Prompt Research

Make the actual usage purposes of LLMs visible. Prompt Research reveals prompt patterns through scientific reproduction and semantic reconstruction: cross-model consistent, privacy-compliant, and exclusively licensed through Rankscale. We estimate prompt volume honestly, not with an unverifiable panel count.

Why trust Prompt Decoding?

Exclusively licensed through Rankscale, Prompt Decoding is the only method that has been scientifically reproduced, cross-model validated, and recognized by the professional community.

Scientific reproducibility

Results independently validated in the OpenAI/Harvard study (NBER Working Paper 34255). Usage purposes and clusters were partially word-identical to Prompt Decoding findings published months earlier.

Cross-model consistency

Applied to both ChatGPT and Gemini, producing very similar clusters across models. Your strategy is not tied to a single vendor.

Community recognition

Presented at the G50 Summit 2025 by Hanns Kronenberg and won 3rd place at the SEO World Championship. Validated by the global SEO community.

Methodological basis

Relies on internal model simulations - revealing realistic prompt data from millions of real queries. Reproducible, privacy-compliant, and without personal data.

On which data is Prompt Decoding based?

Prompt Search Volume uses semantic reconstruction - estimating how often a given intent exists within the model's learned distribution. Think of it as the LLM equivalent of search volume.

• No tracking of real users
• Model-internal probability, not telemetry
• Based on semantic prompt density, not keyword frequency
• Results reproduced in the OpenAI/Harvard study on GPT usage

KEY CAPABILITIES

01 Honest prompt volume estimates

We estimate prompt volume through semantic reconstruction, not a black-box panel number nobody can verify. There is no public equivalent of Google keyword volume for ChatGPT prompts.

02 Cross-Model Consistency

Rankscale applies Prompt Decoding across ChatGPT and Gemini. Both models produce very similar clusters with this method and describe the same markets-so you get consistent, comparable insight across the engines that matter.

03 Methodological Basis

Rankscale uses Prompt Decoding based on internal model simulations in ChatGPT and Gemini. Typical questions, frames, and answer paths become visible: reproducible, privacy-compliant, and without personal data.

Why Use Prompt Research?

Rankscale delivers Prompt Research exclusively: scientifically reproduced, cross-model consistent, methodologically sound, and recognized by the professional community. Make LLM usage purposes visible without tracking real users.

Intents

Giving back strongest intents and clusters for a segment - understand what users are really asking AI engines about your market.

Prompts

Core prompts with trends - see which questions drive AI conversations in your space and how they evolve over time.

Entity

Entity Impression Share - measure how often your brand entity appears relative to competitors across AI engine responses.

FAQ

01 What is Prompt Decoding?

Rankscale licenses Prompt Decoding, a method developed by Hanns Kronenberg to make the actual usage purposes of large language models visible. It is based on millions of real prompts and relies on internal model simulations in ChatGPT and Gemini-reproducible, privacy-compliant, and without personal data.

02 Is Prompt Decoding scientifically validated?

Rankscale's Prompt Decoding was validated when the OpenAI/Harvard study 'Who People Use ChatGPT' (NBER Working Paper 34255, September 2025) identified central usage purposes and clusters-partially word-identical and with comparable frequencies-that had already been published in April 2025 through Prompt Decoding. The method's quality and validity are clearly underscored by this independent reproduction.

03 On what data is Prompt Search Volume based?

Rankscale's Prompt Search Volume is not based on tracking real users. It estimates how often a given intent or question exists within the model's learned distribution of prompts through semantic reconstruction, reflecting how frequently similar questions were present during training and fine-tuning (RLHF). No real-user tracking; model-internal probability and semantic prompt density.

04 How accurate is prompt volume data in this category?

There is no public source that exposes how many people asked a given prompt to ChatGPT the way Google exposes keyword search volume. Tools that claim exact real-world prompt counts from a panel should be asked to show their source. Rankscale estimates prompt volume through semantic reconstruction instead: an honest, methodologically grounded estimate rather than a black-box panel number nobody can verify.

05 Is there a free trial?

Yes. You can try the Pro plan free for 7 days with a limited credit allowance, so you can explore Prompt Decoding and prompt research for your use cases end-to-end before committing. The trial converts to a paid Pro subscription only if you choose to continue, and you can cancel anytime from your account settings.