AI · SaaS · 2026 · Product · Full-Stack
Tracks how ChatGPT, Gemini & Google see a brand — then coaches the climb.
Stack: Firebase · Gemini API · OpenAI API · Perplexity API · Serper · Google PSI
A full-stack SaaS answering the question every brand is starting to ask: when AI engines talk about my market — am I in the answer?
Brands can track Google rankings, but AI answers from ChatGPT, Gemini and Perplexity are a black box. No practical tooling existed to measure AI visibility — let alone a system that tells you how to improve it.
The core design decision came from how AI answers actually behave. When someone asks ChatGPT or Gemini for a recommendation, the engine doesn't return a list of links — it typically compresses everything it can read and verify into a few confident names. You are either in that answer or you don't exist, and until now there was no dial showing which. So I refused to build another audit PDF. The whole architecture bends toward one output: a visibility score and a mention rate you can watch move over time, shown side by side with Google — because businesses already have an intuition for rankings, and pairing the two is designed to make the new metric legible at a glance.
Scanning one engine would have been easier — and misleading. ChatGPT's browsing has typically leaned on Bing's index, Gemini draws on Google's, and Perplexity is built around live web retrieval, so a brand can be present in one answer and absent from the next. That is why the scanner runs across ChatGPT-, Gemini- and Perplexity-style engines, built on the OpenAI, Gemini and Perplexity APIs, with Serper supplying classic Google results for the side-by-side comparison and Google PSI covering the technical health of the pages themselves. Behind that sit the prompt pipelines and scoring logic that turn raw model output — verbose, unstructured, rarely identical from one run to the next — into numbers stable enough to compare from one scan to the next.
The coaching half of the name is the part I care most about. A score without a next step is just a prettier way to feel anxious, so every scan ends in a prioritized action plan — fixes ranked by impact and written to be implemented, not discussed. Where a fix is a file, the tool builds the file: JSON-LD schema so engines know what your pages are, llms.txt to hand AI assistants a curated map of your site, robots.txt to govern crawling. That generator step is a deliberate stance. Most SEO deliverables die somewhere between the report and the developer; here, the artefact ships.
Firebase carries the workflow: it holds each scan's results and the progress state that turns a list of fixes into a coached process. Scan, score, fix, re-scan is only a loop if the system remembers the last pass — the history is what makes the re-scan meaningful. The prompt pipelines, scoring logic, dashboards and billing were all designed and built by me, end to end.
And one thing the product deliberately does not do: promise rankings. Nobody controls Google's index or an AI model's answers, and any tool guaranteeing a ChatGPT mention is selling certainty it does not have. What the Coach offers instead is measurement — a before and after you can trust — and the inputs you can actually control: being indexed, corroborated and quotable.
All projects — Emad Yahya, web developer in Dubai · Live project · Next project: LKYOOB — Private Residences · emaadyahya4@gmail.com