What is CATI? Computer-assisted telephone interviewing

CATI stands for computer-assisted telephone interviewing: a survey method in which an interviewer asks the questions by telephone while software controls the questionnaire. The program shows each question in turn, applies the routing, checks answers as they are entered and stores the data, leaving the interviewer free to manage the conversation.

CATI is used in official statistics, public health surveillance, opinion polling and market research, wherever a spoken interview suits the population and the questions.

How a CATI survey works

To the respondent, a CATI interview is a phone call. Behind it, the software takes each sampled case through five stages.

1. Sample and call scheduling

A study begins with a sample: randomly generated telephone numbers, a customer list or a recruited panel. The software releases numbers for calling and links every attempt to its sampled case. As the Australian Bureau of Statistics (ABS) notes, it can book calls for times that suit respondents and reschedule engaged numbers. Interviewers for the US Behavioral Risk Factor Surveillance System (BRFSS), for example, call seven days a week, rotating attempts across days and times.

2. Introduction and screening

The interviewer introduces the study and confirms that the right person is on the line, who is not always the person who answers. In the BRFSS landline survey, interviewers select a random adult in the household; on cell phones, they interview the adult who answers. Screening questions then establish eligibility and whether the respondent’s quota cell is still open.

3. The programmed questionnaire

Each question appears on screen in order, with interviewer instructions kept separate from the words to be read aloud. The program applies the routing so that, in the ABS’s words, only relevant questions are visible to the interviewer. It can also rotate response options and carry earlier answers into later wording.

4. Data capture and checks

The interviewer keys each answer as it is given, and range and consistency checks can query an impossible value while the respondent is still on the line. For the BRFSS, the CDC builds skip patterns, consistency edits and range checks into the CATI program to reduce interviewer, data-entry and skip errors.

5. Dispositions and quality control

Every call attempt ends with a disposition: completed, refused, no answer, appointment, ineligible and so on. Final dispositions account for the issued sample and are the basis of response rates; AAPOR’s Standard Definitions sets out how to define them and calculate outcome rates. Supervisors monitor calls and review outcomes by interviewer while fieldwork is still running.

What CATI software does

CATI software, often called a CATI system, is the program behind those five stages. Its core functions are:

• Sample management. Importing records, handling duplicates and exclusions, and tracking each attempt against its case.
• Call scheduling. Appointments, callbacks, retry rules and calling windows by time zone.
• Questionnaire programming. Question text, response codes, routing, rotations and answers piped into later questions.
• Validation. Hard checks for impossible values; soft checks that ask the interviewer to confirm unusual ones.
• Quota control. Counting completes by cell and screening out respondents once a cell is full.
• Supervision. Call monitoring, and reports of dispositions and interview length by interviewer.
• Data output. A dataset and codebook, with the call history alongside.

A system applies a mistaken routing rule as consistently as a correct one, so the programmed instrument needs its own testing. Our CATI software overview shows how these functions work when an AI voice conducts the interview.

Advantages and disadvantages of CATI

CATI’s strengths come from control; its weaknesses come mostly from the telephone itself. The table draws on the ABS guide to data collection methods and the studies cited below.

In an early randomised comparison of CATI with a paper questionnaire, Groves and Mathiowetz found few differences in response distributions, some evidence of reduced interviewer variance and little difference in response rates. Interviewers favoured CATI for following the questionnaire and paper for changing earlier answers, so make corrections easy.

Coverage and response are the larger modern concerns. The CDC’s 2024 BRFSS overview cites preliminary data showing 78.0% of US adults were wireless-only in late 2024, and reports that sampling cell phones as well as landlines improved the representativeness of BRFSS data. Pew Research Center reported typical telephone response rates of 6% in 2018, while noting that a low rate raises the risk of error without necessarily making a poll wrong.

CATI vs CAWI vs CAPI

CATI is one of three computer-assisted modes. CAWI, computer-assisted web interviewing, is completed by the respondent online. CAPI, computer-assisted personal interviewing, puts an interviewer face to face with the respondent, with the questionnaire on a laptop or tablet.

Mode affects how people answer as well as who is reached, and switching mode can break a tracker. Our comparison of CATI, CAWI and CAPI sets out the evidence.

Where CATI is used

CATI is older than many researchers assume. In their chapter on mode of data collection, Tourangeau, Rips and Rasinski date the first CATI survey to 1971, when Chilton Research Services conducted one for AT&T. By 2000 they described CATI as the most commonly used method for collecting data by telephone in the United States and Europe.

Official statistics

The ABS uses telephone interviewing in both household and business surveys. Its Labour Force Survey samples about 24,000 dwellings and interviews each household monthly for eight months. Generally the first interview is face to face and later ones are by phone, with an online option.

