Research Methodology

Why this methodology exists

AI answer engines are dynamic systems. Results can change by model version, location, prompt wording, user context, retrieval settings and time. Cited uses a documented methodology so findings can be interpreted as structured observations rather than universal or permanent facts.

1. Define the research question

Each study begins with a clear question, scope and intended use. We identify the platforms, markets, categories, locations and time period being tested before collecting observations.

2. Build a controlled prompt set

Prompts are designed to reflect genuine discovery, comparison, recommendation and trust questions. Exact prompt wording is recorded. Where a study compares platforms, the same core prompt is used across each platform unless a difference is explicitly disclosed.

• Discovery prompts test whether a brand appears for broad category questions.
• Recommendation prompts test whether a brand is actively presented as a suitable option.
• Comparison prompts test how brands are framed against alternatives.
• Trust prompts test which sources, proof points and third-party references shape the answer.

3. Record test conditions

For each observation, we record the platform, date, prompt, location assumption and relevant session conditions. Where practical, important findings are checked across fresh sessions or repeat runs because a single AI answer may not be representative.

4. Classify the outcome

Cited separates different forms of visibility rather than treating every appearance as equal:

• Mention: the brand appears anywhere in the answer.
• Recommendation: the brand is presented as a relevant or suitable option.
• Citation: the answer links to or identifies a source associated with the claim.
• Accuracy: the description of the brand is materially correct and current.

5. Verify evidence

Where a finding depends on external evidence, we prefer original research, official documentation, public datasets and direct observations. Secondary sources may provide context, but they do not replace a primary source when one is available.

6. Report limitations

Every material study should explain what was tested and what was not. A sample of prompts cannot represent every user, and a result observed on one date may change. We do not present small samples as universal market truth or guarantee that every user will see the same answer.

Updates, corrections and repeatability

Time-sensitive research includes a publication or review date. Material corrections are made when identified. Where a study is repeated, Cited aims to preserve the original prompt set and measurement definitions so changes can be compared more meaningfully over time.