Social Media Algorithms and Misinformation: Shaping Online Discourse

Social media algorithms not only determine the content we see online, but they push content creators and unsuspecting users to become more extreme in their views and more resistant to counter information.

In 2009, Facebook, YouTube, and Google began using an algorithm that no longer showed us the latest posts from our family and friends, or the most accurate and referenced web pages on the Internet, but posts and sites most likely to “hook” us and keep us online longer—all so they could collect more user data about us and demand more advertising dollars. The longer we engaged with this content, the more ad revenue they could generate.

As political scientist Barbara F. Walter noted, the content that draws the most eyeballs are posts that put “fear over calm, falsehood over truth, and outrage over empathy. People are far more apt to like posts that are incendiary than those that are not, creating an incentive for people to post provocative material in the hopes that it will go viral.”

Thus, the most engaging posts are filled with anger and hate—controlling emotions that we’re evolutionarily hardwired to respond to, like posts about outrageous political behavior, vitriol at immigrants, or sexist rants like those found in the manosphere. These voices don’t have to believe the hate they’re spewing. Many simply say whatever it takes to garner attention, gain a following, and generate ad revenue. Regardless, once that content is out there, and people consume it, studies show that it makes us more extreme in our views.

Just as troublesome, the algorithm keeps track of the posts we respond to, and it continues to show us similar material, filtering out anything that might contradict it and creating an echo chamber of indoctrination. So, if we think of the internet as a buffet of information, including false information, then we can think of the algorithm as a robot serving us whatever’s on that buffet. And once we’ve sampled a particular item, the robot will keep bringing us plates of that item, regardless if it’s a healthy diet of fact checking or the junk food of misinformation.

So, if you listen to a podcast that dispenses dubious info about vaccines, and the host encourages you to “do your own research,” then whatever “research” you do will merely regurgitate the misinformation promoted in the podcast. And the more misinformed people are, the less chance there is of solving our most pressing issues—because we can no longer agree on the problem.

Social media algorithms are getting gamed to weaponize reality, with massive financial consequences. A recent example I’ve had visibility into recently involved the Cracker Barrel logo change, which was accompanied by a bot-driven social media campaign that knocked off approximately $100 million in market cap — roughly 10.5% — over a few days.

The Algorithmic Mirage

The issue is that algorithms reward velocity, and engagement, but they don’t discern real consumer intent from false flags. As noted by Cyabra’s analysis, 21% of the profiles driving the Cracker Barrel boycott were outright fake — they gamed the algorithms by rallying certain hashtags like #BoycottCrackerBarrel, hashtagged the CEO with coordinated talking points, etc. At the peak of the backlash, 70% of the posts were of identical duplicated content. Because the algorithm saw this as interesting trending content (albeit artificially driven), it surfaced the outrage to many genuine users, who then engaged. While the initial bot network was inauthentic, at the end of the day, 3,268 engagements came from real people. In effect, the algorithm turned the inauthentic bot network into a mainstream media recognition event, which caused Cracker Barrel to hit pause on their rebrand.

Neutralizing Algorithmic Manipulation

For digital strategists, algorithms are treacherous, because when a lot of their engagement is algorithmically-driven, they assume all the engagement is authentic. As a result, I’ve seen the most progressive and forward-thinking teams start to incorporate bot-detection analysis into their overall crisis management strategies. The idea is that when leadership is briefed on the social media crisis, one should not only report what is being said, but importantly, *who* is saying it. By validating that the spike is being driven by accounts that have limited history and that there’s a large percentage of duplicated text in multiple posts, one can make a much more credible assessment. If it can be proven that roughly half of the outrage that’s driving your brand’s search results are actually artificial, this doesn’t totally remove the situation — but it is enough to protect leadership from needing to apologize publicly, or to avoid backing out of a multimillion-dollar strategic decision based on an algorithmic mirage.

Running a digital marketing agency for 20+ years and building WinkedIn as a women’s professional community has put me close to how information travels—and distorts—across platforms.

What I notice most is how algorithms punish patience. The business owners I work with who share nuanced, long-term thinking get far less reach than those making bold, oversimplified claims. Algorithms don’t distinguish between “this is accurate” and “this is engaging”—and that gap is where misinformation quietly takes root.

