To subscribe to the monthly C&E email newsletter and event announcements click here.

Campaigns know they need a presence on social media, but determining what to post in each channel remains a challenge. Generally, rather than explicitly marketing candidates’ ideas and values, campaigns focus on publicizing upcoming events and asking people to donate. This approach leaves voters craving more information, which they’ll seek elsewhere.

So what is the best frame to use for a tweet about healthcare? What are the most effective keywords to use in a Facebook post about the Trump administration? How do you combat a fake news story spreading online? Figuring out what to say on social media, and how to say it, can help move the needle for a campaign.

Knowing the right way to frame social media messages depends on understanding how voters see a candidate and the campaign’s issues. Social media content analysis, a dive far deeper than just listening in real-time, is an essential first step in any message development process.

Content analysis is about understanding the conversation.

Social media content analysis is much more than evaluating a campaign’s own posts and responses to them. Most social media conversations about candidates take place away from the campaign’s own social media channels.

By solely focusing on what people are saying in response to your campaign’s posts, you’ll never know if your voters are talking about other issues elsewhere, or if they are not engaging with your posts because the way you talk about issues is different from the way your audience talks about them.

Take, for example, the heated special election contest in Georgia’s 6th district. A social media content analysis reveals the substance and dynamics of a smear campaign against Democratic candidate Jon Ossoff.

Driven by a handful of blogs, far-right news websites and right-wing social media influencers, Ossoff is being characterized as being in the pocket of George Soros, Al Jazeera and Nancy Pelosi. Knowing this will help his campaign develop more effective messaging that can step around or dissipate this attack.

Accusations against Ossoff were presented via blogs and social media with a mix of fact and fiction, creating a sense that the fiction is also fact. Elements of the attack narrative are posted on extremist blogs and news websites, then tweeted by high-profile “news personalities” like @LouDobbs and other social influencers.

In this case, Dobbs repeated the accusation that Ossoff is funded by Soros, generating thousands of retweets and reaching tens of millions of people. Dobbs triggered the retweets by targeting his tweet directly to Trump’s political supporters using the #MAGA, #TrumpTrain and #AmericaFirst hashtags.

The attacks against Ossoff were given additional context when reported that Ossoff’s film company paid a human smuggler $2,000 in Europe. This story was subsequently shared via social media, steadily stripped of any clarifying context. Using the phrase “human smuggler” tapped into emotionally charged notions of human sex traffickers and coyotes smuggling undocumented immigrants across the border. Then someone retweeted the tweet about the story, adding an assertion that Ossoff is funded by Soros and the smear narrative grows, merging the two story arcs. Like a childhood game of telephone, the messages were merged and twisted, becoming more damning as it spread through social media.

Knowing what’s being said is crucial for figuring out how to respond.

Catching these types of stories early allows campaigns to deflect them before they get out of hand. How to respond is a question the communications team will have to answer. But discovering these stories after they develop, as they fester on the fringes, gives campaigns greater insights into the ongoing conversations happening online. It also allows them to prepare a response in case these fringe conversations develop into a story in the mainstream press.

Alan Rosenblatt, Ph.D., is director of digital research at Lake Research Partners and SVP of Digital Strategy at turner4D.