Unscripted SaaS · AI visibility
Most Brands Have Never Read What AI Says
with Somya Goyal — founder of Stellarcast, building from India for global markets
Every SaaS team has a view on what customers think of them, backed by surveys, reviews and sales calls. Almost none have the equivalent for the intermediary that now sits between a buyer and a shortlist. Somya Goyal founded Stellarcast after fifteen-plus years in quality engineering — eight of them at Accenture — and the problem she describes to host Jeremy Rivera is not a technology gap. It is that nobody has looked.
The Blind Spot Nobody Audits
Stellarcast’s premise is simple to state. When someone asks ChatGPT, Claude, Perplexity or Gemini “what’s the best X?”, the model names a couple of brands. Somya’s description of what her product does is really a description of what most companies are not doing: it “watches whether you are one of them.”
She frames the output as “a system of AI records: that you are being cited, that you are being considered, and that you are being chosen.” Three distinct states — and most brands cannot currently tell you which of the three they are in for any given buying question.
She also draws a line that matters for anyone importing SEO habits into this: “Ranking is different and citation is different. AI citations overlap Google’s top-ten links by only fourteen percent — which I read in the last few months.” She is careful to attribute that as something she read rather than something she measured, and we are keeping that qualification.
Sometimes AI Is Confidently Wrong
“When you want to measure whether the model says your name or not — most brands have never read what AI says about them, and sometimes AI is confidently wrong. It cites data that’s maybe two years old — the pricing, the reviews — even after a lot of fresh data. Where it’s picking that from is something I’m also watching.”
— Somya Goyal, StellarcastThe examples she picks are the two that cost money directly. Stale pricing means a buyer is comparing you against a number you no longer charge. Stale reviews mean you are being represented by a version of your product two years of engineering ago.
“Confidently” is the operative word. A search engine that lacks current information returns nothing, and the absence is visible. A model that lacks current information generates a fluent, plausible answer, and the error is invisible unless you go looking.
Jeremy offered a framing from another guest that fits well: Matt Brooks describes ChatGPT as your least-trained customer support representative — one you cannot train without first listening to the answers it gives people. Somya agreed with the shape of it, adding that the first question is simply presence: “if a user asks about, say, cruelty-free shampoos, are you present or not in the AI answers? That’s the first question. And if not, why?”
Most Tools Stop At The Dashboard
Her sharpest product criticism is aimed at her own category:
“Most of the tools stop at the dashboard. The whole point of Stellarcast is closing the loop — keeping things in a log and moving continuously, rather than a one-time audit or one-time fix.”
— Somya Goyal, StellarcastThe loop she describes runs on four specialist agents: “So it is basically monitor, diagnose, execute, and then prove.” One monitors every engine for how you are cited versus competitors. One diagnoses why you are missing — including which specific model cannot find you, which matters because “you don’t know which user is using which model.” One drafts the fix and, once approved, ships it. The last re-checks the engine and verifies the lift.
Her argument for why this cannot be a one-off is the most portable idea in the interview: “Maybe today the LLM answers in a certain manner, but two or three days down the line it might change its background search, name your competitor, and cross you out.” Your visibility can be revised without anything on your side changing — which makes a point-in-time audit a snapshot of a surface that does not hold still.
Closing The Loop Between Finding And Fixing
Somya is candid that the company is early: “Currently it’s early and building in the open, so we are onboarding a focus group of pilot brands and agencies as design partners.” She is deliberately not publishing pricing yet — “I want to focus on the brands right now, in pilot mode” — and shares what she learns on LinkedIn as she goes.
She is equally clear that automation does not remove the operator: “The human in the loop is always needed.” The product drafts fixes; a person approves them before anything ships to a WordPress, Shopify or Webflow site.
Asked how quickly models update after a change, she declined to overclaim: “That’s something I am watching too. I don’t have concrete comments on that yet.” Jeremy offered his own observations — a community cleanup listing recognised by Copilot in about twenty minutes, other brand updates taking a full day, events appearing almost immediately — and suggested treating the surface as “more liquid” than search. Query deserves freshness, he argued, seems to run unusually high in these systems.
A Monthly AI Mention Audit You Can Run Free
Practitioner guidance from us. Somya describes the product loop; the manual version below is our adaptation of it.
You do not need tooling to start, and the exercise is worth doing once before you buy anything.
Write ten buying questions in your customers’ words. Somya’s observation about how questions changed is the key input: people no longer ask “which is the best brand?” — “they ask ‘will it do the X number of things for me?’” Write capability questions, not category questions.
Run all ten across at least three assistants. Her point that different models “find things in different ways, understand things in different ways, and respond in different ways” means a single-model check tells you very little.
Record three things per answer: were you named, were you described accurately, and what was cited. That maps onto her cited / considered / chosen distinction, and the third column is where the actionable work usually is.
Repeat monthly, in a log. The log is the deliverable. A single run tells you where you stand; a series tells you whether anything you did mattered — which is the “prove” step, done by hand.
Her plain-English target for all of it: “Visibility in AI is simple. Make your brand easy for AI to find, understand, and read. That is the new name of AEO, in very simple English.”
Full recap: Somya Goyal on getting your brand cited by ChatGPT, Claude and Gemini. Stellarcast is at stellarcast.ai, in early access.
More on the content side of this in SaaS SEO and content, and a contrasting view on the framing itself in why “ranking in an LLM” is the wrong frame.