You ran a poll on one of your own posts and 40 readers answered. The draft that cites it has been open for days while you hunt for the sample size that makes a poll safe to publish, and every calculator you try asks for a population size and a confidence level you don't have.
Those 40 respondents picked themselves. They found the page, then they chose to vote, and a sample built from two rounds of self-selection has no margin of error at 40 responses or at 4,000, because margin of error describes random draws and this was never one. So the question worth your time shifts: what sentence can 40 self-selected answers actually hold?
Why Poll Sample Size Has No Magic Number
Search for a poll response threshold and nearly every page serves the same figure: 384 responses, out of Cochran's formula, for a margin of error of plus or minus five points at 95% confidence. The formula does exactly what it promises for the survey it was built for, one where respondents are drawn at random from a defined population and every member has a known chance of being selected.
Nothing about a blog poll meets that condition. Whoever was reading saw the question, and whoever felt like answering clicked. More votes run the same self-selection filter more times; no count converts an opt-in audience into a random draw, so no count unlocks the permission the calculators imply.
Which claims are worth polling at all is a separate, earlier decision, covered in how to poll your own audience for the number. Here, the votes are already in.
Statistical Significance Needs a Probability Sample
Reaching for significance is the right instinct, the same one that keeps a careful team from publishing a coincidence as a finding. Significance tests, confidence intervals, and margin of error all require a probability frame, a population in which each member's chance of selection is known. A reader poll never had one, so those tools have nothing to compute at any response count.
The American Association for Public Opinion Research says this outright: "it is impossible to develop statistically valid margins of sampling error from nonprobability surveys, such as opt-in, online polls." When an editor asks what your poll's significance is, the accurate answer is that the data can't produce one, and the useful things to hand over are the denominator, how respondents were recruited, and the window the poll ran.
LiquiChart's staleness engine applies the same discipline to the numbers it monitors: when it can't defend a verdict on a figure, it flags the figure for review rather than guessing. Holding your own poll to that standard, a claim you can fully defend at its real size, is also where publishing original research without a budget begins.
The natural fallback is scale: maybe enough responses dilute the bias. Pew Research Center checked, benchmarking opt-in online samples against known population values, and measured the gap: "the average absolute error for the opt-in samples combined was about twice as large at 5.8 points," against 2.6 points for probability-based panels. Piling on opt-in responses tightens the estimate around a number that's still biased.
Every team that publishes polls runs on some rule for when a result is ready, usually an unwritten one. Name the rule you actually follow.
One of those four rules lets you publish the day the first responses arrive, because it fits the claim to the votes in hand. The other three keep the draft parked while the count grows toward a bar nobody set.
The rule you reach for first is usually one you inherited without ever choosing it. A team that learned polling from probability-sample methodology treats the response count as the gate, then stalls at every sample too small to clear a threshold that was never theirs to clear. The rule you pick decides how many of your polls you will publish, long before it decides how you word any single one.
The Honesty Ladder for Poll Claims
Start with the smallest real case. Your poll closed at 24 responses and 18 picked the same option. Publish it as 18 of 24 readers who responded. Every word in that sentence is checkable, and you can ship it this afternoon.
Leave the poll open longer and the sentence can grow. At 210 responses with 130 on one option, you can write 62% of 210 readers who responded. The percentage became defensible because the count underneath it grew. That's the honesty ladder: each rung is a wider claim, and the responses you've actually collected decide how high you can go.
Wiring a poll into a post so it keeps collecting is covered in how to create a poll for a blog. The rungs are about what you say once the votes exist.
Match the Claim to the Response Count
With a couple dozen votes, publish integers. A percentage on 24 responses reads like a rate you could project onto a population, and the sample can't fund the projection; the raw count keeps the claim the same size as the evidence.
In the low hundreds a percentage becomes fair, on one condition: the denominator and the audience stay in the sentence. Write 62% of 210 readers who responded, with both qualifiers attached. Strip them and the sentence describes a world you never measured.
Sampling Method Caps the Claim
How you gathered the votes sets a ceiling that no count raises. A self-selected poll can describe the readers who answered and nothing beyond them, so "of professionals" and "of the industry" stay out of reach no matter how many responses arrive. The calculators skip this because they assume a population frame. The moment you decided to ask your own readers, the frame became your own readers, and every honest sentence lives inside it.
Question Wording Shapes the Result
Ask readers what they themselves do, which CMS they run, how often they publish, and each response is a small first-party fact; together they describe the people who answered. Ask readers to estimate what the industry does and you've gathered impressions no response count can firm up. Sorting your claims into self-report questions happens before the poll ever opens.
Disclosure Makes a Small Poll Citable
One line, printed next to the number, does more for its credibility than another thousand votes: Self-selected readers of this blog, N=210, responses collected in March. That line tells an editor or a fact-checker exactly who answered and when, which is everything they need to weigh the number, and supplying it is the price of getting quoted. A count with its limits printed beside it can survive scrutiny; I've never once regretted attaching that line to a number I published.
What journalists screen for before they'll quote a poll, and how to build a methodology page they can verify, is laid out in the four disclosures journalists verify. Soft-gate self-certification in LiquiChart asks respondents to confirm they fit the audience before a vote records, which sharpens the denominator on that disclosure line; it's included in any paid plan.
Trend Polls Grow the Sample Over Time
A poll you run once hands you a fixed count and a fixed date, and both age. A trend poll keeps the same question open across successive periods, recording each period's responses separately, so the denominator you can disclose grows on the audience's own schedule. Repetition also adds evidence a single count can't: a result that holds at 40 responses this quarter and holds again next quarter starts to behave like a pattern, with every period on the record.
Watching a first-party number move across periods is the heart of learning to track audience sentiment over time. Trend polls in LiquiChart are free on every plan.
Self Selected Polls Cannot Support Segments
Past a few hundred responses the temptation changes shape: slice the result by role or industry and report what each group believes. A self-selected sample can't pay for those cuts at any size. The crosstabs come out looking precise, and the precision is hollow, because no probability frame ever stood under the slices; a bigger sample only sharpens the illusion.
Demographic prequalifiers in LiquiChart, available on any paid plan, let a poll require respondent attributes before a vote counts. The judgment that keeps you from over-reading a segment is the same judgment that keeps you from buying a cut your responses can't support, and when the breakdown won't hold, your recruitment disclosure will do more for a fact-checker than any crosstab of an opt-in poll. How to disclose recruitment before demographics is worked through in the citability playbook.
Publish the Sentence Your Sample Supports
Your poll already measured something real about the people who answered it. From the first vote, the only thing between that measurement and publication is a sentence sized exactly to the evidence, with your name under it. The gate was never the count. It was always the claim.
Every week the draft waits for a rounder number, a scoped, disclosed, citable fact sits unpublished on your own page. I held mine back longer than I should have, waiting for a threshold nobody could name. Publish the 40.