When I started Impressions to Ballots one year ago, my goals were to:
Write for 52 weeks about topics that move the needle (pun intended) for political advertisers
Share strategies, data, and insights that would make political advertisers more efficient
Give campaigns information to help determine if their vendors were spending their money wisely
Thank you to those who have been along from the beginning and those who signed up throughout the year.
You can always read past editions at: impressionstoballots.com
We aren’t done. Stay tuned this week for announcements about where this newsletter is headed after Election Day.
What were the most common takeaways?
Measure deduplicated reach and frequency across screens, and use it to move money. 23 issues.
Plan from your audience’s real viewing habits. There is no average voter. 21 issues.
Use every channel together: texts, mail, doors, and relational organizing. 12 issues.
Break down silos and let budget move between TV and streaming. 11 issues.
Outspending the other side doesn’t always win. 10 issues
What have we done differently this cycle?
YouTube - We spent a higher share of budget on YouTube than any prior cycle.
Its user base has grown across all demographics.
Voters are watching a larger share of YouTube on TV.
Targeted Premium Streaming - A higher share of our streaming spend went to targeted audiences on premium streaming apps.
Overall streaming viewership has grown.
Subscriptions to ad-supported tiers have grown, increasing available inventory.
More premium streaming companies offer targeting capabilities than in previous cycles.
Where did we change our mind?
Frequency.
Back in Impressions to Ballots #4, we wrote about wasted impressions: voters seeing the same ads too often based on how much linear and streaming video they watched.
This cycle changed how we define “too often.”
Test after test showed that a message needed 20, 30, or even 40+ exposures to break through, far more than we expected.
Most campaigns don’t keep ads running long enough or optimize delivery well enough to reach those levels. They’re moving on from a message before voters have heard it enough for it to stick.
What were our most data-heavy editions?
50 company finances analyzed
150 distinct financial documents
500+ Claude Cowork tool calls
17.5M Claude Cowork tokens
2,181 creatives analyzed for September only
8,962 transcripts analyzed (counting duplicates and differing creative lengths)
540,182 words read
41 million Claude Code tokens used
How did I use AI to write this newsletter?
According to Pangram, 86% of the 30,000 words in Impressions to Ballots are human-written.
My goal was to write a newsletter that provided value to campaigns and political advertisers, not to create AI slop no one has time to read.
I wrote or dictated every newsletter (thank you, Wispr Flow).
Here’s how I used AI:
Grammar and word choice (the 14% Pangram labeled as AI-assisted).
Data visualization.
Data analysis (see above).
Thought partner for future topics.
Not AI: Huge shoutout to Steven Fitzwilliam for editing every edition, making everything look good, and being very honest when something sucked.
What does this mean for political advertisers?
Take a break. This is a tough industry, and everyone is tired today. Celebrate or mourn, take a breath, and get ready for the next one.
Learn. Campaigns will wind down, but there will be a lot to learn for future cycles. Analyze which strategies worked and which didn’t.
Help someone out. Many hard-working, high-quality staffers are out of a job today. Take a meeting, keep an eye out for opportunities, and open your network to those who've just proven themselves.
My door is open if anyone wants to chat.
Thank you,
Chauncey Southworth