Track Your True Share of Voice in AI Search

SaaS marketing teams rely on tools like Ahrefs and SEMrush to measure visibility. These platforms are excellent for tracking keyword rankings on a traditional search engine results page (SERP). The problem is that AI search engines like Perplexity or Google's SGE do not just present a list of links; they synthesize an answer and cite sources. Your content can be the primary source for an answer even if your page ranks #5 in the organic links. Conversely, you can rank #1 and be completely ignored by the AI-generated answer. Standard SEO tools cannot see this distinction.

Consider this scenario: A B2B SaaS company targets the question “how to calculate customer lifetime value.” Ahrefs shows their guide is ranking at position #3, a respectable outcome. However, when a user asks Perplexity, the AI generates a three-paragraph answer that directly quotes a competitor's blog post and cites it twice. The SaaS company's guide is listed as a minor source at the bottom. Their SEO dashboard reports a win, while in reality, they have zero share of voice in the answer itself. This creates a critical blind spot in their marketing analytics.

The structural failure is that these tools are fundamentally SERP scrapers. They are designed to parse the predictable HTML of a Google results page. AI-generated answers are often dynamic components loaded via JavaScript or returned in complex JSON payloads that these scrapers cannot interpret correctly. They were built for a world of ten blue links, and their architecture is not suited for parsing the output of a large language model. They track document retrieval, not answer synthesis, which is a fatal flaw in the age of AI search.

We built a share of voice tracking system as a core component of our own four-stage AEO pipeline. The approach bypasses traditional SERP scraping entirely and queries the AI engines directly. For a SaaS client, the first step is to define the 100-500 commercial-intent questions that drive qualified traffic. This list becomes the basis for the tracking system.

The technical system is a set of Python scripts running on a schedule using GitHub Actions. Every 24 hours, the script uses the `httpx` library to send asynchronous requests to the Brave Search API and other endpoints that provide programmatic access to generated answers. The JSON responses are parsed to extract both the text of the answer and the list of cited source URLs. We store this raw data in a Supabase Postgres database, creating a historical record of which domains are cited for which questions over time.

With the data collected, a separate process calculates share of voice. It aggregates citation counts by domain for each tracked question, showing a daily trend. The delivered system is a simple dashboard built with Streamlit or a direct connection to your data warehouse. You see exactly which of your pages are being used as sources and which competitors are gaining traction. The entire system runs on AWS Lambda for less than $50 per month, providing data that multi-thousand-dollar-per-month SEO suites currently miss.