Pulse AI NYC X CASTGNY Hackathon

GEO Agent – Generative Engine Optimization for the AI Era This hackathon is about building the GEO Agent, an AI-powered system that measures and improves a brand's Share of Model (SoM) — the percentage of AI-generated answers in which a brand is mentioned. This is a live-fire hackathon challenge with strict live data requirements. 1. The New Reality: From SEO to GEO For 20+ years, brands have competed for Google's 10 blue links (traditional SEO). Today, users increasingly ask AI models directly:
"What are the best international business programs?"
"What's the best CRM for startups?"
"What consulting firm helps international students get U.S. jobs?"
LLMs like ChatGPT, Gemini, Perplexity, and Claude return one synthesized answer.If your brand is not in that answer, you effectively do not exist in that user's consideration set. This gives rise to a new metric:

Share of Model (SoM): The percentage of AI-generated answers in which your brand is mentioned.
Your mission: Build the first version of a tool that measures and improves a brand's SoM. 2. Problem Statement Today, brands generally:

Don't know whether AI models mention them

Don't know which competitors are being recommended

Don't know why those competitors rank

Don't know what content to create to fix it
You are building the GEO Agent — a live AI consultant that:
Audits a brand's Share of Model

Explains why the brand and competitors rank as they do
Recommends concrete actions and content to improve SoM
3. Brand Scope – Choose ONE Real Brand Pick one real company with an actual digital footprint:

Option A: Global Brand

Examples: Nike, Coca-Cola, Airbnb, Stripe, etc.

Option B: Tech Startup

Examples: Linear, Notion, Retool, Webflow, Ramp, etc.

Option C: Education / Program Brand

Examples: Semester at Sea, General Assembly, NYU Florence, etc.

No fictional companies. 4. GEO Agent MVP Requirements The GEO Agent must convincingly answer two core questions:
How well is my brand performing in AI searches?
How can we improve our positioning?
4.1 How Well Is My Brand Performing in AI Searches? The agent must:

Generate at least 5 high-intent queries relevant to the brand

Example: "Best international business programs"
Example: "Top tools for product teams"

Query at least 2 AI systems or search tools live (LLM APIs, Perplexity, web search + synthesis, etc.)
From these responses, the agent must extract:
Whether the brand is mentioned
Which competitors are mentioned
Position / order in the answer

Sentiment (positive, neutral, negative)
Compute a simple Share of Model score:
% of queries where the brand appears
Relative frequency vs competitors
And surface evidence, including:
The actual raw responses

URLs used

Citation references (e.g., which sources supported which conclusions)
4.2 How Can We Improve Our Positioning? The agent must:

Analyze why competitors are ranking, e.g.:

Wikipedia presence?
Domain authority?
More backlinks?
Structured data / schema?
Press mentions?

Identify content gaps, e.g.:

Missing comparison pages?
No listicle presence?
No third-party validation?

Then automatically generate:
A draft Wikipedia-style summary (if appropriate)
A comparison page outline

A suggested SEO / GEO content strategy

Structured data (schema) recommendations
Finally, prioritize actions, such as:
High impact / Low effort
Medium impact / High effort
This must be programmatic, not manually typed advice. 5. What Makes This an AGENT (Not Just a Script) The GEO Agent should:
Dynamically generate queries

Loop over tools (LLM APIs, search APIs, scraping, etc.)
Decide next actions based on intermediate results

Aggregate findings across queries and systems
Produce a structured report

Nice-to-have / bonus capabilities:
Tool calling / tool orchestration
Retry and fallback logic
Model comparison (e.g., different LLMs / search engines)
Autonomous research loops
Monitoring / re-runnable pipeline over time
6. Expected Demo Format At demo time, you must show: Input:
Brand name
Target customer segment (optional)
A live run of the agent
Output:
Share of Model score
Competitor comparison table

Evidence citations (links, quoted passages, raw responses)

Action plan (prioritized recommendations)

Generated content example (e.g., wiki summary, comparison page outline)

Architecture overview (2–3 minutes)
7. Judging Criteria Reliability (30%)
Uses live data and real APIs
Provides proper citations

Produces repeatable results
Shows transparent reasoning (how conclusions were reached)
Insight Quality (25%)
Findings are non-obvious and interesting
Competitor analysis is real, not generic
Strategy logically follows from evidence
Technical Robustness (25%)
Quality of tool orchestration

Level of automation

Error handling strategy

Model usage sophistication (e.g., multi-step reasoning, multiple providers)
Clarity to the CEO (20%)
Non-technical execs can understand the output

Report is structured and easy to scan
Strategy is concrete and actionable

8. What Winning Teams Usually Do Winning teams typically:
Build a query generator → evaluator → synthesizer pipeline
Compare multiple LLMs / search providers

Extract structured entities (brands, competitors, rankings, sentiment) from responses
Build a scoring framework for Share of Model
Produce a polished, "consultant-style" report

Teams that struggle often:
Just ask a single LLM once
Manually copy/paste results
Skip citations

Over-focus on UI instead of reasoning and evidence
9. Optional Advanced Layer (If Time Allows) If the core MVP is complete, consider:
Tracking Share of Model over time

Detecting misinformation (incorrect statements about the brand)
Monitoring new competitor emergence

Suggesting PR placements and outreach opportunities
Building a lightweight dashboard for ongoing monitoring