Does every survey use a sample?

No; a collection may cover the entire defined population.

Is an online poll automatically representative?

No; selection and participation matter.

Can AI help with survey work?

It can assist with drafting or coding, but measurement quality, privacy practices, and analytical validity still require review.

AI Glossary / Statistics and probability / Intermediate

Survey

Definition

A survey is a planned process for collecting information about a population through observations or responses from its units. It can collect information from all units or from a selected sample, depending on its design. A useful survey connects a clear question, defined population, suitable measurements, and an appropriate method of analysis.

What it means

A survey is more than a questionnaire. The questionnaire is one possible collection instrument; the survey also includes deciding whom to cover, how to reach them, how to handle missing responses, and what conclusions the results can support.
Surveys can describe people, households, businesses, transactions, or other defined units. Their quality depends on these design choices, not simply on the number of completed forms. An online poll may collect many responses without representing the population that a business wants to understand.

How it works

Begin with the decision or research question. Specify which quantities will be estimated, such as a proportion using a service or an average satisfaction score. Define the reference period so respondents answer about comparable events.
Define the target population and observation unit. Distinguish a person from a household, account, or transaction. Decide whether the respondent reports for themselves or on behalf of another unit, because this affects what they can reasonably know.
Construct or identify a sampling frame. Check coverage, duplicates, and eligibility. A complete-looking database can still omit relevant groups. If surveying a sample, specify its selection method and retain the information needed to calculate weights and uncertainty.
Choose a collection mode such as web, telephone, paper, or interviewer administration. Accessibility, cost, confidentiality, and likely response patterns influence the choice. Mixing modes may improve reach while introducing measurement differences that need evaluation.
Develop clear, neutral questions with usable response options. Avoid combining two issues into one item or presuming an experience that some respondents did not have. Pilot the instrument and check comprehension before full collection.
Implement contact, follow-up, and data-handling procedures consistently. Track dispositions and missingness. A refusal, an ineligible unit, and an unanswered item are different events and should not be merged indiscriminately.
Validate and analyze the collected data using the design. Apply justified weights, account for clustering or stratification where relevant, and state assumptions behind adjustments for nonresponse. Weighting cannot automatically restore information about groups absent from the frame.
Report the population, dates, mode, question wording, sample selection, achieved responses, and uncertainty. Separate sampling variability from other sources of error. Limit conclusions to what the design and measurements support, especially when moving from descriptions to causal claims.

Three practical examples

Customer research: estimate how often eligible customers use an AI support feature.
Workplace planning: collect employee experiences with a new workflow using clearly defined confidentiality and coverage procedures.
Product evaluation: obtain structured feedback from sampled users alongside observed performance measures.

Think of it this way

Think of a survey as an organized listening process. The questions matter, but so does who gets invited, who answers, and how their answers are interpreted. A loud group of volunteers does not necessarily speak for everyone the organization hopes to serve.

Worked example

Consider an original hypothetical survey of 2,400 eligible customers. Select 240 distinct customers by simple random sampling without replacement, and suppose all selected customers respond. Of the 240, 84 report using an AI feature during the stated month.
The sample proportion is 84/240 = 0.35, or 35%. Each selected customer has inclusion probability 240/2,400 = 0.10, giving a design weight of 10. The estimated number of users is 84 times 10 = 840, also equal to 2,400 times 0.35.
These are estimates of the defined finite customer population under the stated design. They are not exact counts of all users. A suitable uncertainty calculation would account for the sampling design and the fact that sampling occurred without replacement.
If only volunteers had answered an open link, the same 84 out of 240 arithmetic would describe the respondents, but the simple-random-sampling interpretation would not follow. Likewise, partial response would require examining nonresponse rather than silently retaining the complete-response assumptions.

Illustrative case study

Fictional case: Alder Digital wants to know whether customers find an AI assistant useful. Its first idea is an optional pop-up shown only after successful interactions. Analysts recognize that this would miss many dissatisfied users. They instead define the eligible customer population, select a probability sample, and include a clear question about recent use before asking about usefulness. The final report distinguishes feature users from all customers and explains the response pattern, preventing an attractive percentage from answering the wrong question.

Common misconceptions and mistakes

Do not use survey and questionnaire as perfect synonyms. A large response count does not guarantee representative results. Satisfaction among respondents is not automatically satisfaction among all customers. A high response rate does not prove freedom from measurement error, while a low rate requires careful investigation rather than an automatic numerical correction. Survey associations alone do not establish that a product caused an outcome.

Business use

Use surveys when important experiences, intentions, or characteristics are not adequately captured in operational data. Combine them with other evidence when possible. Assign ownership for population definitions, question design, collection quality, and analysis so the resulting figures remain interpretable as they move into dashboards and business decisions.

Check your understanding

What does the 35% estimate refer to?

Answer: Eligible customers' reported use during the defined month under the stated sampling assumptions.

Why is the weight 10?

Answer: Each sampled customer represents the inverse of a 0.10 selection probability.

Would an open volunteer link justify the same design-based inference?

Answer: No.

Related terms

Census
Nonresponse
Measurement error
Questionnaire
target population
sampling frame
probability sampling
survey weighting

Learn next

Study questionnaire testing and sampling design, then examine response dispositions, weighting, and design-based uncertainty. Practice writing a survey methods note that lets another analyst understand exactly what a reported percentage represents.

Sampling design

Sources

Australian Bureau of Statistics, Census and sample: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/census-and-sample Australian Bureau of Statistics, Types of error: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/types-error Statistics Canada, Survey Methods and Practices: https://www150.statcan.gc.ca/n1/en/pub/12-587-x/12-587-x2003001-eng.pdf The customer example is original and assumes complete response.

Previous: Surveillance capitalism Next: Survey errors