Free reading · updated daily
How much the answers AI assistants give moved from one day to the next. One line across every question we ask them — not per assistant, not per topic. The equivalent of a search volatility tracker, for the systems that answer instead of listing.
Today
9
Ordinary movement — 2,578 comparisons
Seven-day mean
6
the band the last week sat in
Highest in the window
15
Aug 4
Answers with citations
5%
of comparisons had sources on both days — the thinnest of the three inputs
A hollow point is a day measured on fewer than 300 comparisons; the line skips it. 61,512 · 5,794 · 45
The scale is the movement itself, not a grade. A low reading is not good news and a high one is not bad news — it is the weather, and what it changes is how much you should read into a single day of your own results.
0–5
The assistants answered yesterday's questions much as they did the day before. A change in your own visibility on a day like this is about your pages, not the field.
6–12
Where most days sit. Individual answers shift — a citation swapped, a brand dropped from one of three — without the field moving as a whole.
13+
Answers changed widely enough that a single day's reading of your own brand should not be treated as a verdict. Wait for the next one before rewriting anything.
Movement a reader would call a different answer. Deliberately not the wording: a model rephrases itself every time it is asked, and a line counting that would read a hundred every day and mean nothing.
The set of domains an answer pointed at, compared against the same question's answer the day before. This is the closest equivalent to what a SERP volatility tracker measures.
Named yesterday and absent today is the largest single move one answer can make, and the one a brand actually feels.
Where an answer ranks or orders what it names, capped at five places — beyond that, one answer cannot move further in a way that means more.
A daily measure of how much AI assistants' answers changed from one day to the next. It is calculated by asking the same questions of the same assistants every day and comparing each answer with the previous day's: the sources it cited, whether a brand was named, and how far that brand's position moved. The reading runs 0–100, and most days sit between 5 and 12.
A SERP tracker measures how far ranked results moved. This measures how far generated answers moved, which is a different thing: an answer has no positions to swap, so what changes is which sources it draws on and who it names. The two can move independently — search results can be stable while the assistants summarising them are not.
No. The reading is the field, not your brand, and it is direction-free: it counts movement without saying which way. What a high reading does mean is that a single day's change in your own numbers is weaker evidence than usual, because the answers around you were moving too.
Because a model rephrases itself every time it is asked. That is sampling temperature, not volatility, and counting it would put the line at its maximum every day. Only changes a reader would call a different answer are counted.
Once a day. The checks behind it run continuously, and the day's reading is computed when that day is complete. The page states how many comparisons produced each point.
The ones our checks run against, aggregated into one line. The number is deliberately not split per assistant here: a per-vendor breakdown invites reading one model's quirk as a trend, and the question this page answers is whether the field moved.
Because they were measured on too few comparisons to carry weight. A reading of 14 from thirty comparisons is a sample, not a spike, so it is published with its count and left out of the line rather than drawn as an event that happened.
The same questions in the words people use speaking to an assistant rather than typing into a box. Each carries the situation it came from, because the situation is what makes it a different question.
That is what this line answers. If the reading is high, the answers moved widely and your own drop is probably part of it. If it is low, the field was quiet and the change is more likely to be yours.
This page can tell you whether the answers moved on that day, and it cannot tell you why. A model update, a refreshed index and a news story all look identical from the outside. What it gives you is the date the movement happened, which is where to start looking.
No, and the difference matters. Those measure how far ranked results moved. This measures how far generated answers moved — which sources they lean on and who they name. Search can be perfectly stable while the assistants reading it are not.
Less than you think, and this page is how to tell. On a high-volatility day one reading of your own brand is weak evidence: the answers were moving around everybody. Wait for the next day before rewriting a page.
Whether an assistant names a brand when asked a question that brand should own.
Generative engine optimisation: the work of being described correctly by systems that answer instead of listing.
A measure stops measuring once it becomes a target — which is why this page publishes what the number cannot claim alongside the number.
