How KI Score queries ChatGPT, Perplexity, and others

This article explains how KI Score sends your prompts to the AI providers. Afterward, you’ll know why some providers are approximations, what role the measurement country plays, and what KI Score adds to your question.

The eight AI providers

Monitoring lets you choose from eight AI providers. The dashboard labels them as follows:

• “ChatGPT”
• “Google AI Overview”
• “Perplexity”
• “Gemini”
• “Copilot”
• “Grok”
• “Claude”
• “DeepSeek”

You choose which providers your runs query in Zeitplan (Schedule) under KI-Anbieter (AI providers). New websites start with ChatGPT and Google AI Overview selected. How many providers you can select at once depends on your plan: Essential and Advanced allow 3, Premium allows 5, and Enterprise allows all eight. Each additional provider costs one more credit per run, active prompt, and measurement country.

API queries with web search

KI Score doesn’t open chat apps or read browser interfaces. Every provider is queried through its models via API, routed through the OpenRouter service, and always with web search turned on. That way the AI can draw on current web pages and cite sources, much like it would in a search situation. Perplexity uses its own built-in web search.

Web search is the reason you can see in Responses and Sources which domains and URLs an answer cites. For ChatGPT, Copilot, Grok, Claude, and DeepSeek, the measurement also returns the searches the model runs in the background. You’ll find them under Fanouts and in the answer details; see Fanouts: the searches behind AI answers.

The KI Score measurement queries providers the same way. Which providers it uses is set centrally by KI Score.

What “approximation” means

Three providers are approximations of a product interface:

• ChatGPT is an API measurement with web search that approximates the ChatGPT interface. The ChatGPT app itself is not queried.
• Google AI Overview uses a model to approximate the AI overviews in Google Search. KI Score does not read real AI Overviews from Google Search.
• Copilot is also a model approximation, not the Copilot app.

The names in the dashboard don’t carry a separate label for this. The dashboard glossary explains it under Näherung (approximation). Answers from these providers can differ from what a person sees in the respective app or search engine. In return, KI Score sends every query using the same defined procedure, which makes results easier to compare with one another.

In Responses, the Modell (Model) column shows which provider delivered an answer. Hover over the entry to see the exact model identifier KI Score stored for the answer. A competitor’s comparison panel under Competitors also shows it in the KI-Anbieter und Modelle (AI providers and models) section. Because the models behind the providers can change, this article doesn’t name any versions.

Your prompts’ measurement country

Every monitoring prompt has at least one Messland (Country), and at most 10. KI Score passes the country along with the query as geographic context, so the answer is measured from that country’s point of view. If a prompt has several measurement countries, KI Score measures each one separately. So in every run, each active prompt is queried exactly once per measurement country and selected AI provider, and each measurement country costs credits like a separate prompt.

When you add prompts, the country you chose during onboarding is preselected as the measurement country. In the Prompts hinzufügen (Add prompts) panel, you can change it in the Messland (Country) field, and in a CSV import with the Messland (Country) column. For existing prompts, change the measurement countries in the prompt editor or by clicking the flags in the Messland (Country) column of the prompts table, and for several selected prompts at once with Messländer bearbeiten (Edit countries). A change applies from the next run; answers already measured keep their original context. In Responses, the Land (Country) column shows which measurement country an answer comes from.

A measurement country always covers a whole country. If a question should refer to a city or region, write the place into the prompt, for example “Which tax advisors in Cologne are worth recommending?” A place name in the prompt stays part of its text. How strongly an AI provider adapts its answer to the measurement country can differ from provider to provider.

The KI Score works independently of this: the market measurement uses the market’s measurement area, not your prompts’ measurement countries. To learn more, see Which location is measured?.

What gets added to your prompt

KI Score doesn’t send your prompt unchanged. So that rankings can be evaluated, KI Score appends this instruction, in German, to every prompt:

Antworte bitte in Form eines Rankings der besten Ergebnisse. Nenne die Namen klar nacheinander (1., 2., 3. ...), damit ich die Platzierungen auswerten kann. Vermeide langen Fließtext.

In English, it asks the AI to answer as a ranking of the best results, to list the names clearly one after another (1., 2., 3. …) so the placements can be evaluated, and to avoid long running text. From the order of the brands named, KI Score determines the position at which your brand and your competitors appear. As a result, answers are usually more list-like than a typical chat conversation.

Tip: You don’t need to ask for a list or ranking in your prompts yourself. Phrase the question the way your customers would ask it.