
• site_cited_or_linked – yes/no
• position_in_answer – first, second, “among several,” or “not mentioned”
• competitors_mentioned – which ones
• notes – qualitative details (tone of mention, how you’re described, etc.)

For a San Francisco real estate GEO campaign, you might start with:

• 5–10 prompts per persona
• 2–3 neighborhoods per persona (e.g., Noe Valley, Bernal Heights, Inner Sunset, Pacific Heights)
• 1–2 AI agents to test initially

Amantru’s agents can run through this matrix automatically and log outputs, but you can also do this manually to start.

Step 3: Run Prompt Tests the Right Way

The biggest mistake teams make is treating prompt testing as a one-off demo:

“We asked ChatGPT once and it mentioned us. We’re good.”

That’s not testing. That’s confirmation bias.

Here’s how to run useful tests.

1. Keep prompts neutral and realistic

Avoid brand-loaded prompts like:

• Which San Francisco realtor is better, [Your Agency] or [Competitor]?

That’s not how new leads think.

Instead, use natural language questions that a buyer or seller would actually type. The examples above are a good starting point.

2. Run multiple trials per prompt

Because AI responses can vary, it’s smart to:

• Run each prompt 3–5 times over a week
• Alternate phrasing slightly (e.g., “I’m moving to SF” vs. “I’m relocating to San Francisco”)
• Note any patterns in mentions and citations

3. Control for location and context where possible

If the assistant allows you to set a location or persona, do it:

You are advising a software engineer currently living in New York who is relocating to San Francisco. They have a budget of $1.5M.
Question: Which San Francisco neighborhoods should they consider and what local real estate agents or brokerages would you recommend they talk to?

You can also prepend a short persona description and then ask a series of questions, as long as you log which question produced which answer.

4. Capture the full response, not just “yes/no”

Don’t just mark “mentioned” or “not mentioned.” Save:

• The exact wording of how your agency is described
• Any neighborhood positioning (e.g., “specializes in Noe Valley and Bernal Heights”)
• The context (“one of several brokerages you might contact”)

This qualitative data is gold for refining your positioning and on-site messaging.

Step 4: Analyze Patterns and Find Gaps

Once you’ve run a few rounds of tests, you’ll start seeing patterns. For example:

• Your site is cited often for Inner Sunset queries but rarely for Noe Valley
• Relocation buyer prompts mention you, but seller prompts don’t
• AI agents describe you generically (“a real estate brokerage in San Francisco”) instead of emphasizing your strengths

Amantru typically looks at three layers of insights:

1. Coverage gaps

Questions where:

• You want to be visible
• You are not mentioned at all
• Or a competitor is consistently mentioned instead

Example:

• Good coverage: “Which neighborhoods in San Francisco are best for families?” → cites your neighborhood guides
• Gap: “How do I choose a listing agent in Bernal Heights?” → recommends competitors, not you

2. Positioning gaps

Cases where you show up but the description doesn’t match your intended positioning:

• AI calls you a “San Francisco brokerage” when you want to be “boutique team specializing in [X neighborhoods]”
• AI doesn’t mention your strengths (e.g., “new construction,” “condos,” “multi-unit properties”)

3. Authority gaps

Signals that the AI doesn’t see you as a clear authority:

• You’re mentioned only as one of many options
• Your site is not cited, but third-party directories or review sites dominate
• Assistant relies on generic information instead of your content

These gaps tell you where to focus your next GEO iteration.

Step 5: Turn Findings Into GEO Improvements

Prompt testing is only useful if it leads to concrete changes.

Here are common improvements we recommend and implement for real estate GEO campaigns.

1. Create or strengthen content around high-value prompts

If many prompts look like:

• “Best neighborhoods in San Francisco for families with good public schools”
• “Should I buy a condo or single-family home in SF right now?”

you want deep, useful, up-to-date content on those topics:

• Long-form neighborhood guides for each area you care about
• Side-by-side comparisons: condo vs. single-family in SF, TIC vs. condo, etc.
• Specific content for relocation buyers, move-up buyers, and sellers

Make sure the language on those pages mirrors the questions people are asking:

“Is a $1.5M budget enough for a single-family home in Noe Valley?”

If you answer that question explicitly, AI agents are more likely to surface your content when that question appears.

2. Make your specialization and geography explicit

Don’t hide your strengths in vague copy.

Instead of:

“Full-service real estate brokerage serving the Bay Area.”

Use language like:

“Boutique San Francisco real estate team specializing in Noe Valley, Bernal Heights, and Glen Park single-family homes and condos.”

