How scores work
The AI Visibility score
Each kind of check (for example Agent Search Visibility or AI Chat Visibility) gets its own score from 0 to 100, made of five parts with fixed weights:
| Part | Weight | What it measures |
|---|---|---|
| Frequency | 40% | How often the answers named you: answers that named you out of all readable answers. |
| Sentiment | 25% | How the answers that named you described you: positive 100, neutral or mixed 50, negative 0. |
| Position | 15% | Where you appeared in the answers that named you: first 100, second or third 85, fourth to tenth 60, lower 35. |
| Cross-engine | 10% | How many of the engines in this kind of check named you at least once. |
| Coverage | 10% | On how many of your questions you were named at least once. |
- Only answers S6S could read count. An answer that failed or couldn't be read is left out, never counted as “not named”.
- Questions that name your brand are left out: they test whether AI recognises you, not whether it recommends you.
- Being mentioned only as the thing others are compared against doesn't count as being named.
- Kinds of check are never blended into one number. The main card shows plain counts instead (“Named in X of Y answers”).
Measured, early read, or no score
A score is only as good as the answers behind it, so each kind of check has a bar for “measured”:
| Kind of check | Measured needs at least |
|---|---|
| Agent Search Visibility | 6 readable answers from 2 engines or apps across 3 questions |
| Google AI Overviews | 6 readable answers from 2 engines or apps across 3 questions |
| AI Chat Visibility | 4 readable answers from 2 engines or apps across 2 questions |
| Local & Maps | 3 readable answers, with your business found in them |
- Measured: the bar is met, and the score is shown.
- Early read: some readable answers, but fewer than the bar. Any number is shown muted and is directional only.
- No score: no readable answers yet, or none could be read. This is never shown as 0.
- A measured 0 is shown as 0: AI was asked enough and didn't name you.
These bars are calibration targets and can change as S6S collects more answers per check. A kind of check that wasn't refreshed by the latest check keeps its last result, with its date, and is held to today's bar.
Comparing checks over time
The Report compares the leading score with this brand's own normal swing, using only earlier checks that ran the same way. A change inside the normal swing is never called progress. How normal swing is read
The Progress tab's run history, the Dashboard and the brand list show each check's Agent Search Visibility score: the same number the Report shows for that check, scored the same way. Per engine they show how many answers named you, as the Report does.
The AI Readiness grade
AI Readiness grades the inputs you control in three parts: Website basics, Reputation and Reasons to pick you. How much each part counts depends on the kind of business:
| Kind of business | Website basics | Reputation | Reasons to pick you |
|---|---|---|---|
| Local business | 30% | 50% | 20% |
| SaaS / B2B | 35% | 35% | 30% |
| Dev / API | 50% | 25% | 25% |
| Industrial / B2B | 40% | 40% | 20% |
| Online shop | 35% | 30% | 35% |
| Publisher | 35% | 45% | 20% |
From parts to a letter
- The grade is the weighted average of the parts that have a score. A part with nothing measured yet drops out and the others are reweighted; it is never counted as 0.
- Letters: A from 85, B from 70, C from 55, D from 40, F below 40.
- Site access comes first: if your robots.txt blocks an AI search crawler (OAI-SearchBot, PerplexityBot or Googlebot) from your homepage, the grade is held at D or lower (at most 54) until it is fixed. The block itself adds no points, so it never raises a grade or creates one. Blocking AI training crawlers such as GPTBot costs nothing.
- A basic that works adds no points; a broken one costs them. Some rows are shown for reference and never graded.