2025: The year AI hype met reality 🥂

Before we pop the champagne and head into the new year, it’s time to look back at a year that changed everything for us as marketers. In this final issue of 2025, we’re moving past the "speculative excitement" of AI and diving into the cold, hard data. From the FTC’s major crackdowns to the rise of "Answer Engine Optimization," I’m recapping the five stories that defined AI marketing this year. Plus, I’m sharing the final results of my n8n agent build—and a confession about why I might be scrapping the whole thing already. Let’s close out the year with a dose of reality.

Here are the top five marketing-related AI stories of 2025, cited from non-biased news organizations, academic journals, and regulatory bodies.

In late 2025, the Federal Trade Commission (FTC) took aggressive action against deceptive AI-driven marketing. This included suing companies like Air AI for making misleading claims about "AI-powered business growth" and issuing warning letters to firms regarding AI-generated consumer reviews. Simultaneously, the Trump Administration’s late-year Executive Order sparked a massive debate by aiming to preempt state-level AI regulations to favor rapid development.

The marketing world faced a "traffic cliff" as AI Overviews in search results began reducing traditional click-through rates (CTR) by up to 30–40%. According to studies from Semrush and Amsive, marketers had to pivot from SEO (Search Engine Optimization) to AEO, focusing on getting their brand cited within the AI's summary rather than just ranking as a link.

A major study by iHeartMedia and Critical Mass Media in 2025 revealed a growing "algorithmic fatigue." Nine out of ten consumers reported that it is important to know if the media they consume was created by a real person. This led to a "Cheers and Jeers" culture in marketing, where brands like The Onion were praised for banning AI content, while others faced backlash for "AI-slop" in their feeds.

2025 marked the transition from "Assistant AI" to "Agentic AI." Reports from Forbes and Gartner highlighted that 73% of marketing departments now use generative AI, but the leaders are those deploying autonomous agents. These agents move beyond drafting copy to independently managing lead research and cross-channel budget adjustments.

According to Deloitte’s 2025 Connected Consumer study, trust has become a "product feature." As brands integrated hyper-personalization, they hit a wall with consumer privacy concerns. Academic research published in the Journal of Advertising highlighted that "algorithmic mediation" often erodes consumer autonomy, leading to a new marketing requirement: transparent AI governance.

Final Project Wrap-Up: AI Agent Workflow Complete

This project build is officially finished! In the last newsletter, I successfully executed the agentic workflow I built in n8n. The goal—to receive an email summary of the week's top AI marketing news—was achieved. However, the final output required a slight refinement: despite my initial filter settings, the summary still included some stories that weren't specific enough to AI marketing.

The good news is the fix was simple.

Refining the Workflow for Precision

My solution involved updating the "message a model" node. Previously, this node was only used to format the selected stories into an itemized list. I modified the prompt to give it a second, crucial task: re-assessing the relevance of each story for my specific audience (marketing professionals interested in AI).

I instructed the model to label each item as "Relevant" or "Irrelevant." I then added a subsequent node to automatically filter out everything labeled "Irrelevant," ensuring the final email message contained only highly relevant items.

And it worked! The agent successfully pared down a list of 225 stories to a highly focused few based on the new, stricter parameters. I have officially built a functional AI agent!-----See below for a full recap of the build process!

I wanted a system that could scan a ton of AI + marketing content for me, filter out the noise, and hand me newsletter-ready insights every other week. This is the exact workflow I built—and what each node does.

What it does: Starts everything automatically
When: Every 2 weeks on Sunday at 9:00 AM

I didn’t want this to rely on memory or discipline. This trigger kicks off the entire workflow on a fixed schedule, whether I’m busy, traveling, or deep in another project.

What it does: Stores my list of 9 RSS feed URLs

This node is basically my source control. All of the publications I trust live here, which makes it easy to swap sources in or out without touching the rest of the system.

Think of it as: My curated reading list

What it does: Separates the 9 RSS URLs into individual items

The next steps need to evaluate each feed one at a time. This node takes the list and breaks it into single URLs so they can be processed properly.

Analogy: Dealing cards from a deck—one URL per card

What it does: Pulls articles from each RSS feed

This is where content actually enters the system. For every feed, this node grabs:

At this point, I’m sitting on hundreds of raw articles.

What it does: First-pass filter using keywords

I only keep articles that mention both “AI” and “marketing.” This isn’t meant to be smart—just fast.

Why I added it:
AI evaluation costs money. This step filters out obviously irrelevant content before it ever hits a model.

Think of it as: A bouncer checking basic requirements at the door

What it does: The intelligent filter

This is the brain of the system. For each article, the model decides:

If yes, it:

If no, it returns “NOT RELEVANT.”

This is where context, judgment, and writing happen.

What it does: Deletes anything marked “NOT RELEVANT”

Once the AI has weighed in, this node throws away everything that didn’t make the cut.

Result: Only high-signal, newsletter-ready entries remain.

What it does: Combines all entries into one list

Each article starts as its own item. This node bundles them together so they can be sent as a single output.

Analogy: Putting all your groceries into one bag instead of carrying them one by one

What it does: Emails me the final output

I get an email with:

From there, I review, pick my favorites, and publish.

Input: 9 RSS feeds with hundreds of articles (I can always add more RSS feeds to have more options for the agent to choose from.
Process:
Keyword filter → AI judgment → Formatting
Output:
An email with 5–15 curated, newsletter-ready insights from articles about AI in marketing

What this really gives me:
More time thinking and writing—and way less time scanning tabs.

You read that right—I might be abandoning this agent. But not because it failed to meet expectations or doesn’t work. In full transparency, while developing this project, I simultaneously created an alternative that accomplishes the same tasks in half the time. It is still an agentic AI workflow, but I approached it with a more intuitive, "vibe-coding" methodology. What else was I supposed to do during a five-hour flight delay at the airport? In the next newsletter, I will detail exactly what I built, the process behind it, and the specific tool I utilized.

As always, stay curious and have fun!

Best,
Skyler Neal