Answers
What is silicon sampling?
Silicon sampling is the practice of asking a language model to stand in for human survey respondents. It is a research method, not a brand. Lewsearch is one implementation: a census-grounded synthetic panel with a published miss against poll toplines. Academic papers on silicon samples, including NIM's 2024 marketing-research note, did not evaluate Lewsearch.
Definition
A silicon sample is a set of answers produced by a model that has been prompted, or otherwise conditioned, to respond as if it were a person or a population. Researchers use it when they want a fast, cheap read before or beside a human fielding. The name comes from the methods literature, not from Lewsearch.
Grounding and elicitation
Two design choices dominate results. Grounding is who the model is asked to be: a one-line persona, a census microdata record, or a persistent agent with memory. Elicitation is how the question is asked: a single prompt, a verbalized multiple-choice draw, or a tabulated panel run. Lewsearch grounds on census-derived records and elicits a full panel. That is one choice among several. Changing either choice can move the number.
Prompt sensitivity and missing variance
Silicon samples are sensitive to wording, option order, and the profile you attach. They also tend to under-disperse: agents agree with each other more than a human sample would. If a vendor only shows a confident mean and no spread, ask whether the method recovered human variance or collapsed it.
Validation against human data
A silicon sample is only as useful as the human comparison. Lewsearch publishes mean absolute error against named public-poll toplines. That is a retrospective benchmark, not a warranty on your concept. See methodology and the evidence index. Census grounding is not the same as interviewing a human probability sample.
Buyer checklist
- Is the unit MAE, correlation, thematic overlap, or something else?
- How many questions were scored, and how many respondents per question?
- Was the test retrospective, held-out, or prospective?
- Can you download the scored file?
- What uses does the vendor refuse: legal sample, conversion forecast, WTP?
| Idea | Silicon sample in general | Lewsearch panel |
|---|---|---|
| What it is | Model answers standing in for people | Census-grounded agents, one instrument, tabulated shares |
| Typical use | Exploratory screen | Panel study, message test, focus group on top of the panel |
| Published miss | Varies by paper and prompt | About 10 pp as shipped on 1,521 scored questions |
FAQ
- Is silicon sampling the same as a synthetic panel?
- Related, not identical. Silicon sampling is the general method of using model-generated answers as stand-ins for people. A synthetic panel adds a fixed respondent file, a shared instrument, and tabulated option shares. Lewsearch sells the panel shape.
- Did NIM endorse Lewsearch?
- No. NIM published a methods note on using silicon samples in marketing research. That paper is a source for the category, not a test of this product.
- When is silicon sampling useful?
- Exploratory screens: instrument pretesting, message ranking, and concept shortlists before you spend on humans. It is not a probability sample and not a substitute for a legal, regulatory, or journalistic fielding.
- What should a buyer ask any vendor?
- What was measured, on how many questions, at what n, against which human source, and is the file public? If those cells are empty, treat the method as unvalidated for your use.
Check the work
Related answers
Canonical: https://lewsearch.com/answers/what-is-silicon-sampling