How to Zap in 5 mins⚡

Hello marketers. Welcome to AI Marketing School, where we dish out the latest and greatest in AI-powered marketing.

In this week’s issue:

1. AI Marketing Update: Meta joins the AI search movement.

2. The Stack: Building useful automations in minutes.

3. Consultant’s Corner: The rise of modular AI.

Onwards!

AI MARKETING UPDATE

Meta Wades Into AI Search

The "Google killer" conversation is heating up.

OpenAI’s ChatGPT Search has sparked plenty of debate, offering a glimpse into what AI-driven search could be — but also exposing its weaknesses.

Now people have had a chance to use it, ChatGPT Search seems to struggle with the short, navigational queries that dominate Google’s search traffic.

For example, if you want to find the score for a sports game, it basically doesn’t work, as per TechCrunch’s tests.

That does beg the question, why would you use it for that anyway?

In my tests, it’s very effective when there’s sufficient information for it to find — like news which has been written by a few different outlets.

One key limitation to point out is that ChatGPT Search tends to repeat itself.

It’ll return some stats, you’ll ask for more, and it’ll return the same list. You often have to open a new chat, modify the prompt, and try again to get more stats.

Anyway, things in AI search is hotting up again, as Meta has quietly entered the race, not just building a search engine but positioning itself to wield AI across generative models, social media platforms, and search indexing.

Could this trifecta — AI search, social integration, and generative AI — the three rings to rule them all — create an all-powerful knowledge system?

Meta’s Vision: An Ecosystem on Lock

With Meta AI already embedded in WhatsApp, Instagram, and Facebook, the company wants to make the search experience native to its platforms.

Recent moves like its partnership with Reuters for real-time news answers signal Meta’s commitment to making its AI a self-sufficient.

Soon you might be able to search the web in real-time, explore social media in real-time, and use gen AI capabilities, from one interface — something no company is yet to accomplish.

Meta AI itself is a fairly recent release (it still isn’t out in the EU yet), and I’d wager many haven’t tried it.

It’s great for returning info directly from Instagram and Facebook, which you can use to track trends, locate viral content, find influencers, and much more.

Unlike OpenAI, which leans on Bing for web access, Meta is now building its own search index, which will grant it more flexibility.

This will further risk Google’s dominance and build a ‘walled garden’ around Meta’s products which could be a masterstroke in the long-term.

For example, imagine searching for “best restaurants nearby” in WhatsApp and getting immediate recommendations, complete with links to book or order within the app.

If this catches on, businesses and publishers will find it harder to drive traffic outside these platforms.

For marketers, it raises crucial questions:

• How do you optimize for visibility within these closed systems?

• Will these platforms prioritize their own content over independent creators?

Meta’s track record suggests the latter isn’t far-fetched.

The takeaway? We really need to keep our eyes on the ball for how to optimize content for this new form of gen AI search.

THE STACK

Buid Useful Zapier Automations in 5 Minutes

Zapier automations are easier to build than ever thanks to their AI copliot, which works extraordinarily well. If you’ve never tried Zapier, now is the time.

This short tutorial will show you how to build a simple automation that uses Zapier, Gmail, ChatGPT, and Slack to create an automated content pipeline.

The goal of this automation is to take email content, turn it into content automatically using AI, and send it to your Slack via personal message.

In this example, we’ll take an email newsletter (which I generated and sent to myself from my own email), feed it into ChatGPT, add a custom prompt to turn it into a funny social media post, then send it to Slack via personal message.

You can implement this with any email, e.g. newsletters you follow, company updates, email content you send yourself, and Google Alerts.

First, you’re going to need:

• Zapier account (free to build the Zap and test it, but unfortunately you have to pay to Publish it).

• An OpenAI account — you’ll need an API key and to top up your account with a small amount of credits.

Let’s go:

Step 1: Create a “Zap”

You’ll want to head into Zapier to create an automation, named a Zap. You’ll be presented with a blank canvas:

Now, in the Copilot section, this is the prompt I used to create the automation framework. It’s also easy to build manually.

