What Is a Survey for Press Coverage?
A survey for press coverage is original research commissioned by an organisation to produce findings that are newsworthy in themselves, published with enough methodological detail for a journalist to judge whether the numbers can be reported.
It differs from internal market research in one way that shapes every step: the result will be read by people who have no reason to trust the sponsor. A reporter deciding whether to use your figure is asking who paid, who was asked, how they were found, what exactly they were asked, and how precise the number is. Professional standards from AAPOR in the United States and the British Polling Council in the UK set out what those answers should include. This guide applies them to a PR survey. It complements our guides to writing a press release and pitching a journalist by email.
How to Run a Survey for Press Coverage at a Glance
| Step | Action | Expected result |
|---|---|---|
| 1 | Write the news question and define the population | One sentence a journalist could use as a headline, and a precise “who” |
| 2 | Choose the sample source and record whether it is probability or opt-in | A named supplier and a stated recruitment method |
| 3 | Set the sample size from the smallest subgroup you will quote | A target n and an expected margin of error |
| 4 | Draft neutral questions and fix their order | A final questionnaire you will publish in full |
| 5 | Field the survey with quality checks | A cleaned data file with exclusions documented |
| 6 | Weight, analyse and test differences | Findings that exceed sampling error |
| 7 | Publish the method and tables, then pitch | A methods page and data tables on a stable URL |
Time: typically several weeks from question to release (our estimate; fieldwork length depends on the supplier). Tools: a survey platform or research agency, a spreadsheet or statistics package, a web page for the method. Cost: set by the sample supplier; ask for a quote that states the recruitment method. Skill level: intermediate, with a researcher involved.
What You Need Before Starting
- A news question tied to something journalists already cover; see what a news hook is.
- A research partner or panel supplier who will disclose how respondents are recruited.
- A named owner for the methodology statement, and sign-off that unfavourable results will still be disclosed.
- A page on your site to host the method and tables, such as an online newsroom.
- A short list of reporters on the beat; see how to build a media list.
Step 1: Write the News Question and Define the Population
Write the finding you hope to report as a question with an unknown answer, then define precisely who can answer it. A survey built backwards from a desired headline is what reporters are trained to distrust.
Process: state the question in one sentence (“What share of UK small-business owners have used an AI tool for bookkeeping in the past year?”), then define the population. AAPOR asks researchers to “be specific about the decision rules used to define the population,” including location, age and other characteristics. “Consumers” is not a population; “adults aged 18 and over in Great Britain” is. Expected result: a question and a population definition you will print in the method. Verify: a colleague can say who would and would not qualify. Common mistake: generalising from a narrow sample, such as reporting customers' views as the public's.
Step 2: Choose the Sample Source and Say What It Is
Decide whether you will use a probability sample (people selected from a known frame with a known chance of selection) or an opt-in panel (people who signed up to take surveys). Then disclose it.
AAPOR's standards ask researchers to state which applies and, for a panel or list, to name “the supplier of the sample or list”. The choice matters for accuracy. Pew Research Center found that in an opt-in survey, 20% of adults under 30 agreed the Holocaust is a myth; on its probability-based panel, “Rather than 20%, we found that 3% of adults under 30 agree with the statement.” Pew says the gap is what it would expect to see if there were “a large number of bogus respondents in the opt-in poll” claiming to be under 30. Expected result: a named supplier and recruitment method. Common mistake: calling an opt-in sample “nationally representative” without explaining how it was weighted.
Percent, as reported by Pew Research Center; first three rows are adults under 30, last row is all Americans
Source: Pew Research Center, “Online opt-in polls can produce misleading results, especially for young people and Hispanic adults”, March 5, 2024.
The opt-in and probability figures for the Holocaust question come from different surveys a month apart, as reported by Pew. The submarine question was designed to detect respondents who answer without reading.
Step 3: Set the Sample Size From Your Smallest Subgroup
Choose the sample size by working back from the smallest group you intend to quote, not the total. Precision falls quickly as groups shrink.