The deeper issue is that communities absorb this distortion over time. Inside WinkedIn, I’ve seen women internalize false narratives about business growth—that speed equals success, or that visible momentum means real momentum—largely because that framing dominates their feeds. Correcting it takes sustained trust-building that no algorithm will ever prioritize.

The practical response I’ve landed on: stop trying to win the algorithm’s game and start building owned spaces where your message isn’t filtered through engagement metrics. A community, an email list, a direct relationship—those channels let truth travel on its own terms.

I’m Runbo Li, Co-founder & CEO at Magic Hour.

Algorithms don’t spread misinformation. People do. Algorithms just reveal what people actually want to engage with, and it turns out humans are wired to engage with things that make them feel something, whether it’s true or not.

I spent years at Meta working on consumer social products. Here’s what I learned: the algorithm is a mirror, not a puppet master. It optimizes for engagement because engagement is the only honest signal of what people value with their time. The uncomfortable truth is that outrage, fear, and tribalism generate engagement because those emotions are deeply human. Blaming the algorithm is like blaming a thermometer for the fever.

That said, the mirror has a magnifying effect. I watched this firsthand when I was posting AI-generated videos daily and reaching over 200 million people. One video I made was clearly an AI creation, stylized and artistic. People in the comments still argued about whether it was “real.” The algorithm didn’t cause that confusion. But it did put that content in front of millions of people in hours instead of days. Speed and scale are the real variables that changed.

The actual problem isn’t algorithmic. It’s literacy. Most people consuming content online never developed the skill of evaluating sources, checking context, or distinguishing entertainment from journalism. We’ve handed billions of people a firehose of information with zero training on how to drink from it.

If you want to fix online discourse, stop trying to neuter algorithms and start investing in media literacy at scale. Teach people to be better consumers of information. The platforms that win long-term will be the ones that help users develop judgment, not the ones that try to make editorial decisions on their behalf.

Censorship dressed up as “content moderation” just pushes misinformation underground where it festers without counter-arguments. Sunlight is still the best disinfectant. The algorithm isn’t the disease. It’s the X-ray showing us how sick the patient already was.

The real power of algorithms is that they industrialize attention. A misleading post used to depend on chance, but now systems can identify the exact emotional patterns that make it spread and then keep feeding it into similar audiences. That changes misinformation from an occasional problem into a scalable behavior model. Once that happens, discourse becomes less about truth and more about which narrative architecture triggers the fastest network response.

The most actionable fix is to audit amplification pathways, not just individual posts. Platforms should examine how false claims move from niche communities into broader visibility, where recommendation loops accelerate them, and which ranking signals consistently privilege distortion over verification. We need governance that focuses on system behavior, not just content cleanup after the damage is done. When accountability sits upstream, platforms can preserve open expression while reducing the structural rewards that make misinformation so effective.

I run IT for a university of 60,000 students, and the part of this I recognise is not the ranking. It is the residue.

A field experiment published in Nature this year put about 5,000 X users on either the algorithmic or the chronological feed for seven weeks. The algorithmic feed changed who people followed. Switching them back did not change it back. The follows stayed.

That asymmetry is what most commentary misses. We debate ranking as a setting you can turn down, with media literacy as the counterweight. It behaves more like a write operation.

I see the duller version every week. A misconfiguration rolls back in an afternoon. The accounts, entitlements and dependencies it created while running do not roll back with it. ISO 27001 and our NORA architecture accreditation govern persistent state for that reason, not settings, and our audits chase what a system wrote rather than what it was set to.

Misinformation then arrives as ordinary traffic from accounts the reader chose. By then the feed has done its work upstream. The intervention worth arguing about is who is permitted to audit the graph it already built.

Misinformation spreads on social platforms because it is engaging, not because the platform designed it to spread. The algorithm optimized for engagement. That is not the same goal, but it produces the same result.

Most responses treat this as a supply problem: remove the bad content, label it, reduce its reach. The supply is infinite. Content moderation at scale is a rearguard action.

The interventions that have actually moved the needle are not removal. They are friction. Twitter’s retweet-with-comment prompt did not take anything down. It added one deliberate step before amplification. That step changed the psychological state of the user from reactive to considered. Misinformation travels fastest when people do not pause. Adding pause is not a content decision. It is a design decision.

