When someone asks ChatGPT "who is the best auto insurance agent in Fresno," does your client's name come up? When a homeowner asks Claude "what insurance company covers wildfire damage in California," does the answer mention your client's agency? If you do not know the answer to these questions, you are already behind.
AI assistants are becoming the new front page of Google. Over 100 million people use ChatGPT weekly. Millions more use Claude, Gemini, and Perplexity as their first stop for recommendations and research. These platforms do not show 10 blue links. They give one answer. If your business is not in that answer, you do not exist in that channel.
We built a monitoring system that tracks exactly what AI assistants say about our clients, their competitors, and their industries. Here is how it works and why every business needs one.
The Problem: A Recommendation Channel You Cannot See
Traditional SEO is measurable. You can check your Google rankings, track keyword positions, monitor click through rates. You know where you stand because the data is visible.
AI recommendations are invisible. There is no "AI ranking" dashboard. There is no Search Console for ChatGPT. When a model recommends your competitor instead of you, you have no way of knowing unless you manually ask the same question yourself. And even then, AI responses vary based on context, phrasing, and model version.
This means businesses are losing potential customers to AI recommendations they never see, and they have no strategy to influence those recommendations because they do not even know what is being said.
We built a system to make the invisible visible.
The Architecture: 5 Agents, 4 Platforms, 50 Prompts
Our monitoring system for a California insurance agency consists of five Python agents that run automated queries across four AI platforms, scoring and tracking brand mentions over time.
Agent 1: Prompt Generator. Maintains a seed library of 50 prompts across 6 categories: auto insurance, home insurance, wildfire coverage, earthquake coverage, renters insurance, and brand queries. Each category has prompts at different specificity levels. Generic ("best auto insurance in California"), local ("who sells car insurance in Fresno"), and brand ("tell me about [Agency Name]"). The prompt generator rotates through these on a daily cycle, varying phrasing slightly to avoid caching effects.
Agent 2: Query Runner. Executes prompts against ChatGPT (via API), Claude (via API), Gemini (via API), and Perplexity (via API). Each query is logged with the exact prompt, timestamp, platform, and model version. The agent handles rate limiting, retries, and error logging.
Agent 3: Brand Mention Detector. Parses every response for mentions of our client's agency name, agent names, phone numbers, addresses, and website URLs. Also detects mentions of 12 tracked competitors. Uses fuzzy matching to catch partial mentions, abbreviations, and name variations. Scores each mention on a scale of 1 to 5: 1 is a passing reference, 5 is a direct recommendation as the top choice.
Agent 4: Sentiment Analyzer. Evaluates the context around each brand mention. Is the AI recommending this business? Warning against it? Mentioning it neutrally in a list? This matters because a mention is not always positive. We have seen cases where an AI assistant mentions a business in the context of a negative review or a warning.
Agent 5: Dashboard Generator. Aggregates all data into a Flask web dashboard that updates daily. Shows brand mention frequency by platform, sentiment trends over time, competitor comparison, and prompt category performance. The client can see at a glance which AI platforms mention them, how often, in what context, and how they compare to competitors.
The Prompt Seed Library
The 50 prompts are not random questions. They are carefully designed to mirror how real consumers ask AI assistants about insurance.
Auto insurance prompts (10): - "Who has the cheapest car insurance in Fresno?" - "I need auto insurance for a teenage driver in California" - "Best insurance agent near me for a clean driving record" - "What is the minimum car insurance required in California?" - "I got a DUI, who will insure me in Fresno?" - Plus 5 more variations covering specific scenarios
Home insurance prompts (10): - "Who sells homeowners insurance in the Central Valley?" - "Best home insurance for older homes in Fresno" - "My home insurance was cancelled, who can help?" - "How much is home insurance in Fresno California?" - Plus 6 more variations
Wildfire prompts (8): - "Who insures homes in California wildfire zones?" - "My insurance dropped me because of wildfire risk, what do I do?" - "Best insurance company for wildfire coverage in California" - Plus 5 more variations
Earthquake prompts (7): - "Do I need earthquake insurance in Fresno?" - "Who sells earthquake insurance in California?" - Plus 5 more variations
Renters insurance prompts (7): - "Cheapest renters insurance in Fresno" - "Do I need renters insurance in California?" - Plus 5 more variations
Brand queries (8): - "[Agency Name] reviews" - "Is [Agency Name] a good insurance agency?" - "[Agent Name] insurance Fresno" - Plus 5 more variations including competitor brand queries
What We Found: The First 30 Days
The initial scan revealed patterns that traditional SEO would never show.
