How answers are made
Exactly how Spotlit produces and reads AI answers, and what that means for how you use the numbers.
What you'll get from this page
- Know where every answer comes from.
- Know how mentions, positions and scores are worked out.
- Know the limits before you act on a number.
Where answers come from
Every answer is written by Spotlit's own AI model from a live web search, at the time of the scan. That is the same basic way AI assistants build answers: search first, then write from what they found.
One question, step by step
- 1
Search
We run a live web search for the question, leaning to your chosen country, and keep up to 10 results. If a search comes back empty, we retry once, then search without the country lean.
- 2
Write
A large language model (LLM) writes an answer from those results, in your chosen language.
- 3
Read
We find each tracked brand's name in the answer, and its place when the answer lists brands. We record the pages the answer cited.
- 4
Classify sources
Each cited site is sorted into a kind (review site, community, list, reference, other, yours or a rival's) and checked for your brand name.
A scan does this for each of your questions, plus the one about your brand. A few questions are asked several times in the same scan, and some in slightly different wording, to measure how much answers vary.
How the numbers are worked out
- Visibility rate: answers that named you, divided by all answers in the range.
- Share of voice: your mentions, divided by all brand mentions in those answers.
- Average position: your mean place in the answers that ranked you.
- Visibility score (one scan, 0–100): points per buying question, then scaled so small gains near zero still show.
| Your place in an answer | Points for the visibility score |
|---|---|
| Named in places 1–3 | 10 |
| Named in places 4–10 | 6 |
| Named at place 11 or lower | 3 |
| Named, but not in a ranked list | 2 |
| Not named | 0 |
The question about your brand by name is left out of the visibility score, because it is easy to win.
Country and language
Your market changes two things: the search leans to pages from your country, and the answer is written in your language. It is not a reading taken from inside that country, the way a person there would see it.
Limits to keep in mind
- Answers vary. The same question can name different brands on different runs. We don't show a precise “AI rank”; read rates and trends.
- Name matching is text matching. An unusual spelling of a brand can be missed, and two similar names can be confused.
- Fan-outs are predicted. The searches on AI searches come from our AI model, not from any AI app.
- A mention is not a visit. We measure mentions and citations, not clicks, traffic or leads.
- No proven levers. Studies have not shown that llms.txt or structured data alone get a brand named more often.
- Change is not cause. Results records what changed, not why.
See also
Was this page helpful?