"How many responses do I need?" is the first question every researcher asks and the one most survey tools answer worst. The honest short answer for a general consumer study is 384. The honest long answer is: it depends on the precision you need and the subgroups you plan to compare - and both are simple to work out.
The formula, briefly
For a proportion estimate (e.g. "what share of people prefer A?"), the required sample at 95% confidence is n = 1.96² × p(1-p) ÷ e², where p is the expected proportion (use 0.5 for the safe maximum) and e is the margin of error you can tolerate. Population size barely matters until you are surveying more than a few percent of the whole population - a survey of Saudi consumers does not need more responses than a survey of Bahraini consumers for the same precision.
The lookup table
| Margin of error | 95% confidence | 90% confidence |
| ±10% | 97 | 68 |
| ±5% | 384 | 271 |
| ±4% | 600 | 423 |
| ±3% | 1,067 | 752 |
| ±2% | 2,401 | 1,691 |
Read it as: with 384 responses, a result of "62% prefer A" means the true value is very likely between 57% and 67%. If that range is precise enough for your decision, 384 is enough. Notice the brutal economics at the bottom of the table - halving the margin of error quadruples the sample. Precision is bought at a steeply rising price, which is why you should decide the decision first and the sample second.
The subgroup trap
The formula applies to every group you will analyze separately, not to the total. If you need to compare men vs women across three age bands, that is six cells - and a 400-response study leaves ~65 per cell, with a margin of error near ±12%. This is the most common sample-size mistake we see: studies sized correctly for the total and uselessly for every comparison the researcher actually cared about.
Size for the smallest cell
Decide the finest comparison you truly need, size that cell to at least 100 (±10%), and let the total follow. If the total gets too expensive, cut comparisons - not precision.
Practical presets
| Study type | Recommended n | Why |
| Quick directional read / pilot | 50-100 | Detects big effects; fine for iterating on ideas |
| Standard consumer study | 384-500 | ±5% on the total, headroom for light subgroups |
| Segmentation / tracker wave | 1,000+ | ±3% on total, ±10% on 4-6 segments |
| B2B / niche audience | As many as feasible | Precision yields to reachability; report the achieved margin honestly |
Sample size buys precision, not truth - a perfectly sized survey with biased questions is precisely wrong. Pair this with our guide to writing good survey questions before you field.