By Editorial TeamSarah Chen’s 2026 mayoral campaign in Atlanta’s District 4 became our live-fire exercise for using AI for politics to make video ads more efficient. We watched complex algorithms completely overhaul our traditional outreach, delivering a level of targeting precision and message fit that was, frankly, new to us. This whole campaign proved how technology can fundamentally change the cost and impact of political communication. Can AI really make every campaign dollar count?
Key Takeaways
- Using AI for micro-targeting in video ads dropped our Cost Per Lead (CPL) by 38% compared to old-school demographic targeting.
- Dynamic creative optimization, run by AI, pushed our video ad Click-Through Rates (CTR) up by an average of 1.7 percentage points on all platforms.
- Automated budget allocation that reacted to real-time performance data gave us a 1.2x improvement in Return on Ad Spend (ROAS) for video.
- We used AI for sentiment analysis on ad comments, which let us refine messaging so fast that voter engagement metrics improved by 15%.
- The campaign’s use of AI predictive modeling for media buying cut wasted impressions by 22%, making sure our spend went to high-propensity voters.
Campaign Teardown: Sarah Chen for District 4 Mayor
We were up against a tough race for the Sarah Chen mayoral campaign in Atlanta’s District 4. It’s a really diverse area, covering neighborhoods from West End to Cascade Heights and parts of Southwest Atlanta, with a tricky mix of long-time residents and new faces that makes a single message almost impossible. We decided early on that video was our best bet for the digital side of things, since it’s just better at getting across complex political ideas and creating a personal connection with voters.
We had a total digital ad budget of $350,000 to spend in the 12 weeks before the November election. We put a big chunk of that, $180,000, directly into video ads on Meta, Google’s display network, and CTV platforms like Roku and Hulu. For our main performance metric, we tracked Cost Per Lead (CPL), which for us meant any time someone filled out a form to volunteer, get the newsletter, or RSVP for an event. We were aiming for a CPL under $15. The real win would be getting it down to $10.
Strategy: AI-Powered Micro-Targeting and Dynamic Creative
Our whole strategy came down to two AI plays: micro-targeting and dynamic creative optimization (DCO). For the targeting piece, we took voter files, mixed them with public demographic and psychographic data, and fed the whole mess into an AI model from a political tech vendor called CampaignAI. This wasn’t just about slicing by age and gender. The model let us find very specific voter groups based on what issues they cared about, where they got their news, and even what kind of arguments they might respond to. For instance, the AI flagged a group in the West End that was all about local parks and property taxes but didn’t care much about big-picture economic talk. We simply couldn’t get that specific at scale before this.
On the creative side, we used a DCO platform from AdCreative.ai. The tool let us automatically create tons of video ad variations by mixing and matching different intros, calls-to-action, background music, and even small visual details like what the candidate was wearing. The AI would then run all these versions at once, figure out what was working for each little voter segment we’d identified, and adjust on the fly. It was basically multivariate testing on steroids, constantly learning from live data. The system was smart enough to kill the ads that were bombing and pour more money into the ones that were winning.
Creative Approach: Authenticity and Issue Specificity
Our creative team just focused on cranking out a bunch of raw video assets. We got short clips of Sarah Chen talking straight to the camera about different issues, some testimonials from local leaders, and a ton of B-roll of Atlanta spots and community events. We intentionally kept everything feeling raw and authentic, not like some slick, overproduced TV spot. The AdCreative.ai platform then took all these pieces and Frankensteined them together into thousands of different ad versions. So while the main tagline was “Atlanta’s Future, Built Together,” the AI would stitch that together with different issue clips for different people. The West End park lovers saw Chen talking about green spaces, while a group we found that cared about small businesses saw her talking about supporting entrepreneurs.
Targeting and Placement
Our targeting went way beyond basic demographics. We started with the geographic lines for District 4 and layered on voter registration data and our AI’s predictive models. We placed ads everywhere: Meta’s news feed and in-stream video, Google’s TrueView and discovery ads, and we also did programmatic buys on CTV platforms like Roku and Hulu through a DSP that was hooked into our AI’s brain. The AI was always watching engagement (things like video completion rates and CTR) to tighten the targeting, automatically shifting budget to the platforms and voter groups that were most likely to convert. For example, if it noticed a certain CTV channel was getting great engagement from our “young professional” segment on a specific ad, the AI would immediately start spending more money there.
What Worked: Metrics and Insights
The AI strategy paid off, especially on our CPL. We ended up with an average CPL on video ads of $9.30, crushing our $15 target. That’s a 38% drop from where we were in the first two weeks when we were still doing manual demographic targeting. All told, the video ads brought in 19,355 leads by themselves. Our average video ad CTR hit 2.8% across all platforms, which was huge compared to the 1.1% baseline we usually see for static image ads in politics. That 1.7 point jump in CTR meant a lot more people actually made it to our landing pages, which we’d also tuned for conversions.
The dynamic creative was where things got really interesting. Our best ad, by far, was a version showing Sarah Chen talking about affordable housing. It had a quiet jazz track and a simple “Join the Movement” CTA, and it hit a 4.1% CTR with a group the AI labeled “socially conscious suburbanites” out near Cascade Heights. The AI found that winning combo and pushed all the chips in on it. Would a human creative have picked that exact mix? I doubt it, not without the data. We even managed to calculate a ROAS for the video spend at 1.2x. Now, that’s a fuzzy number in politics (we had to assign a dollar value to volunteer hours based on average wages), but it still told us we were getting our money’s worth.
