PRODUCT
Pay directly from the pricing page
Plans can be bought without talking to anyone. The price you see and the price you are charged come from the same published table, read at the moment you click, so the two cannot drift apart.
ChangelogWhat changed,
What changed,
week by week.
Product changes and changes to the methodology, dated. A change to how something is measured is listed like any other.
METHODOLOGY
Questions that name your brand stop inflating your numbers
Asking "what does this brand offer?" always mentions the brand: that is the question, not a finding. Those questions now feed what AI understands about you and are kept out of citation rate, share of voice, share of recommendation and average position. They also stop counting against your question limit.
MEASUREMENT
A brand is found under every spelling it answers to
Otterly, Otterly.ai and OtterlyAI are one competitor, not three. Spellings are now derived automatically, with two guards: a name that is also a common word generates nothing, and the stem needs four letters.
MEASUREMENT
What the assistant says it searched for
When an assistant is measured inside its app it shows what it was doing, not the query it sent. Those steps are now captured and the subject of each one is extracted, so you can see what the buyer's words were, not your catalogue's.
PRODUCT
Inception: declare what you want AI to say
The product measured where you stand and had nowhere to keep where you want to get to. You can now set a target on a question, or a value proposition across many questions, and the following week says whether it moved. A target the measurement cannot verify reads as "not measured", never as "not reached".
PRODUCT
An alert when your brand stops being named
The week you disappear is the one you want to hear about first, and it was the one thing the alerts did not say. Three guards keep it honest: zero after zero is a state and not an event, a minimum of five answers per engine, and a week with no measurement is never reported as an absence.
MEASUREMENT
ChatGPT measured inside the app, as a customer sees it
The searched half of ChatGPT is now collected in a real signed-in session, from the market being measured, next to Claude and Meta AI. The remembered half stays on the API, because search cannot be turned off inside an app and a real session could never measure the training corpus.
PERFORMANCE
The weekly audit runs in minutes, not hours
Fifty questions used to take sixteen minutes of queuing because the runs went one at a time. They now run in parallel, with a deadline on each call and a guard that never pays twice for the same question.
METHODOLOGY
Share of recommendation: appearing is not being chosen
Citation rate asks whether you appear. Share of voice asks what slice of the mentions is yours. Neither tells being one of ten from being the answer. Share of recommendation counts the choices, on the same scale, and what counts as a choice was fixed in writing before 616 answers were re-read.
MEASUREMENT
Share of voice is computed, never typed
The figure is arithmetic over the week, and for two weeks it had been counted over the batch being read. A competitor showed at 38.5% where the measurement gives 6.2%. It is now computed in one place, refreshed nightly when it drifts, and a short denominator raises a flag.
MEASUREMENT
The Claude a buyer sees is the Claude we measure
The API and the app are not the same product. Measured side by side on the same question: 7 sources against 16, and only 24% of the brands named in common. The app is now collected in a real session, and every measurement declares which way it came.
WHEN A CHANGE TO HOW SOMETHING IS MEASURED CHANGES YOUR NUMBERS, IT IS LISTED HERE WITH THE DATE, AND THE WEEKS BEFORE IT ARE NOT SILENTLY REWRITTEN.