The harder problem: nobody owns that design decision in a durable way. Ranking and amplification choices are made by teams, approved by committees, then maintained by nobody. When a false story reaches 80 million people, the question of who approved the ranking that surfaced it is almost never answerable. That is an accountability gap. Friction and accountability together — that is the combination that changes the physics of how misinformation moves.

Social media algorithms enable micro-targeting.

Years ago, misinformation was mostly the same for everyone. Now it’s different. People don’t all see the same false claims anymore. Algorithms learn what we click on and what keeps us paying attention. That information helps build detailed audience profiles.

For example, someone searching for anti-aging skincare may start seeing misleading posts that exaggerate the risks of retinol. The claim already fits something they are concerned about. That alone can make it seem more believable. Many people share it without verifying if it’s a fact.

People can end up talking about the same topic but from completely different sets of information. That is one reason online discussions become so divided. Each group thinks it has the full story because that’s all the algorithm keeps showing them.

Over time, trust starts to erode.

Agreeing on basic facts becomes harder. Correcting misinformation also takes much longer.

A boil-water notice we never issued had spread beyond our supply area before nine in the morning. Twenty-three years running communications for a regional water utility, and what algorithms do to misinformation isn’t ideological; it’s that a warning outperforms a reassurance every time. Ours is the boring post. The rumour was shared tens of thousands of times; our correction reached 1,100 accounts, and we had to pay to push even that. Give verified operators free reach to their own followers during a live incident and half of this becomes administrative rather than epistemic. The shaping effect follows from the asymmetry: whoever is alarmed sets the terms, and everybody else is answering. What I can’t settle from outside is how much is ranking and how much is ordinary sharing — the published research disagrees with itself, and the remedy depends on which.

I think algorithms change how misinformation spreads, but also change what it looks like. A claim that doesn’t get a good response in one form will be changed, made shorter, made into a picture, turned into a personal story, or have the contested part of it taken away. The authors are learning from their audience and changing the message over and over again until they find the version that people pay attention to. In that way, the algorithm is not just distributing misinformation but also encouraging misinformation to be edited for maximum effect.

This matters because the claim that ends up spreading may be wildly different from the original. It has already survived a series of small public experiments showing which language is met with attention and which facts are met with resistance. By the time most people see it, they are seeing a version that has been optimized for the response it gets. I think this makes it harder to address because often people fact check the factual claim, while in social media it has already been proven which form of that claim is the most effective.

Operationally speaking, social media algorithms also mean misinformation can become a customer-service problem long before a business knows it’s happening. Take claims and accident management. It’s not uncommon for customers to turn up on a call already armed with information they’ve seen repeated across social media, whether that’s about their rights, what compensation they can expect or what the insurer is legally required to pay. The customer thinks they know what to expect and demands an instant response based on what they’ve read online, rather than the reality of their situation. This puts further pressure on already stretched frontline teams to not only correct the misinformation but also to do so without losing the customers’ trust. Add in the power of social media algorithms – where posts that invoke a stronger emotional response (positive or negative) are distributed more widely – and those misconceptions can be shared thousands of times before a business’ more measured explanation catches up.

From an operational point of view, the key learning for me is that you cannot approach misinformation as purely a marketing problem. Frontline teams should be aware of common misconceptions appearing online so they can address them confidently when questioned. I’d take customer-service teams’ recurring questions and objections and feed them back into our communications and content planning, tackling those questions head-on with clear, straightforward information before the misinformation spreads.

Algorithms do not need to prove something is true. They just need to make it seem familiar. When something seems familiar, people stop questioning it. As a coach, I see how social media can affect people. When people see the same thing over and over, it starts to feel normal. If people they trust post something, it feels even more true. The algorithm is not making people believe something, but it shows them the same information many times. This matters because people start to care more about what their friends think than about what is true.

The problem with social media is not just that false information spreads fast. It is also that when people see something many times, they think it must be true because their friends believe it. Social media companies can help by giving people more control over what they see. They can also make it harder for false information to spread. The bigger problem is that people need to learn that just because something seems familiar, it does not mean it is true.