Pew's explainer gives the numbers: “A simple random sample of 1,067 cases has a margin of error of plus or minus 3 percentage points,” but a subgroup of about 160 has “a margin of error of plus or minus 8 percentage points”, and plus or minus 16 points for the difference between two answers within that group. The margin for a difference between two figures is “generally about twice” that of a single figure. Process: list every subgroup you might headline (by region, age, job role), estimate its share of the sample, and increase the total or boost that group until its margin is acceptable. Expected result: a target n with a margin stated for the total and key subgroups. Common mistake: headlining a regional finding from a few dozen respondents.
Plus or minus percentage points at 95% confidence, Pew Research Center worked example
Source: Pew Research Center, “5 key things to know about the margin of error in election polls”, September 8, 2016.
Figures assume simple random sampling. Weighting and opt-in samples change precision; AAPOR says non-probability surveys should report precision only with a description of the model used.
Step 4: Draft Neutral Questions and Fix the Order
Write questions that do not suggest an answer, offer balanced response options, and place your topic questions before anything that could prime respondents. You will publish them, so write them to be read.
Process: avoid loaded wording (“How concerned are you about the growing problem of...”), include “don't know” where reasonable, and keep screening questions honest. AAPOR asks for “The exact wording and presentation” of questions and “any preceding contextual information that might reasonably be expected to influence responses”. The BPC asks members to publish “the exact questions asked in the order they were asked”. Expected result: a final questionnaire. Common mistake: asking about your product category immediately after showing respondents your brand.
Step 5: Field the Survey With Quality Checks
Run the survey with checks that catch careless, duplicate and automated responses, and log every exclusion.
AAPOR lists the checks it expects researchers to disclose: “attention checks, logic checks,” exclusion of respondents “who straight-lined or completed the survey under a certain time constraint,” screening “for evidence that it originated from bots or fabricated profiles,” and “measures to prevent respondents from completing the survey more than once”. A question with a known answer, like Pew's submarine licence item, is a simple trap. Do not replace or top up respondents with AI-generated answers; our report on whether a PR survey can use AI respondents covers Pew tests across nearly 300 questions in which AI estimates “differed from their human counterparts by an average of 12 percentage points” (Pew). Expected result: a clean file and an exclusions log. Common mistake: deleting inconvenient responses without a pre-set rule.
Step 6: Weight, Analyse and Test Differences
Weight the data to known population figures if needed, then report only differences larger than sampling error.
Process: record the weighting variables and their sources, since both AAPOR and the BPC require them. Before writing “more X than Y”, check that the gap exceeds the margin for a difference, which Pew says is about twice the single-figure margin. Report percentages with their base, such as “42% of 1,012 adults”. If a finding contradicts your hypothesis, keep it; the BPC says that if later findings “cast doubt on, or differ significantly from those that have been published,” they must also be released. Expected result: a short list of defensible findings. Common mistake: calling a 2-point lead “more likely” in a sample of 1,000.
Step 7: Publish the Method and Tables, Then Pitch
Put a methods statement and full data tables on a stable URL before the release goes out, and link it from every pitch.
The BPC requires its members' published polls to state the “Client commissioning the survey; Dates of interviewing; Method of obtaining the interviews”, the universe, the percentages and the “Size of the sample and geographic coverage,” with full question wording and a link to tables “Whenever it is practical”. Members must post full tables and sampling details “within 2 working days”. AAPOR requires its Section A items at release and Section B items “within 30 days of any request”. Do both at release. Then pitch the one finding that matters most to each reporter's beat; our report on how editors decide what to cover explains what makes data usable. Expected result: a methods page that answers every question a reporter would ask. Common mistake: sending a headline percentage with “full results available on request”.
Classification from BPC and AAPOR rules, not a measurement
Source: British Polling Council, Objects and Rules; AAPOR Disclosure Standards, read October 10, 2026.
Bar length ranks immediacy (longer means sooner); it is ordinal, not a measured quantity. BPC rules bind its members' social and political polls in the UK; AAPOR's are professional standards.
How to Verify It Worked
- Disclosure complete: your methods page names the sponsor, the fieldwork company, dates, population, recruitment method, sample size, weighting and full question wording.
- Numbers defensible: every headline finding exceeds its margin of error, with the base stated.
- Coverage accurate: articles that use the data describe the sample correctly. Track pickups with alerts; see how to monitor brand mentions with Google Alerts.
