Data-led PR campaigns: finding, analysing and packaging original data
A data-led PR campaign turns a dataset into one clear finding, publishes it with a full method note and offers journalists cuts for their own region or beat. The method note matters as much as the headline, because journalists check where a number came from.
In short
- Data-led content is the default digital PR format, pitched by 95.9% of practitioners in BuzzStream's State of Digital PR 2026.
- Data can come from four places, which are your own records, public datasets, a commissioned survey or collected listings, and each has a different cost and weakness.
- A consumer survey for PR usually has 1,000 to 2,000 respondents, a practitioner norm that matches Researchscape's 2024 guidance of at least 1,000 for consumer studies.
- YouGov's Anthony Wells told IJNet in 2018 that a sample of 1,000 carries a margin of error of about 3 points and that subgroups under 100 need extreme caution.
- The AAPOR Transparency Initiative lists what a credible survey discloses, including the sponsor, exact question wording, population, dates, sample size and weighting.
A data-led campaign has three parts: a dataset nobody else has put together, one finding a journalist can use as a headline, and a method note that lets them trust it. Find or collect the data, analyse it until you have a single clear story, then package it with regional or sector cuts so that many outlets each get their own version.
It is the standard format. In BuzzStream’s State of Digital PR 2026, 95.9% of practitioners said they pitch data-led content. Demand is there too: in Cision’s State of the Media 2026, a survey of 1,899 journalists, 47% asked PRs for data or research, a figure relayed by BuzzStream because Cision’s full report is gated. Choosing the idea is covered in digital PR ideation. This lesson is about the data. With nearly every practitioner pitching the same format, the data itself is where a campaign is won or lost, because a reporter offered several data stories has to decide which numbers can be trusted and explained.
Four places to get data
| Source | What it is | Strength | Weakness |
|---|---|---|---|
| Your own records | Sales, searches, bookings, prices or usage from your business | Nobody else has it, and you are the obvious authority | It describes your customers, not the population, and may need legal sign-off |
| Public datasets | Government, statistics office and regulator data | Free and trusted | Anyone can use it, so the value is in your analysis |
| Commissioned survey | Questions put to a panel by a research firm | You choose the question | Costs money, and quality depends on the sample and the wording |
| Collected listings | Prices, reviews or listings gathered from public pages | Fresh and specific | The source’s terms and the law may limit use, and the sample can be skewed |
Each row trades one advantage for one weakness, and the weakness decides how the finding can be worded. Your own records support a statement about your customers and nothing wider, so a headline that slides from “our users” to “people” has claimed more than the data holds. Public data is trusted because of where it comes from, but the same tables are open to everyone, which puts all the originality into the way you combine or rank them. A survey lets you ask exactly the question you want answered and then makes you responsible for how it was asked. Collected listings need a check on what the source allows before anyone starts gathering them.
Practitioners commonly name Censuswide, OnePoll and Pollfish as panel providers for PR surveys. Survey fieldwork runs to roughly $1,500 to $5,000 as a rule of thumb, and whole campaigns are often quoted at $3,000 to $15,000 with a lead time of three to eight weeks. None of those is a measured average. They are useful for a first conversation about budget and for noticing a quote that sits far outside the range, and they should not be read as a price list. On those rough figures the fieldwork can be a large part of a small campaign and a modest part of a big one, so the choice of source in the table is also a choice about where the money goes.
From data to one finding
Data is not a story. Speakers at Digitaloft’s Digital PR Summit in May 2026 put it plainly: data on its own does not land, and it needs a narrative. A working method starts by writing the question first. “Which city is cheapest for first-time buyers?” gives the analysis a direction, where a dataset opened without a question can produce a pile of observations and no finding. Next, clean the data and note every decision. Removed outliers, merged categories and excluded dates all go in the method note later, and they are far easier to record as you make them than to reconstruct when a journalist asks.
Then look for one of four shapes: a ranking, a change over time, a gap between groups, or a surprising total. These are the forms a headline can take, so finding one of them in the data is the same as finding the story. Pick one headline finding and two or three supporting ones. More than that and the pitch has no centre. Finally, check the finding against a second source where one exists. If your number contradicts official figures, find out why before a journalist does. There may be a good reason, such as a different definition or a different period, and knowing it in advance turns an awkward question into a line in the method note.
In BuzzStream’s 2026 report, 60.8% of practitioners said they use AI for data analysis. It speeds up the work, but the numbers still need checking by a person who will put their name to them. The check matters more here than in most uses of such tools, because a wrong figure in a pitch is published under your brand and, if it is caught, corrected in public under your brand as well.
The methodology standards journalists expect
We found no survey that measures how often journalists reject a story for a weak method. What exists is guidance written for reporters and standards written for researchers, and both tell you what a careful journalist will look for. Accuracy is on their minds: half of the journalists in Cision’s 2026 survey named accuracy and combating misinformation as their top challenge. On sample size, the practitioner norm for consumer surveys is 1,000 to 2,000 respondents. Researchscape, a survey firm, recommended in December 2024 at least 1,000 respondents for consumer studies, 500 or more for broad business audiences and at least 150 for narrow business ones. Those are a vendor’s rules of thumb, and they match common practice.
