Which LLMs we track and why

Documentation Index

Fetch the complete documentation index at: https://docs.getcited.in/llms.txt

Use this file to discover all available pages before exploring further.

Cited tracks brand visibility across five AI platforms: ChatGPT, Perplexity, Gemini, Claude, and Grok. Each platform is queried using a specific model via its API. This page documents which models are used, what each costs, and why these five were selected.

​

The five platforms

Each platform is tracked using a specific model and API configuration, chosen to balance cost, accuracy, and response quality. As of April 2026: Grok’s “Platform only” status means it is tracked for individual brand dashboards but is not yet included in the Cited Index pipeline. The Index currently runs 4 platforms. Grok integration into the Index is planned for a future edition. Perplexity uses two models: the Cited Index runs sonar (the cheaper model, ~83% cost reduction) for monthly benchmarks, while daily brand dashboards use sonar-pro for richer response quality.

​

Why these five

Three criteria determine whether a platform is included in Cited’s tracking: User adoption. The platform must have meaningful consumer adoption for product and category queries. ChatGPT, Perplexity, Gemini, Claude, and Grok each have distinct user bases asking the kinds of questions where brand visibility matters — “best CRM for startups,” “skincare brands for oily skin,” “travel bags under 5000.” API availability. The platform must offer a programmatic API that supports the query patterns Cited needs — full text responses to natural-language prompts, with parseable output for mention extraction. All five platforms meet this requirement. Cost viability. Per-query cost must be sustainable at scale. At current pricing, the five tracked platforms range from $0.002/query (Gemini) to$0.034/query (Grok). Grok’s higher cost is a factor in why it is not yet included in the monthly Cited Index refresh.

​

Platforms not tracked (and why)

Google AI Overviews (AIO): Tested via SerpAPI in the India market. Result: approximately 3% of queries triggered an AI Overview — too sparse for meaningful benchmarking. AIO data from earlier testing is preserved. Tracking may resume as Google expands AIO coverage. Microsoft Copilot / Bing Chat: Considered but deprioritized. Copilot’s search behavior overlaps substantially with ChatGPT (both use Bing-derived retrieval) and its standalone adoption in the Indian market is lower than the five tracked platforms.

​

Model selection rationale

Cited does not always use the most expensive or most capable model for each platform. The goal is to match the model tier that typical consumers encounter, not the most capable tier available. For Perplexity, the Cited Index uses sonar rather than sonar-pro because the 83% cost reduction makes monthly refreshes across 253 brands economically viable, and testing shows comparable mention rate accuracy between the two models. For Gemini, gemini-3-flash-preview is chosen for its cost-efficiency. The 29-second average response time is notably slower than other platforms, which the team is investigating. Model versions change as providers update their offerings. When a model is deprecated or replaced, Cited validates the replacement against historical data before switching. Model changes are documented in the platform’s operational changelog.

​

Related concepts

• How ChatGPT search works — the most widely used platform
• How Perplexity ranks sources — the most citation-focused platform
• Cited Index benchmarks — where the per-platform data comes from
• Refresh cadence — when and how often queries run

​

Frequently asked questions