For example, let's say I control a popular app which installed on people's devices -- thousands of developers/corporations/other entities are in this position. I insert some new code in the app (perhaps via an automatic update) that sends a Google query for some search term from a pool of similar terms I have chosen. I can easily mimic the HTTP headers sent by a common web browser, thereby making my request indistinguishable from one that is intentionally submitted by a user, chosen from a list of common browser headers.
By having large swathes of users all appearing to search for the same or similar terms I can make it appear to, say, the Economist, that the public at-large is interested in a given topic. From there, the Economist will make assumptions about the population at-large and then make conclusions about what this means. Then they will publish these "findings" as a story. Many readers will be swayed by the Economist's conclusions. At that point I will have effectively influenced what those readers now believe. All it took were some phoney searches.
Later, I could remove the code in the next auto-update.
By the time something like this was discovered by a user MITM'ing the app traffic and reverse engineering the app, chances are I would have already accomplished my objective: to influence people -- perhaps investors -- to believe a story like the one reported here by the Economist. (Note: I am not suggesting the Economist's conclusions are wrong in this case.)