Health surveillance

The BRFSS completes more than 400,000 adult interviews a year, which the CDC says makes it the largest continuously conducted telephone-based health survey system in the world. In 2024, 54 states and territories used CATI systems for it, and the core questionnaire took 17 minutes on average.

Polling and social research

Pollsters and social researchers use CATI when a telephone sample suits the population. Election polls show the trade-offs in public, as we discuss for Australia and New Zealand.

Market research

In market research, CATI suits studies built from a customer or business list, and business-to-business work where an interviewer has to get past a switchboard to the right respondent.

Designing a CATI study that holds up

The software executes a design; it does not supply one. Five practices do most of the work.

Specify the routing, then test it with cases

Programming turns a questionnaire into decisions about what is read aloud, what only the interviewer sees, which answers are accepted and what happens next. Specify response codes, reference periods, rotations, eligibility rules and the handling of uncertainty.

Then test with deliberately chosen cases: an ineligible respondent, a value at the edge of a range, someone who corrects an earlier answer, and someone who declines a question that controls later routing. Compare each exported record with the expected result. A question can look right on screen while writing the wrong value.

Keep “don’t know”, “refused” and “not asked” distinct

Suppose a survey asks how many appointments a respondent attended last month, then asks about the most recent one. Zero should skip the follow-up. “Don’t know” may not, because the respondent could still describe the latest appointment, and a refusal is a third state. Coding all three as zero produces clean-looking data by discarding real distinctions.

Decide in advance which conditions justify a hard check, such as an impossible date, and which only a confirmation prompt, such as a large but possible number. Treating every unusual answer as an error can push interviewers towards more conventional answers.

Write clarification rules with the questions

The same words do not guarantee the same understanding. In laboratory experiments beginning with Schober and Conrad (1997), respondents answered from fictional scenarios, such as a receipt for a floor lamp when asked about furniture purchases. Where a scenario fitted the survey concept straightforwardly, accuracy was nearly perfect either way.

Where it did not, clarification decided the result. In the authors’ later summary, respondents answered these atypical cases correctly 28% of the time under strict standardisation and 87% when interviewers could explain concepts freely, though conversational interviews took more than three times as long.

For CATI, agree approved definitions, neutral probes and a boundary on what the interviewer may explain before fieldwork. A respondent asking whether a student living away from home counts as a household member needs the study’s rule, not an improvised one. Cognitive testing finds these ambiguities early, and questions should be written for the ear.

Use dispositions and monitoring for quality control

The BRFSS shows CATI inside a wider quality process. The CDC’s overview describes BRFSS-specific interviewer training, the capability to monitor interviewers at every site and, in some states, verification callbacks. Disposition codes tabulated by interviewer were the primary means of quantifying interviewer performance.

Patterns need interpretation before judgement. Short interviews may signal a routing fault, rushing or a subgroup with fewer questions; one interviewer’s higher refusal rate may reflect the shifts assigned. Record every correction and the cases it affected.

Protect comparability across waves

Reusing a questionnaire does not by itself keep a tracker comparable; the frame, selection, contact protocol, weighting and mode can all change. In 2011 the BRFSS added cell phone interviews and replaced post-stratification with raking, and the CDC advises that its 2024 data are not directly comparable with years before 2011. Telephone interviewing continued; comparability did not. Before changing an established procedure, run old and new protocols in parallel on comparable samples.

A pre-fieldwork checklist

• Frame and selection. The sample source, its coverage gaps and any within-household selection rule are documented.
• Routing. Every skip, rotation and eligibility rule is specified and tested against expected output.
• Missing data. Don’t know, refused and not asked have separate codes.
• Clarification. Approved definitions and neutral probes sit with the questionnaire.
• Dispositions. Outcome codes map to AAPOR final dispositions.
• Methods record. Changes to frame, protocol, weighting or mode are logged wave by wave.

What is AI CATI?

AI CATI is CATI in which the interviewer is an AI voice. Software still controls the questionnaire, routing and data capture, and the respondent still answers aloud. Unlike an IVR survey, which plays recordings and takes keypad presses, an AI interviewer can repeat an item, offer an approved definition or ask a scripted follow-up.

The design goal is an interviewer’s responsiveness without a human listener or the limit of interviewer hours. That is a goal to test, not a demonstrated result. Check recognition and coding against recordings, adherence to wording and clarification rules, breakoffs and sensitive items, and compare with human CATI in a split sample before moving a tracker; evaluating AI voice interviewers sets out how.

catilyst is built for that kind of study: AI voice interviewers calling in parallel, following the researcher’s questionnaire and routing with approved probes and scripted clarification. Quotas are screened during the interview, calling windows and retry rules are set per study, and completions appear in a live view, with structured responses and CSV export. See how it works.