Methodology
The data for this report come from the Reuters
Institute Digital News Report surveys. In the 2024
and 2025 surveys we asked respondents who said
they used social or video networks for news to tell
us where they paid most attention when it comes to
news (news brands and journalists, news creators,
etc.). We then asked respondents to name some of
the news brands and individual creators that they
paid most attention to using open text fields.
In this report we analyse in detail these open responses from both
the 2024 and 2025 surveys, counting mentions of both individuals
and news brands across the dataset to identify the most
frequently mentioned in each category. We supplement these
lists with desk research and consultation with country experts.
In the 2025 survey, all respondents who said they used Facebook,
X, YouTube, Instagram, Snapchat, or TikTok for news in the last
week were asked, for one randomly selected network they used,
which sources they paid most attention to (traditional news
media/journalists, digital-first news outlets not associated with
traditional media, creators/personalities who mostly focus on the
news, creators/personalities who occasionally focus on the news),
and then, for each source type, to name up to three. In the 2024
survey, respondents were presented with a slightly different set of
sources and, if they were selected, were asked to name up to three
mainstream outlets or journalists, and up to three alternative
news sources or online personalities or celebrities. See the Digital
News Report website for the full questions.
5
These questions were designed to provide enough breadth in terms
of different social networks and different source types to be able to
map the ecosystem and construct our typology. These data cannot
be used to reliably estimate the proportion who pay attention to
each individual or brand, though they can be used to highlight a
selection of creators and influencers who are among the most
widely encountered in each country. Moreover, surveys capture
people’s self-reported behaviour, which does not always reflect
people’s actual behaviour due to biases and imperfect recall.
We used open text response boxes to collect the data for several
reasons. First, because in many countries the most popular news
creators and influencers have not yet been identified by previous
research. Second, because it would likely not be possible to fully
capture the broad and fragmented nature of this ecosystem using
a fixed listed of response options. And third, because we wanted
to adopt an audience-centric approach whereby respondents
could enter names that they considered news sources to them,
even if they did not meet accepted standards or definitions within
academia or the journalistic profession. This means that many of
the names we list here would perhaps have been excluded under
a more top-down approach.
Although open text responses enable an audience-centric
approach, and research shows that they can be a more accurate
way of measuring people’s news exposure (Guess 2015), they are
more challenging to analyse at scale. This is because misspellings,
shorthand references, low-quality responses, and other errors all
have to be cleaned and recoded, and groups of responses have to be
merged into meaningful categories. We used two parallel processes
to clean and process the data. We applied a range of commonly
used text editing and clustering algorithms using OpenRefine, and
supplemented this with manual editing and checking within the
software by the authors. We also used ChatGPT5 to process and
recode the original data, to identify the most mentioned individuals
and news brands, and to provide pen-portraits for each in English.
Previous research has shown that large language models can very
reliably recode open survey responses (Mellon et al. 2024) and
annotate short texts across multiple languages (Heseltine and
Clemm von Hohenberg 2024). Both sets of responses were
manually checked against the original data, with any discrepancies
resolved by the authors. We also added follower/subscriber counts
on social media for each individual as a rough proxy for the size of a
particular individual’s audience. We also asked country experts to
review all information and provide a sense check. It is important to
acknowledge, however, that this kind of analysis, even when
undertaken by humans, can be subject to error and necessarily
involves judgement calls. This means that, in addition to the usual
error and uncertainty associated with surveys (see the Digital News
Report methodology for a specific description) there is the potential
for additional error in the order of the names in the lists, and on
either side of the necessary cut-off point. Our lists are inclusive in
terms of being faithful to the names mentioned by respondents.
We removed just a handful of actors, sports stars, and celebrities if
we were sure they did not post on any news-related issues.
We applied the same process in 24 countries, which were chosen to
provide geographic diversity (Europe, Americas, Africa, and Asia-
Pacific) as well as a good mix of different media systems. The
countries chosen were United States, Canada, Mexico, Brazil,
Argentina, Colombia, United Kingdom, Norway, Netherlands,
Germany, France, Spain, Czech Republic, Poland, Australia, India,
Indonesia, Philippines, Thailand, South Korea, Japan, Kenya, Nigeria,
and South Africa. In order to achieve the biggest sample of names,
we used data from open fields in both our 2024 and 2025 surveys, in
all countries except Nigeria, Kenya, Colombia, Mexico, Indonesia, and
the Philippines where data were drawn only from our 2025 survey
because the questions were not asked there in 2024.
Data from India, Kenya, and Nigeria are representative of younger
English speakers and not the national population, because it is
not possible to reach other groups in a representative way using
an online survey. The survey was fielded mostly in English in these
markets, and restricted to ages 18 to 50 in Kenya and Nigeria.
Findings should not be taken to be nationally representative in
these countries.
Reuters Institute Digital News Report methodology: https://
reutersinstitute.politics.ox.ac.uk/digital-news-report/2025/
methodology
5
Although slightly different questions were used in 2024 and 2025, we believe the data are broadly comparable and, given that the primary purpose is to map this space, that the added depth
from two years’ worth of data outweighs any comparability issues.Reuters Institute for the Study of Journalism | Mapping News Creators and Influencers in Social and Video Networks
8