Most of the sensationalist claims present in science news seem to originate from press releases, not from newsrooms. A recent study shows that the original findings presented in a peer-reviewed journal article appear exaggerated in popular news stories almost always, if they are exaggerated in a related press release. If the press release does not exaggerate the original findings, journalists only rarely add exaggeration to their stories. The study in question focused on health related science news.
The findings are correlational, but the authors imply causation as previous research suggests journalists rely mostly on press copy in writing science news. The article was authored by Petroc Sumner, Solveiga Vivian-Griffiths, Jacky Boivin, Andy Williams, Aimée Davies, Jack Ogden, Leanne Whelan, Bethan Hughes, Bethan Dalton, Christopher Chambers, all of Cardiff University,
Christos Venetis, of University of New South Wales and Fred Boy, of Swansea University (names not in original order).
The results are published online in an article of the journal BMJ (formerly known as The British Medical Journal), and are freely available here.