The Unexpected Findings
These were the emoticons staring back at me at the end after NotebookLM AI completed the analysis: 📉 💸 😡
I was drowning in this sea of negativity. Was this just the nature of Reddit, an outlet for user to vent?
While that’s a possibility, it’s not the full picture.
My decade long experience in PPC taught me that ad platform has ALWAYS had its own unique challenges and repeated complaints could reveal patterns that, when analyzed properly, offer deep insights on what they are and how to solve them.
The Approach
To dig deeper into the data and uncover actionable insights, I turn to the powerful capabilities of LLMs AI such as Claude, ChatGPT and NotebookLM.
These tools now offer both larger context to handle the large and better ingestion capabilities (Search, Urls or Uploading data)
However it can still be tedious as the large volume of data makes it challenging to upload (Over 302 posts were considered) and hard to filter out only valuable posts and comments from the memes and jokes, which is crucial to avoid misleading insights.
Instead I took the following approach:
- Used PRAW to politely get posts with all variations of platform name (Fb Ads, Meta Ads..etc)
- Excluded irrelevant posts and comments through a process I've developed
- Flattened the output while still respecting the threaded nature of comments
🌟 At the end of this post I've also provided the links to NotebookLM so you could interact with the above curated dataset for more juicy insights 🌟
The Insights & Solutions
Common Issues across Facebook, Google and Linkedin Ads
Bot Traffic
Across all platforms was quite a bit of concern with spam leads, click fraud and bot traffic.
This not only directly impacts ROAs with fake conversions but can cause each platforms Algorithm to be get trained on the wrong signals as these would be counted as "Conversions" on the platform
Solutions - Network Exclusions
A quick solution is to reduce exposure to extended networks. Platforms typically have additional network we could opt out of:
- Meta's Audience Network placements
- LinkedIn audience network
- Google's display and search partner networks
- Google's Mobile app category
These actions were repeatedly advocated across many threads and have noted to reduce spam leads.
Some users even opted to stop targeting specific countries where cases are severe.
Key actions includes regularly checking placements for where ads are displayed to remove spammy sites. Finally some have opted to engage click monitoring / specialist bot detection companies.
Facebook Ads
An Unknown Killer Lurks!
Something mysterious is brewing for Facebook ads since Late Jan to Feb 2025,
Multiple Reddit posts were spawn about inconsistent performance around the same time.
There are speculations about data restrictions and algorithm changes - however nothing conclusive.
Roller Coaster Meta Ads Performance
Meta Ads - and all went dark again
FB Ads huge drop
WORST Month for Meta Ads - JANUARY 2025
Anyone Else Seeing a Major Drop in Meta Ad Performance?
Reported Solution - Creative Refresh
Multiple users have reported that a consistent creative refresh have helped them keep afloat
Linkedin Ads
High CPCs and Costs
The most common comment was that linkedin Ads is expensive - some quoting 4X vs Google Search CPCs for the same niches and $20+ CPCs for some audience is not unheard of.
This also reflects the premium nature of the platform where we could easily reach professionals.
Solutions - Manual Bidding
While manual bidding is a quick solution where quote but there was an entire thread dedicated to
LinkedIn Ads not respecting the manual bid setting hence monitoring is key!
Solutions - Be Targeted
To ensure the most is made out of this higher CPC enviroment. We could serve ads to only highly qualified audience through Match Audience. We could upload a list of contacts and targeting could be based on either high-quality contact lists or even leads gathered from offline events. To expand on this audience, lookalike could also be considered.
Solution - Be Nurturing
A multi layered approach also works well - where a "cold layer" is run where cheaper cpc are (traffic ads, video view ads, influencer ads, single image ads, carousel ads)
and a retargeting campaign is then used to for converting via lead gen form ads with strong value offering like eBooks, Webinars..etc.
And lastly to ensure these expensive leads are well followed up with a email nurture sequence.
Solution - Balance Value with Cost
Also given the known high costs on Linkedin - we could also be strategic about which higher value products to promote on Linkedin and which to avoid. Ensuring only high ticket items are promoted is key to a healthy ROI.
Google Ads
Match type creep
Google has really altered keyword matching. Even trusty exact match keywords can trigger ads for loosely related searches resulting in bad matches. This has quite a few comment grieving about this.
Google seems to be moving from a keyword based model to an "intent" based model for Search Themes and this has quite a few negative response.
Solution - Negative Keywords
Deeper lists of negative keywords could help, however as of 2025 due to privacy concerns much of the search term returns are listed as unknown. Hence supplementing this process with keyword research is key.
Geotargeting Default
The default setting, "People who show interest in your targeted locations," may lead to ads being shown to users outside the intended geographic area. For example, someone in Singapore watching an American TV show could be tagged as "interested in America" and see ads targeted to the US
Solution - Correct Choice
"Presence In" for some cases makes more sense - eg, if your campaign is for a service like car-washing service. No matter how interested someone is the place you are located, they are unlikely to travel cross states or country.
Usefulness of the Research Approach
The results yielded tallies with my decade long experience in this space and have reveal quite a few common effective levers I've use.
It has also returned really unexpected results on Facebook - as usually I'm first to self-reflect and think it is some stone I've left uncovered if performance is not going well instead of thinking it could be a platform issue.
What's more exciting is we could extend this process of getting reddit comments to gain insights on many other things
- Learn how users compare between products/brand (eg, only retrieve comparison posts)
- Social listening for trends in comments ( segment analysis through time)
- Finding pain points of users for certain products
- Invert the filter I used for quality posts to find ....memes and jokes only
..... The list goes on!
Bonus - More Insights you could think of?
Interact with the curated dataset and find additional insights with these links to the notebooklm!
😡Google Ads: Experts Discuss Strategy, Frustrations, and the Future