How to Choose a Sample Approach
| Situation | Recommended approach | Why |
|---|---|---|
| Headline claim about the general public | Probability-based panel | Pew found large errors in opt-in samples on some groups |
| Professional audience (for example, IT buyers) | Vetted B2B panel with screening and disclosed recruitment | Few probability frames exist; disclosure lets reporters judge |
| Your own customers | Customer list survey, reported as customers only | Honest about the population |
| Small budget | Fewer questions and subgroups, larger n per group | Precision matters more than breadth |
| Need results this week | Postpone, or report as indicative with full caveats | Rushed fieldwork weakens quality checks |
Common Problems and Fixes
| Problem | Likely cause | Fix |
|---|---|---|
| Reporter asks for the method and you cannot answer | Supplier did not disclose recruitment | Get recruitment, weighting and exclusions in writing before fieldwork |
| Implausible result among young respondents | Bogus respondents in an opt-in sample | Add attention and known-answer checks; review exclusions |
| Subgroup result swings wildly | Too few cases | Do not headline it, or boost the subgroup |
| Finding contradicts your message | Hypothesis was wrong | Publish it; suppressing it breaches the standards cited |
| Coverage misstates the sample | Release summarised it loosely | Put the population and n in the first paragraph of the release |
What Not to Do
- Do not write the headline first and the questions second. Consequence: leading questions that reporters spot. Correction: start from an open question, as in Step 1.
- Do not use AI-generated respondents. Consequence: figures that do not reflect people. Correction: field to real respondents and disclose the source.
- Do not hide the sponsor. Consequence: lost credibility when it emerges. Correction: name the sponsor and fieldwork company in the release and method.
- Do not quote subgroups without bases. Consequence: false precision. Correction: give n for every figure.
- Do not release only the favourable questions. Consequence: a breach of the BPC rule on related findings. Correction: publish full tables.
Frequently Asked Questions
How many respondents does a PR survey need?
Enough for the smallest group you plan to quote. Pew's example shows about 1,067 respondents give a margin of about 3 points overall, while a subgroup of about 160 gives about 8 points. Work back from your subgroups.
What must I disclose with survey results?
At minimum the sponsor, fieldwork company, dates, method, population, sample size and percentages, plus full question wording and tables where practical. These follow the British Polling Council and AAPOR disclosure standards.
Are opt-in online panels acceptable?
They are widely used, but they must be disclosed as opt-in and checked carefully. Pew found opt-in samples can include many bogus respondents, especially among young adults, producing results far from probability-based estimates.
Can I report a margin of error for an opt-in survey?
AAPOR says non-probability surveys should report precision only with a detailed description of the model used, its assumptions and how the measure was calculated. Otherwise, describe the sample without a margin.
Should I publish results that do not support my message?
Yes. The BPC rules require members to release later findings that cast doubt on, or differ significantly from, published ones. Selective release also undermines trust if a reporter asks for full tables.
Can AI-generated respondents replace a real sample?
Not for a published number. In Pew tests summarised in our report on AI respondents, synthetic answers missed the human results by 12 points on average.
Sources and Methodology
This guide is built from the British Polling Council's rules, AAPOR's disclosure standards and two Pew Research Center explainers listed below, read on October 10, 2026, together with Pew's September 2026 study of synthetic respondents. We did not commission or field a survey for this guide, and we did not test any panel supplier. BPC rules bind its members' UK social and political polls and AAPOR's are professional standards; neither is law. Timings and costs vary by supplier, and the sample-size figures are Pew's worked examples under simple random sampling.
- trade body British Polling Council, Objects and Rules, rules 2.1 to 2.3 on disclosure. Read October 10, 2026.
- standards American Association for Public Opinion Research, Disclosure Standards, items for immediate disclosure and data-quality checks. Read October 10, 2026.
- research Pew Research Center, 5 key things to know about the margin of error in election polls, September 8, 2016. Read October 10, 2026.
- research Pew Research Center, Online opt-in polls can produce misleading results, especially for young people and Hispanic adults, March 5, 2024. Read October 10, 2026.
- research Pew Research Center, Can AI stand in for human survey takers? Not really, September 30, 2026. Read October 10, 2026.
- guide Media Morpheus, Can a PR Survey Use AI Respondents?