ChatGPT mentioned our client by name in 23% of relevant prompts. Not bad, but it means 77% of the time, someone asking ChatGPT for insurance recommendations in Fresno is not hearing about our client's agency. The mentions were concentrated in brand queries (naturally) and wildfire coverage (where the client has strong content).
Claude mentioned the client in 31% of prompts. Higher than ChatGPT, particularly for local queries. Claude appeared to weight local business directories and review sites more heavily.
Gemini mentioned the client in 18% of prompts. The lowest of the four, but Gemini heavily favored businesses with strong Google Business Profile data.
Perplexity mentioned the client in 42% of prompts. The highest, likely because Perplexity explicitly retrieves and cites web sources, and our client has strong web presence.
Competitor analysis: The leading competitor appeared in 38% of ChatGPT responses, significantly ahead of our client's 23%. That gap represented a measurable competitive disadvantage in the AI recommendation channel.
The GEO Strategy: Getting AI to Recommend You
GEO stands for Generative Engine Optimization. It is the practice of structuring your online presence so that AI assistants are more likely to mention and recommend your business. Here is what we implemented based on the monitoring data.
FAQ content matching AI prompt patterns. We analyzed which prompts generated zero mentions and wrote content specifically answering those questions on the client's website. When someone asks ChatGPT "my insurance was cancelled because of wildfire risk, what do I do?" and ChatGPT does not mention our client, we create a page that directly answers that question with specific, helpful information.
Schema markup. AI assistants that retrieve web data (especially Perplexity and Gemini) parse structured data. We added comprehensive schema markup to every page: LocalBusiness, InsuranceAgency, FAQPage, Review, and Service schemas. This makes it easier for AI crawlers to extract and attribute information.
Citation optimization. AI models learn from and retrieve data that is widely cited across the web. We ensured the client's NAP (Name, Address, Phone) data was consistent across 40+ business directories. Inconsistent NAP data confuses both traditional search engines and AI models.
Review volume and recency. AI assistants heavily weight review data when making recommendations. We implemented a systematic review request process that increased the client's Google review velocity from 2 per month to 8 per month. More reviews mean more training data that includes the business name in a positive context.
Content depth on specialty topics. The monitoring showed that AI assistants were more likely to recommend businesses with deep content on specific topics rather than thin content across many topics. We doubled the client's wildfire coverage content from 2 pages to 6 pages, each addressing a specific scenario.
Results After 90 Days
After implementing the GEO strategy and continuing to monitor:
- ChatGPT mentions: 23% to 34% (11 point increase)
- Claude mentions: 31% to 39% (8 point increase)
- Gemini mentions: 18% to 29% (11 point increase)
- Perplexity mentions: 42% to 51% (9 point increase)
- Competitor gap on ChatGPT: 15 points behind to 4 points behind
The wildfire coverage prompts showed the most dramatic improvement. After publishing detailed wildfire FAQ content, ChatGPT began recommending our client as the first or second option for wildfire related queries. This directly correlated with the content we published.
How to Build Your Own Monitoring System
You do not need to build the full 5 agent system to start. Here is the minimum viable approach:
- Write 10 prompts. Think about what your ideal customer would ask an AI assistant about your industry and location. Include brand queries, generic queries, and competitor queries.
2. Test manually across platforms. Ask each prompt on ChatGPT, Claude, Gemini, and Perplexity. Screenshot the results. Note whether your business appears, in what context, and who else is mentioned.
3. Create a spreadsheet. Track date, platform, prompt, whether you were mentioned, mention quality (1 to 5), and which competitors appeared. Run this monthly.
4. Identify gaps. Which prompts never mention you? Those are your content priorities. Write content that directly answers those questions.
5. Automate when ready. Once you understand the patterns, invest in automating the monitoring with API calls and scheduled scripts.
If You Do Not Know What AI Says About You, You Are Flying Blind
Five years ago, a business that did not track their Google rankings was at a disadvantage. Today, a business that does not track their AI recommendations is in the same position. The channel is different, but the principle is identical: you cannot improve what you do not measure.
AI assistants are not a future trend. They are a current reality that is actively shaping which businesses get discovered and which ones get ignored. The businesses that monitor and optimize for AI recommendations now will have a compounding advantage over those that wait.
We build monitoring and optimization systems for businesses across every vertical we serve. If you want to know what AI says about your business, and more importantly, how to influence it, we should talk.