Vizor

Inspiration

My mom runs her own private dental practice called DoveDentalSmiles. Given the rise of people searching for services like dental cleanings on LLMs, I realized that small businesses like hers lose their most important marketing factor - visibility. Instead of getting potential prospective clients, the industry giants continue to prosper off of their brand recognition, pushing out smaller businesses from getting a larger market share. Big businesses are able to pay thousands a month to optimize their advertising with SEO(search engine optimization) and GEO(generative engine optimization), which prioritize website visibility in the LLM age. So you must be thinking, why doesn't my mother just do the same? Well, when it comes to optimizing for LLM visibility, its simply not a viable option for her company since the existing proprietary solutions are so expensive ($2k/month). Essentially, you can think of it this way: the rich keep getting richer, while people like my mother get left behind. However, with Vizor, I found a way to level the playing field.

What it does

Vizor is an affordable generative engine optimizer that measures and improves visibility metrics [position-adjusted word count, citation-share-of-voice, number of citations, and tonality analysis] for individual websites on LLMs. Vizor determines a user's given visibility in LLM outputs relative to competitors, and gives the user the freedom to choose actions (e.g., add a FAQ section, integrate relevant statistics, etc.) that best optimize future visibility. Each potential action gets a visibility score, which is calculated by simulating how the changed content shows up in LLM output assuming competitor sources stay the same (A/B testing).

How we built it

We used GPT-5 to generate potential relevant user queries based on the semantic content of the user's article, and we ran GPT-5 mini for the main model outputs, primarily for speed. Based on some relevant papers, we calculated visibility metrics like Position-Adjusted Word Count and Citation-Share-of-Voice, which are metrics we display on the dashboard and factor into our overall visibility ranking. We used a fine-tuned BERT rating model for the sentiment analysis (how the product is discussed in LLM output). We used Common Crawl to get the raw text from given URLs. For the optimization side, we used the visibility metric pipeline to determine the estimated impact on visibility of 12 different predetermined actions.

Challenges we ran into

Took us forever to settle on an idea. We also had a very convoluted project structure at first that we simplified to ensure we finished on time.

Accomplishments that we're proud of

Understanding some pretty technical papers, and whiteboarding our entire technical spec.

What we learned

Ways to optimize content for generative engine outputs. Without a very detailed and specific technical spec, AI IDEs can get quickly misaligned.

What's next for Vizor

Building a better web scraper that is able to get more metadata from websites. Common crawl doesn't give us much to work with. Instead, we would build our own BeautifulSoup parsing algorithim. We would also add support for more search engines besides ChatGPT. Hopefully, we can deploy it to a live website soon.