Quick disclosure: I used AI to help explain the stats bits and clean up the writing here. The idea to pair these signals, the judgment calls, and the bad puns are still mine.
Being in the digital marketing space for quite a while, it’s already a given that having the right targeting is important to achieving campaign objectives. It has been proven consistently that effective audience targeting improves ROI and reduces wasted spend by focusing on the people most likely to convert.
However, the landscape has been constantly changing, and with the advent of AI, strategies have begun shifting toward the targeting innovations offered by social media platforms. AI is already reshaping marketing by analyzing behavior at scale and helping predict which audiences will engage.
That said, intent-based channels like search remain highly effective. People who turn to search engines signal clear intent, which makes them especially valuable compared to passive exposure in social feeds. Given that the average person is exposed to hundreds or maybe thousands of ads per day, I believe a strong omni-channel strategy is essential to cut through the noise and drive real impact.
The Spark
It got me thinking: if AI is reshaping marketing, why not use it to assist in building a new way to measure digital interest? That idea became the starting point for this experimental dashboard.
Of course, there are caveats. The data I used — Google Trends and Facebook Interests — isn’t absolute. It doesn’t include demographics, and it can miss nuances like language variations or regional differences, especially in the Philippines. The one-year timeframe I considered (2023) also means we’re seeing relatively short-term patterns rather than long-term cultural shifts.
Still, combining these signals is a logical approach. From experience, cross-platform data often provides deeper insight into audience behavior than relying on a single channel. While not foolproof, this experiment shows how marketers can begin piecing together a bigger picture of digital affinities using the tools already available.
How the Project Works
The idea is simple. I took the interest categories Facebook uses for ad targeting1 2 — “Beer,” “Card games,” “Coffee,” and a couple hundred more — and looked each one up on Google Trends for the Philippines across 2023. AI-assisted scripts in Colab did the fetching, cleaning, and organizing, using pytrends3 to pull the data automatically.
That gives every interest a heartbeat: 52 weekly numbers showing how much it was searched, week by week. Some heartbeats are steady, some spike around events, some quietly rise or fade across the year.
The interesting part isn’t any single heartbeat — it’s which ones beat together. When two interests rise and fall on the same weeks, there’s usually something connecting them. When one climbs while the other drops, they pull in opposite directions. And when they ignore each other entirely, they’re just strangers. That’s the whole dashboard: pick an interest, and see what moves with it — and you can narrow the time window to see how those relationships shift across the year.
The Method (a little math, no pain)
I asked AI to help me measure this with some honest math — here’s how “moving together” actually gets measured, and it’s more approachable than it sounds.
Two heartbeats are compared with a Pearson correlation4 — a single number from +1 (they rise and fall in perfect lockstep) through 0 (no relationship) down to −1 (perfect opposites).
So how strong is “strong”? Here’s the trick: square the number. That value — r2, what statisticians call the coefficient of determination5 — is the share of the week-to-week movement the two interests actually have in common:
- Strong — a correlation of 0.60 or higher. That’s about 36% of the movement shared: a third of the wiggle lining up, which is a lot for messy, real-world search data.
- Moderate — between 0.30 and 0.60 (roughly 9–36% shared).
- Inconclusive — below 0.30. Not enough in common to call it a relationship.
Those cut-offs aren’t arbitrary; they sit close to the effect-size conventions statisticians (Cohen) use for correlations, just rounded to clean numbers.6
Two honest catches are built into the tool:
Sample size. A fixed cut-off can’t tell how much data a number came from. With only a handful of weeks, even a big correlation can be a coincidence — so any window shorter than about 8 weeks gets flagged. Below that, there simply aren’t enough data points to trust the pattern.
The calendar. This is the big one for the Philippines: almost everything spikes around Christmas and dips during Holy Week. So two interests can look related purely because they both follow the national calendar. Treat strong matches as leads worth exploring, not proof of a real connection.
Reading the Map
Some pairings just click. “Beer” sits next to seafood, wine, card games, and restaurants — basically a Friday night in data form. “Advertising” lines up with newspapers, marketing, community issues, and writing — the natural neighborhood of anyone in the business.
Others are less convincing (and funny even). Under “Family,” one top match was reptiles — maybe parenting groups with pet lizards? For “Hotels,” it was rabbits and tattoos, which is likely just the calendar (or plain randomness) having a laugh — unless there’s a subculture I’m not aware of!
So What?
There’s no shortage of data today, and the big providers have built sophisticated systems for surfacing patterns and guiding decisions. This project is my way of showing that even at an experimental level — raw data points, a bit of scripting, some honest math — a curious marketer can pull out insights that matter.
The value isn’t enterprise-grade tooling. It’s being curious enough to connect the dots, and disciplined enough to know when a “connection” is really just December. That’s what makes this experiment meaningful to me — not because it’s perfect, but because it treats data as a playground for ideas.
Sources
1Vixen Digital. (2020). Facebook Ads: The Complete Interest Targeting List [PDF infographic]. https://www.vixendigital.com/wp-content/uploads/2020/06/Facebook-Ads-Complete-Interest-Targeting-List-Infographic.pdf
2Interest Explorer. Facebook Interests List. https://interestexplorer.io/facebook-interests-list/
3GeneralMills. pytrends: Pseudo API for Google Trends [Software]. GitHub. https://github.com/GeneralMills/pytrends
4Britannica. Pearson’s correlation coefficient. https://www.britannica.com/topic/Pearsons-correlation-coefficient
5Britannica. Coefficient of determination. https://www.britannica.com/science/coefficient-of-determination
6Cohen, J. (1992). A power primer. Psychological Bulletin, 112(1), 155–159. https://web.mit.edu/hackl/www/lab/turkshop/readings/cohen1992.pdf