Margin of error and subgroups come next. In a 2018 IJNet article for journalists, Anthony Wells of YouGov said that a sample of 1,000 carries a margin of error of about 3 points, so a difference of a few points may mean nothing. He advised extreme caution with groups of fewer than 100 people and said groups under 50 should not be used for comparisons. This is where regional cuts often break: a national sample of 2,000 can leave a small city with a few dozen answers. The very thing that makes a survey useful to local outlets, a separate figure for each place, is the thing the sample is least able to support, so decide which cuts you intend to publish before fieldwork and size the sample for the smallest of them.
Who was asked matters as much as how many. The same article warns reporters off open polls that anyone can answer, such as a poll on social media or your own site. Wells’s advice was to rely on polls that use random or quota sampling to represent a population. A large number of answers from people who chose to respond describes those people and nobody else, which is the reason for the warning. As for what to disclose, the Transparency Initiative of AAPOR, the American association for public opinion research, lists what its members disclose with any published survey. In plain terms, that covers the following.
- who sponsored the research and who carried it out
- the exact wording of the questions and answer options
- the population studied and where
- how people were sampled and recruited
- how and when the data was collected
- the number of completed responses, and the sampling error where it applies
- any weighting, and how it was calculated
- quality checks, known limits of the design, and any use of AI in collection or processing
You are not bound by those rules, but they are a ready checklist. In the UK, Full Fact pointed out in 2012 that the Market Research Society’s code requires members to supply the technical information needed to judge published findings. A journalist can ask for your data tables, so have them ready. For non-survey data, no equivalent standard exists for indexes and rankings. The working equivalent is: name every source with its date, list each metric and its weight, show how the score was calculated, and say what was left out. An index is a set of choices about what to count and how much each thing matters, and a reader who can see those choices can judge the ranking, while one who cannot has only your word for it.
Packaging the campaign
Journalists should not have to ask for anything. A complete package starts with a page on your site holding the findings, the charts, the full tables and the method note. This is what earns the link, and it should be good enough to work as a linkable asset after the campaign ends. Alongside it goes a short pitch with the headline number in the subject line. See how to pitch journalists. The page and the pitch do different jobs: the pitch gets the story considered, and the page gives the reporter something to cite and gives later writers something to find.
The package also needs regional or sector cuts. A table by city, region or industry gives each local or trade outlet its own story. It needs a quote from a named expert who can be checked. Summit speakers in 2026 noted that journalists now look for invented experts, so a real name with a verifiable role has become part of the evidence. And it needs charts a newsroom can reuse, with the raw data on request. Decide early where links should point. Which pages need links explains how a campaign page passes value on to commercial pages through internal links. If that decision is left until coverage arrives, the links will go wherever each journalist happens to choose.
Why original data is worth the effort
There are two reasons, with different strengths of evidence. The stronger one is that recurring data earns for years. BuzzStream cites Bankrate’s annual Emergency Savings Report at more than 8,000 linking domains and Preply’s subtitle survey at about 800. A report that returns every year gives journalists a reason to come back to the same page, and each edition adds to the links earned by the ones before it. That changes the sum for the first edition, since its cost is spread over every later year in which the report is repeated.
The weaker reason is that it may help in AI answers. A 2024 paper by researchers at Princeton and IIT Delhi found that adding statistics, quotations and source citations to content raised its visibility in generative answers by up to 40% in a lab benchmark. That test used 2023 systems and on-page changes only, so treat it as a direction, not a forecast. How AI search chooses sources has the wider evidence. It would be a mistake to commission research on the strength of that paper alone. It is better read as a possible second benefit of work you would do for the coverage in any case.
Common faults
Six faults are worth checking for before anything is sent, and most of them break a standard described above. A headline built on a subgroup of 40 people falls below the size Wells said should not be used for comparisons. A leading question produces the answer the brand wanted, which the disclosure of exact wording is designed to expose. A difference of two points reported as a trend can sit inside the margin of error. A campaign may have no method note, or one that hides the sample size. Old public data may be presented as new. And an index may have weights chosen to put a favoured city on top. Any of these can cost you the story, and a journalist who catches one is unlikely to open your next pitch. That second cost is the larger one, because it falls on campaigns you have not yet run.
With the data ready, build a media list around the regional and sector cuts you have, so that each table has a set of outlets it was made for. The digital PR tactic page covers costs and campaign types, and digital PR agencies and tools lists suppliers. Terms are defined in the glossary.
Common questions
How many people do I need to survey for a PR campaign?
For a national consumer survey, 1,000 to 2,000 is the practitioner norm. Researchscape's 2024 guidance suggests at least 1,000 for consumer studies, 500 or more for broad business audiences and at least 150 for narrow ones.
How much does a data-led PR campaign cost?
Campaigns are often quoted at $3,000 to $15,000, which is a typical range and not a measured average. Survey fieldwork alone is roughly $1,500 to $5,000 as a rule of thumb.
Do I need original data, or can I use public data?
Public data works if your analysis is new. Re-cutting official figures into a ranking or an index is a standard format. Credit the source and say what you did to the numbers.
What should a methodology note include?
Who paid for and ran the research, where the data came from, the dates, the sample size, who was asked and how, the exact questions, any weighting, how any score was calculated and what the data cannot show.
Why do journalists reject data stories?
No reliable survey measures this directly. Journalism guidance warns reporters about open polls anyone can answer, small subgroups, differences inside the margin of error and missing method details, so those are the faults to avoid.

