Discovery begins with the questions, not the brand
At SaidTrue we treat discovery as a listening exercise. Before we look at anything else, we look at the questions. Not brand-name lookups, but the questions a customer already has in mind: who should I use for this, and is this business reputable. Our own site makes the point plainly — Google isn't dying, but it's no longer the only front door, because customers increasingly ask AI who to use and whether a business is reputable. Discovery is finding out what comes back when they do.
How we work through it, step by step
We fix the question set first, so the same questions go to ChatGPT, Gemini, Perplexity and Claude. Then, for every answer, we record two things: what was said, and what was sourced — which pages an engine leaned on to reach its conclusion. That second half carries most of the signal. An answer with no visible source is only a claim, and an answer built on the wrong page is a claim that will keep repeating.
With the answers and their sources side by side, we compare. The engines do not agree with each other, and that disagreement is itself a finding: the same business can be described differently depending on which assistant a customer happens to ask. We then score the result using the method set out in How SaidTrue Scores AI Answers, and publish it as a public scorecard — what was said, what was sourced, and where the truth diverges. Monthly AI monitoring follows, because these answers are not static.
Why we do it in that order
Because the difficulty is context, not retrieval. Atlan's Discovery documentation puts the general problem well: "Most data teams don't have a search problem. They have a context problem: assets exist, but nobody knows what they mean, who owns them, or whether they can be trusted." The same holds for an AI answer about a business. The business exists online; the open question is whether an engine can say what it means, who stands behind it, and whether it can be trusted.
Working from questions to answers to sources to score keeps the report auditable. Anyone can read the question, read the answer, follow the citation and see the gap for themselves. That is why SaidTrue publishes its scoring method rather than asking anyone to take a number on faith — a visibility report is only worth the reasoning you can check underneath it.
References
- Discovery | Atlan Documentation — docs.atlan.com