Fanouts: the searches behind AI answers

While an AI provider answers one of your monitoring prompts, it often sends its own queries to its web search. KI Score calls these background searches fanouts. This article shows where to find them, how to read the key figures and cards, and what the data can’t tell you.

What fanouts are

A fanout is the search text a model sends to its web search while producing a monitoring answer and reports back in KI Score’s measurement. Identical search texts count as one fanout, regardless of upper or lower case. Fanouts show which terms the AI searches for before it answers, which tells you what searches your content should be found for.

Fanouts describe KI Score’s API measurement. The providers’ own apps, such as ChatGPT, may search differently.

Where to find fanouts

In the sidebar, open Fanouts in the Sichtbarkeit (Visibility) group. The page shows the fanouts for the selected website, for the last 30 days by default.

You can also see the fanouts of a single answer under Responses. The answer details of a completed answer include a Fanouts (n) section listing every search in the order the model ran it, repeated searches included.

Topics, prompts, or list

Three tabs in the page header set how the Alle Fanouts (All fanouts) table is grouped:

• Themen (Topics): fanouts by the topics of your prompts
• Prompts: fanouts per prompt
• Liste (List): all fanouts without grouping

Groups are ordered by occurrences. For each fanout, the table shows the columns Fanout (the search text), Modelle (Models, the providers that ran this search), Vorkommen (Occurrences), and Antworten (Answers). You can sort by Fanout and Vorkommen.

The key figures

The strip above the table shows three values for the selected period:

• Verschiedene Fanouts (Distinct fanouts): how many different search texts the providers sent
• Vorkommen (Occurrences): all background searches in the completed answers, each search counted once
• Antworten mit Fanouts (Answers with fanouts): how many completed answers include at least one reported search, out of all answers in the period

The Antworten column in the table counts the different answers in which a fanout appeared.

Brands and word pairs

Two cards sit next to the table:

• Marken in Fanouts (Brands in fanouts): your brand and your tracked competitors whose name or alias appears in the search texts, with the number of occurrences. Discovered brands you don’t track don’t count here.
• Häufige Wortfolgen (Frequent word pairs): the most frequent word pairs across all search texts in the period, without filler words

Each card shows at most ten entries.

Filtering, searching, and exporting

• Period: In the filter bar, choose a period such as Letzte 7 Tage (Last 7 days), Letzte 30 Tage (Last 30 days), or Gesamter Zeitraum (Entire period).
• Filters: Use + Filter to narrow down by Thema (Topic), Tags, Nennt Marke (Mentions brand), Modell (Model), and Prompt. Zurücksetzen (Reset) removes all filters.
• Search: The Fanout suchen… (Search fanouts) field narrows down the table only. The key figures, brands, and word pairs still cover the whole period with the filters you set.
• Export: The download icon in the page header (screen reader name “Alle gefilterten Fanouts exportieren”, export all filtered fanouts) saves the file fanouts-YYYY-MM-DD.csv. It contains every matching fanout under the current filters and search, regardless of the table page you’re on, with the columns Fanout, Vorkommen, Antworten, Modelle, and Zuletzt (last seen), plus Thema or Prompt when the table is grouped.

Which providers deliver search texts

In the current measurement setup, ChatGPT, Copilot, Grok, Claude, and DeepSeek report their search texts. Gemini, Google AI Overview, and Perplexity don’t deliver search texts, so their answers don’t count toward Antworten mit Fanouts. That doesn’t mean these providers didn’t search.

You choose which providers your monitoring queries in the schedule; see Choosing AI providers for your monitoring.

Why the list can be empty

• No run yet: Fanouts appear only after your first monitoring run.
• No answers in the period or for your filters: Choose a different period or reset the filters.
• No reported searches: The answers come only from providers that don’t deliver search texts, or the models didn’t report any searches for these answers.
• Older answers: Answers from before fanouts were introduced in October 2026 contain no search texts. KI Score doesn’t query them again after the fact or add estimated searches.
• Search or filters: None of the fanouts match your search text or filters.

What fanouts are not

• Not search volume: Vorkommen counts how often the providers ran this search in your measurements, not how often people search for it.
• No extra cost: Fanouts come with your regular monitoring runs. The page doesn’t use any credits.

To learn how KI Score queries the providers, see How KI Score queries ChatGPT, Perplexity, and others.