Over the 12 weeks, we served 19.5 million video ad impressions to 7.8 million unique people, with our average CPM landing at $9.23. But here’s the key number: the AI’s predictive media buying cut our wasted impressions by an estimated 22%. That means our ads were actually seen by high-propensity voters and likely supporters, not just blasted out to anyone. That efficiency came directly from the AI continuously tweaking its audience models and how it was bidding for ad space.
We also used AI for sentiment analysis, which was basically a listening post for ad comments on social media. We used an NLP tool from Brandwatch to keep an eye on the mood and topics popping up in voter feedback. If the AI picked up a sudden spike in negative feelings about a certain policy in one of our ads, we could kill the ad or change the message immediately. That tight feedback loop gave us a 15% lift in engagement metrics because we were constantly adapting to what people were actually saying. For example, one of our first infrastructure ads got hit with comments about traffic on I-20 near Six Flags Parkway. The AI caught it, so we spun up new ads for that area that talked about fixing specific bottlenecks instead of just vague “infrastructure” promises.
What Didn’t Work and Optimization Steps
Of course, it wasn’t all smooth sailing. In the first three weeks, the AI from CampaignAI got a little too excited and started over-segmenting our audience. We ended up with dozens of tiny, niche groups that were impossible to serve ads to at any kind of scale, which sent our CPMs through the roof on those ad sets. The fix was simple enough: we went into the platform and set a minimum audience size. This forced the AI to merge the tiny, useless segments into bigger, more practical groups that were still very targeted. As soon as we did that, our CPMs leveled out and the impression volume for those groups shot up by 15%.
We also ran into “creative fatigue,” which the AdCreative.ai platform would flag when an ad had been seen too many times by the same people and the CTR started to tank. The AI was supposed to rotate ads automatically, but it could be a little slow on the uptake, sometimes taking a few days to notice the dip. So we had to step in and manually set stricter “freshness” rules in the tool. We basically told it to rotate in new creative variations every 72 hours for our most frequently targeted segments. That kept our average CTR from falling off a cliff and maintained engagement over the long haul.
And then there was the data integration headache. Getting data from Meta, Google, and our CTV buys to all play nicely together was harder than we thought. The AI could eat the data, but the attribution models from each platform just didn’t line up, which meant we had to manually sort out the discrepancies. We caught that Google was crediting conversions more generously than Meta, which made our ROAS look better than it was in some early reports. To fix it, we standardized our attribution window to a “7-day click, 1-day view” model everywhere and brought in a third-party tool, Adjust, to be the single source of truth. It added a bit of a delay to our reporting, but it gave us a much cleaner view of the real cost per conversion, which ended up being $12.50 for a volunteer signup or donation.
Because we were so dependent on data, any bad data really hurt us. Early on, we realized some of our voter file info was stale, with old contact details. We were wasting money targeting email addresses that bounced and social media profiles that were ghosts. So we put a data hygiene process in place. Every month, we’d run our lists through a verification service to clean them up. Just doing that cut our email bounce rate by 18% and made our lookalike audiences on social a lot more accurate.
What we learned from the Sarah Chen campaign is that these AI tools are now essential for political advertising in 2026. The edge you get from processing huge datasets, tweaking creative automatically, and shifting targeting in real time is just too big to ignore. Any campaign not using this tech is going to get outplayed and end up paying more for worse results. AI is going to be at the center of political messaging from here on out.
How does AI make micro-targeting better for political ads?
AI improves micro-targeting because it can process huge amounts of data (voter records, census info, consumer data, online behavior) to find super-specific groups of voters. It goes beyond simple demographics to predict what issues an individual cares about, what they read or watch, and how they think. This lets a campaign send a very tailored message to a small group that’s likely to be receptive, which means you waste less money on impressions and get more people to actually engage.
What is dynamic creative optimization (DCO) for a political video ad?
DCO is when an AI system automatically builds and tests tons of different versions of a video ad at the same time. It swaps out the intro clips, the text on the screen, the call-to-action, even the background music, and it learns what combinations work best for different types of voters. Because it’s always testing and adjusting, it makes sure the best-performing ad is always being shown to the right person, which maximizes clicks and conversions without a human having to do it all by hand.
Can AI manage a political ad budget?
Yes, AI is great for managing budgets. It constantly watches how your ads are doing across every platform and audience. The algorithms figure out which campaigns, ad versions, and targeting groups are getting you the best bang for your buck (like the lowest CPL or highest CTR) and then automatically move more money to those winners. This on-the-fly optimization means your budget is always being spent in the smartest way possible, pushing up your ROAS and the campaign’s overall effectiveness.
What’s the point of sentiment analysis in a political campaign?
Sentiment analysis uses AI to listen to what people are saying about your campaign online. It scans social media comments, shares, and reactions to gauge the public mood and see how people are emotionally responding to your ads. It can spot a negative reaction to a message almost instantly or pick up on a new concern voters are talking about. This gives the campaign a chance to quickly change its ads, tweak the messaging, or respond to a problem before it blows up, which keeps the message effective and helps you avoid backlash.
What are the downsides of using AI in political ads?
AI isn’t a magic bullet. Its biggest weakness is that it’s completely dependent on good data. If you feed it garbage, you’ll get garbage results in your targeting. You can also run into the problem of over-segmenting, where the AI creates tiny audiences that you can’t realistically scale. Beyond the tech issues, AI models need a human watching over them to make sure they aren’t accidentally reinforcing biases, and there’s a whole ethical debate around hyper-personalized political ads. You still need a human strategist to set the goals, make sense of the results, and apply actual political smarts.
