Guide 3 of 6

Understand your first run

A run finished and the screen is full of numbers. Here is what each one is built from, and which of them is not true yet.

Read time 6 min · Product behavior verified on 2 Sep 2026

What a run actually is

One question put to one engine gives one answer. That answer is stored, analysed, and becomes one row. A run is all of those pairs, played once.

i
40 prompts on 3 engines is 120 answers, not 40. This is the number that explains how long a run takes, how your usage adds up, and why turning an engine off on a prompt is a real saving.

Engines are set per prompt, so the arithmetic is not uniform. A question that only matters inside Google can run on two engines while the rest of your set runs on three.

What "analysed" means

The raw answer is not the product. Every answer is read and five things are extracted from it.

ExtractedMeaning
MentionedWhether your brand is named in the answer at all. This is the yes or no behind your visibility score.
PositionWhere you come in the answer when several brands are named. First recommendation, or the fifth one down.
SentimentHow you are talked about, when the answer says enough about you to tell.
Other brandsEveryone else named, in their order. This is what your share of voice is measured against.
Cited sourcesThe pages the answer points to. These are the pages the engine is reading about your market.

Every one of these stays attached to the answer it came from, so any number on your dashboard can be traced back to the exact text that produced it. If a figure surprises you, open the answer.

One answer opened: engine tabs, rank history, the four per-answer metrics, the full captured text and the brands it named
Figure 1One prompt, opened on one engine. Every number on this panel is derived from the text underneath it, which is shown in full with your brand highlighted. Switch engine with the tabs at the top: they often disagree, and that disagreement is the useful part.

When a pair fails

An engine can time out, refuse, or return something unusable. That pair is retried twice, a minute apart then five minutes apart. If it still fails it is marked as failed and left alone.

A failed pair is not a hole in your dashboard. Your current state is read per pair, from the last time that pair came back successfully across your recent runs, so one bad Monday usually leaves the numbers untouched. The next run creates the pair again from scratch. A pair only disappears from the picture if it has failed on every recent run, which is a real signal rather than a glitch.

Baseline against trend

This is the most common misreading of a first run, and it is worth being precise about.

Every arrow on your dashboard compares one prompt on one engine with the previous time that exact same pair was measured. Not this run against the last run as a whole, and not your brand against an average. Pair against its own past.

AfterYou can readYou cannot read
One runWhere you stand: which prompts name you, who wins the ones that do not, what gets citedAny direction at all. There is nothing to compare against.
Two runsYour first real movement, prompt by promptA pattern. Two points is a line, not a trend.
Four or moreWhether what you shipped worked, and what is drifting on its ownNothing much. This is where the product starts paying.
!
A new prompt starts its own history. Add a question today and it has no previous measurement, so it shows no movement until next week, even though everything around it does.

Why the numbers moved after you subscribed

If you saw results before entering a card, you saw a preview, and a preview is deliberately smaller.

PreviewYour first full run
PromptsUp to 30Every active prompt you kept
EnginesThe 3 core onesEvery engine you enabled
RepeatsNever. It runs once.Every Monday

So a visibility score that shifts between the two is expected: it is a different denominator, not a change in the market. Treat your first full run as the real baseline and ignore the preview from then on.

Reading the Prompts table

This is the screen you will live in, so three columns are worth learning properly.

ColumnHow to read it
EnginesOne icon per engine that ran this question. A highlighted icon means that engine named you. Six greyed icons on a purchase question is your most actionable row.
RankYour best position across the engines that cite you, not an average. One engine putting you first is a real fact and we do not dilute it.
A dash instead of a rankNo engine gave you a place on that question. Either none of them named you, or they mentioned you in passing without ranking anything. The Absent filter isolates the first case, which is the shortest list of things worth fixing.

Sort by rank and read from the bottom. The worst rows on your buying questions are the shortest path to a decision, and they are what the action plan is built from.