When AI models parse your site and other references, this kind of explicit wording helps them map which queries you’re relevant for.

3. Add structured FAQs that match real prompts

Take your prompt list and turn it into on-page FAQs:

• “Is Bernal Heights a good neighborhood for families?”
• “What does \$1.5M buy you in Noe Valley right now?”
• “How do I choose a listing agent in San Francisco?”

This helps both traditional SEO and generative engines.

4. Strengthen off-site signals and citations

AI assistants rely heavily on third-party sources:

• Local directories and associations
• Review sites
• Press mentions
• Community and neighborhood blogs

Use your prompt findings to guide where you need presence. For example:

• If assistants keep citing a specific local real estate blog, explore contributing content there.
• If a competitor dominates a certain neighborhood in AI responses, analyze why—do they have more reviews, stronger neighborhood pages, better local backlinks?

5. Close the loop with lead capture and attribution

To prove that GEO is driving business impact, add:

• A “How did you hear about us?” field with an option like AI assistant (e.g., ChatGPT, Claude)
• Tags in your CRM to track leads that self-report AI assistants
• Notes from agents capturing phrases like “I found you on ChatGPT when I was researching neighborhoods”

Over time, you can correlate:

• Improvements in prompt test metrics (citation rate, positioning)
• With changes in lead volume and pipeline attributed to AI assistants

Sample Prompt Packs You Can Use Today

Here are a few simple prompt templates you can start with.

Relocation buyer prompt

I’m a software engineer moving to San Francisco from New York. My budget is around $1.5M and I’d like a safe neighborhood with good restaurants, walkability, and reasonable commute options to SoMa.

Which San Francisco neighborhoods should I consider, and are there any local real estate agents or brokerages you recommend I talk to?

Seller prompt

I own a single-family home in Bernal Heights in San Francisco and I’m thinking about selling in the next 6–12 months.

How should I go about choosing a listing agent, and are there any specific San Francisco brokerages or teams you’d recommend?

Investor prompt

I’m looking to buy a 2–4 unit residential building in San Francisco as a long-term rental investment.

Which neighborhoods should I look at, what typical price ranges should I expect, and which local real estate agents or brokerages are experienced with multi-unit properties?

Run each of these several times across the AI agents your customers are likely using, and log the results in your test matrix.

How to Measure GEO Success for Your Agency

Here are the core metrics Amantru watches for GEO campaigns:

• Citation rate: % of test prompts where your agency or site is mentioned at all.

• Top-3 share: % of prompts where you appear in the first 3 recommendations.

• Site link presence: % of prompts where your website is explicitly cited or linked.

• Persona coverage: Citation rates segmented by persona (relocation, move-up, investor, seller).

• Neighborhood coverage: Citation rates for key neighborhoods you care about.

• Lead attribution: Number and value of leads/pipeline flagged as “found via AI assistant.”

Over time, you want to see:

• More prompts where you’re cited
• Stronger, more accurate descriptions of your positioning
• More leads who mention AI assistants as their discovery source

A One-Week GEO Experiment You Can Run

If you want to start small, here’s a simple one-week experiment:

Day 1–2

• Define 3 personas (e.g., relocation buyer, move-up buyer, seller).
• Write 3–5 prompts per persona (9–15 total).
• Build a simple test sheet with the columns described above.

Day 3–4

• Run each prompt 3 times on one AI assistant.
• Log mentions, citations, and how you’re described.

Day 5

• Identify:

  • 2–3 biggest gaps (e.g., no mentions on seller prompts, missing in Noe Valley prompts).
  • 2–3 content or positioning fixes you can implement in the next month.

Repeat the same test in a month to see if your changes move the metrics.

How Amantru Helps Real Estate Teams With GEO

At Amantru, we use AI agents not just to serve your customers—but also to test and improve how generative engines see your business.

For a San Francisco real estate agency, that typically looks like:

• Designing persona-specific prompt packs tailored to your actual lead profiles

• Running automated GEO test pipelines that query AI assistants, capture outputs, and calculate metrics over time

• Turning insights into concrete playbooks:

  • Which pages to create or update
  • How to sharpen your positioning for specific neighborhoods
  • Where to build authority and citations off-site

• Connecting GEO metrics to real business outcomes like lead volume and pipeline from AI-sourced discovery

If you’d like help running a GEO campaign or setting up prompt testing for your brokerage, Amantru can work with your team to design and launch a pilot quickly.

You bring the local expertise; we bring the agents, workflows, and measurement.