On making stuff up

Well, no. "Most" don't, which is why you should read the story before writing the hed, no matter what you want to believe.

The second graf tells you what you need to know about the hed. How many people are "aware" of the Ryan plan? About 72%. Multiply .72 by the number of those who oppose the plan -- 49% (.49), traditionally considered a little less than "most" -- and you get a reasonable idea of the overall proportion who oppose the plan: about 35%, or slightly more than a third.

A third of what? Good question. It's one of the essential things this brief is missing. We also can't place the reported results in a temporal context -- when was the "poll" taken? -- or judge how good of a guess about public opinion it represents (that's your "margin of error"). All of that could be described in a single clause that describes the sample size, the population, and the time in the field. For example:

The telephone survey of 900* registered voters was conducted Aug. 16-19.
 
If someone tells you there isn't room to spare, your answer is simple: Any poll that isn't worth a dozen words to explain its mechanics isn't worth reporting.

Observational aside here. I don't think the Freep -- the purportedly liberal side of the  Detroit JOA -- is deliberately baking survey data to favor the Party of Chardonnay and Socialism. It's more likely that the Freep doesn't understand that you're supposed to do a little basic arithmetic before you talk about public opinion; after all, lots of successful journalism consists of making culturally congruent remarks that have little to do with the evidence at hand. But organizations that don't want to be mistaken for Fox News ought to be a little more careful.

* It'd be nice to report the maximum margin of sampling error at 95% confidence, but you can approximate that for yourself given the sample size.