By the Wavefield Research team · Published Sep 5, 2026
Ask for year of birth when analysis needs precision, and a single-choice age band when a screener or quota needs to act on the answer. Never let bands overlap, align them to the population targets you will weight to, always offer Prefer not to say — and ask early only when age gates who gets in.
That is the whole method in one paragraph. The rest of this guide is the reasoning, the band sets that keep weighting possible, the placement rule the usual advice gets wrong, and practical wording for the demographic survey questions that follow age onto almost every questionnaire.
Every way to ask age in a survey is a trade between precision, comfort, and what the answer must do mid-survey. The principle that settles most arguments: collect at the finest grain the respondent will tolerate, because fine data can always be banded later and banded data can never be un-banded.
Swipe the table sideways to see all columns →
| Format | Wording | When it wins | What it costs |
|---|---|---|---|
| Year of birth | “In what year were you born?” (four digits, open numeric) | The precision default. Ages never go stale in a tracker, exact ages compute forever, and any band set can be derived later. | Feels slightly more personal than a band; a few respondents skip it. Require all four digits, or 74 and 1974 will coexist in your data. |
| Exact age | “How old are you?” (open numeric) | Simple and universally understood. Fine for one-off studies where the fieldwork window is short. | Ages in a long tracker drift stale, and typos produce 9-year-olds and 199-year-olds — validate the range. |
| Age bands | “Which age group are you in?” (single choice) | The screener-and-quota workhorse: bands feed quota cells directly, feel less intrusive, and answer in one tap. | Irreversible. Bands can never be un-merged, re-cut, or averaged. Choose them to match your weighting targets or regret it at analysis. |
Two mechanics regardless of format: skip dropdown menus (a 60-item scroll is hostile on every device), and validate numeric entries so a mistyped 9 or 199 cannot enter the data quietly.
The mechanical rules first, because breaking them is embarrassingly common: bands must not overlap (a scale offering 18–24 and 24–30 gives a 24-year-old two homes), must be exhaustive from your minimum age up through an open top band (65+), and should read in one consistent direction. Then the rule that actually separates professional questionnaires from templates: choose bands that map onto the population targets you will weight to. The census-aligned set — 18–24, 25–34, 35–44, 45–54, 55–64, 65+ — exists because published population counts come in those breaks. Ask in 20–29/30–39 instead and raking to census targets becomes impossible without assumptions your report will have to confess.
Generation labels deserve their own caution. Gen Z, millennial, and boomer bands make lively reporting, but they age (a “millennial” band drawn in 2020 is wrong by 2026), their boundaries are contested, and Pew Research Center itself has largely stopped using them outside genuinely generational analysis. The working compromise: collect birth year or census bands, and translate to generation labels at reporting time if the audience expects them. Collection is forever; labels are a formatting choice.
Standard advice says demographics go at the end of the questionnaire, where personal questions cost the least trust and an abandonment loses the least data. The advice is right — and it quietly breaks for age, because age is rarely just a profiling variable. In most commercial studies it is a screener (under-18s must terminate), a quota dimension (age cells filling toward targets), or both. A quota cannot act on an answer given after the respondent finishes, so when age does mid-survey work, it must be asked early, as a banded single choice the quota engine can watch in real time.
The resolution is one question, placed by its job: age goes in the screener when it gates or fills quotas, and in the end-of-survey demographics block when it merely profiles. What it should never do is appear twice — respondents notice, and the two answers will disagree just often enough to embarrass the dataset. On Wavefield the agent programs this from the brief: a terminate on the under-age band, quota cells on the rest, and the answer flows into crosstabs and weighting without a second ask.
A note on minors, since the youngest band is usually a terminate: most commercial research stops at 18, both because panels rarely admit minors and because children's-privacy law (COPPA in the US, for under-13s) makes collecting their data a specialist exercise with parental consent requirements. Terminate mechanics matter more than wording here — a screen-out should read like a normal survey ending, not an accusation. Our runner deliberately shows the same Next button and closing page whether a respondent completed or screened out, so the seventeen-year-old and the panel professional trying to guess the screener both learn nothing.
The same logic — wording for comfort, categories for analysis — extends to the rest of the demographic block. For each: the practical wording, and the analysis reality the template libraries skip.
“How do you describe your gender?” — Woman, Man, a self-describe option, and Prefer not to say. The analysis reality nobody states: categories below roughly 30 effective base cannot be significance-tested, so small groups appear in tables but not in tested comparisons. Design the wording for respect, and the analysis plan for base sizes.
Ask in brackets, name the household and the year (“What was your total household income in 2025, before taxes?”), and expect the highest refusal rate of any demographic — 10 to 15% choose Prefer not to say, which is why the option must exist. Align brackets to round census-style breaks so weighting stays possible.
Highest level completed, in 5 to 7 steps from “Less than high school” to “Graduate or professional degree.” Resist the urge to enumerate every credential: education is usually a weighting and crosstab variable, and census-aligned steps serve both.
Use your census bureau's current categories with a multi-select and a self-describe write-in — in the US, that framework is set by OMB standards, updated in 2024. Deviating from official categories feels creative until weighting day, when your data no longer maps onto any population benchmark.
Ask the level you will actually analyze — state or province, not street address. Finer geography raises privacy stakes and refusals while adding nothing a crosstab will ever use. If respondents arrive from a contact list, geography often rides in as embedded data and needs no question at all.
Two block-wide disciplines. Ask only what the analysis plan will use: every demographic question spends respondent goodwill, and a question no crosstab ever touches bought nothing with it. And give every sensitive item an explicit Prefer not to say — the refusals it absorbs would otherwise surface as abandonment or fiction. Where each question type fits mechanically is covered in survey question types; what the finished block feeds into is how to analyze survey results.
Year of birth, when analysis precision matters: it never goes stale across tracker waves, requires no mental arithmetic under a deadline, and any band set can be derived from it later. Ask exact age for short one-off studies, and use age bands when age drives a screener or quota, because bands feed quota cells directly. The one rule that overrides all preferences: data collected in bands can never be made finer afterward.
The census-aligned set is the safe default: 18–24, 25–34, 35–44, 45–54, 55–64, 65+. It matches published population targets, so weighting works, and most syndicated research uses it, so your results compare. Marketing bands (18–24, 25–34, and so on with a 16–17 extension) and generation labels are fine for reporting flavor, but collect in census-aligned bands and relabel at reporting time, not the reverse.
It depends on what age does in the study. If age gates eligibility or fills quota cells, it must go early, in the screener, as a banded single-choice — a quota cannot act on a question asked after the respondent has finished. If age is purely a profiling variable, it belongs at the end with the other demographics, where personal questions cost the least trust. The common mistake is asking it twice; once, in the right place, is enough.
Because the alternative is worse: forced disclosure produces abandonment and fabricated answers, both invisible in the data. A Prefer-not-to-say rate is honest missingness you can see and handle — those respondents simply carry weight 1 in raking and sit outside age-banded tables. Expect low single digits for age; if the rate runs higher, the question's placement or the survey's trust story needs attention before the wording does.
Related: survey screening questions · survey question types · survey weighting, explained · Likert scale examples
Brief the agent and it builds the block: age screener with terminates, quota cells on census-aligned bands, weighting targets wired to match. From $99 per study.
Start a study