Build an Automated Share of Voice Tracking System for AI Search

Education marketing teams often rely on traditional SEO tools like Ahrefs or SEMrush. These platforms are built for classic search engine results pages, tracking keyword rankings of blue links. They can report that your university's page for 'best online MBA programs' is at position #2, but they cannot see if Google's SGE or Perplexity is citing three of your competitors in the generated answer for that same query.

A marketer might try to supplement this with brand monitoring tools like Brand24, but these are designed for social media and news articles. They scrape public, static HTML pages and do not programmatically interact with AI chat interfaces. This approach misses the entire dataset of AI-generated answers, creating a massive blind spot where prospective students are forming opinions.

Consider a university marketing team tracking their new data science certificate. Their SEMrush report shows strong rankings, but enrollment is flat. The problem is that when a user asks an AI assistant about top certificates, the generated text summarizes a competitor's program and links to them directly. The team has no visibility into this lost impression because their tools are measuring an obsolete format. They spend budget optimizing for a SERP that fewer users see.

The structural issue is that existing tools are built on web crawlers, which are designed for a world of static documents. AI search engines are interactive applications that generate content in real time. Measuring visibility requires a system that can manage sessions, submit prompts, and parse unstructured, conversational text. This is a data pipeline problem, not a simple ranking check.

The first step is a discovery process to map the queries that drive enrollment. We would work with you to identify the critical 'brand vs. brand' and 'brand vs. topic' searches, from specific degree programs to general queries like 'is an online degree worth it?'. We also define the target AI engines (Google SGE, Perplexity, Copilot) and the key competitors to track. This audit produces a concrete tracking specification and a data schema for the results.

We would build a Python system scheduled by GitHub Actions to run these queries daily. For AI engines with official APIs, the system uses the `httpx` library for efficient, asynchronous calls. For those without, it uses Playwright for browser automation to simulate a real user's query and capture the full response. The raw JSON or HTML outputs are stored in a Supabase database using pgvector for semantic analysis of the generated text.

A parsing layer then processes the raw data, extracting your brand mentions, competitor mentions, and any cited URLs. This structured data feeds a live dashboard built in a tool like Metabase, which you own completely. The system shows SOV trended over time and can trigger alerts for significant shifts. You receive the full source code and a runbook, and the system typically runs for under $50/month on your own cloud infrastructure.