Provenance on every panel
Each panel states whether its data came from the tracker, Google Analytics, Search Console or a placeholder. Sample data is labeled until real runs replace it.
A method you can check.
Robin Knows is a measurement, not an opinion. The same questions go to the same engines on a fixed schedule, and every answer is kept. Here is exactly what happens between your brand name and the number on the dashboard.
1
Write the buyer questions
We start with about 50 prompts, written the way a buyer asks an assistant, not the way an SEO tool lists keywords. They cover discovery, comparison and vertical-specific needs for your category.
The set is versioned and does not change mid-quarter. Editing the questions would break comparison with earlier months, so new questions arrive as a new version with a clear break in the trend.
2
Ask each engine, with search on
Every prompt goes to ChatGPT and Claude with live web search enabled, and to Google AI Overviews, which we sample on its own schedule. Perplexity and Gemini follow.
Each prompt runs three times per engine and the results are averaged. Model answers are random by nature, and one run is an anecdote. Three runs smooth the noise without hiding it.
3
Parse every answer
A language model reads each answer and returns structured data: which brands appeared, in what order, with what sentiment, and which sources were cited.
The original answer is stored next to the parsed result. If a number looks off, you can open the exact text it came from.
4
Compute the metrics
Mention rate, share of voice, citation rate, position, sentiment and the gap report are calculated from the structured data, against a watchlist of you and your competitors.
These are plain calculations, not another model's opinion. The same inputs always give the same outputs.
5
Read it, and act on it
You get a dashboard for the full picture and a monthly readout for the headline: where you stand, what changed, and what to do next.
The gap report turns into a content plan. The citation report shows which of your pages the engines already trust.
The parts that keep the numbers honest.
No false comparisons
If there is no comparable earlier period, we say so rather than print a percentage change against a partial window.
Each engine on its own terms
Google AI Overviews is sampled monthly over a subset of questions, so it is reported as its own strand and never blended into the headline.
Things people ask us.
Not here? Mention it when you join the beta.
Which AI engines do you measure?
ChatGPT and Claude with live web search, and Google AI Overviews. Perplexity and Gemini are built behind the same interface and will be switched on as they are validated.
Is this the same as SEO rank tracking?
No. There is no ranked list to track. An answer is generated for each question, so we measure how often you are named and cited across a fixed set of questions instead.
Why not just ask ChatGPT myself?
You will get one answer, shaped by your phrasing, and a different one tomorrow. A fixed question set, repeated runs and a stored history is what turns an anecdote into a trend.
Do you edit or summarize the answers?
Never. Verbatim engine quotes are evidence, so they are shown exactly as returned. Summaries and metrics sit beside them, never in place of them.
What does the beta include?
We are working with a small group of teams first. Join the list and tell us about your brand; we will confirm what the beta covers for you when we reach out.
Find out what the assistants say about you.
We are onboarding a small group of teams first. Tell us about your brand and we will be in touch about a first report.
Read by a person.
We look at your brand and category, then email you if there is a fit. No payment details are collected.