It’ll then ask whether you want to auto-map the steps to the canvas, as so:

If everything works when you add the steps to the canvas, the first step should look like this:

You’ll need to connect the Gmail account you want to use. Then, in the Configure tap, you’ll add the email address you want to target in the Search keywords.

You can add any keywords that locates the emails you want to repurpose content from, e.g. “Google Alerts.”

Go ahead and test it. It should pull records matching your criteria:

Step 2: Connect ChatGPT

In the next step, you’ll connect your OpenAI account and add in your API key, which is super-easy. It’s probably best to use GPT-4o.

In Setup, make sure “Conversation” is the Action step, which should have been filled out by Copilot.

In the Configure window, you’ll need to use the “+” to map the Body of the email to the field. Before the Body Plain field, you’ll want to insert your prompt.

It should look like this:

This essentially adds the prompt to the beginning of the email body.

You can use whatever prompt you want.

You could write up Temu promo email into a philosophy essay, analyze a business’s newsletters for content and tone, take a newsletter like this and turn it into social media posts or something else — use your imagination (within fair use/copyright obviously!)

Go ahead and test the step.

Step 3: Connect the Slack Output

This step is pretty easy. In Step 3, you just connect your Slack account and send your ChatGPT output as Direct Message.

In the Configure tab, use the field mapping to map the “Reply.” See below:

The Results

Here’s the result. Whenever you receive an email that matches your criteria, it’ll output whatever you set ChatGPT to produce into Slack.

This has tons of potential uses.

You can create a dedicated email address to essentially collect information to route through your content machines. The downside is that you have to pay — Zapier is fairly pricey, ChatGPT credits are cheap for this use case.

But the time-saving potential of this is huge, and I’ve almost definitely not thought of all the uses!

Maybe some of you can think of some cool stuff?

CONSULTANT’S CORNER

2025: Will Small Models Dominate?

Companies seem to be moving beyond massive, one-size-fits-all models like GPT-4.

Instead, we’re observing a shift toward modular AI — small, specialized models that are integrated into flexible, full-stack solutions — or small language models (SLMs).

Microsoft, Salesforce, and Writer are leading this movement of late, offering targeted small language models (SLMs) that can handle specific tasks with greater efficiency.

As Sébastien Bubeck, Microsoft’s VP of generative AI, recently put it, the new direction is about focusing on “the basic ingredients” necessary for each task.

Writer’s Palmyra models illustrate this concept well, combining accuracy and domain-specific knowledge in a stackable format.

For example, Palmyra-Med excels in analyzing medical data and research, while Palmyra-Fin is tailored for financial reports and regulatory documents.

Palmyra X enhances structured text generation, making it suitable for legal and compliance tasks, while Palmyra Vision integrates NLP with computer vision for tasks like image captioning and visual data analysis.

Writer’s AI Studio can also integrate with systems like your CRM, ad platforms, or analytics tools to do things like:

• Adjust ad spend based on live performance data.

• Generate a full campaign — from posts to emails — starting with just one piece of input.

• Pull insights from your datasets, whether it’s customer sentiment or sales forecasts.

You can try it for free, it’s really impressive stuff and will prove useful for marketing agencies.

Writer RAG tool: build production-ready RAG apps in minutes

• Writer RAG Tool: build production-ready RAG apps in minutes with simple API calls.

• Knowledge Graph integration for intelligent data retrieval and AI-powered interactions.

• Streamlined full-stack platform eliminates complex setups for scalable, accurate AI workflows.

Learn more about our production ready RAG tooling here.

Rethinking Marketing with Modular AI

What if you could design AI tools that don’t just work for your clients, but seem to work with their voice, goals, and unique challenges?

Modular AI makes that possible.

For example, imagine spinning up a bespoke AI model for each client that truly embodies their brand.

Instead of fighting to coax a general-purpose model into generating on-brand content, you’d have an AI that inherently understands the subtleties of their tone, industry terminology, and so on.

Models Built for Campaigns, Not Just Clients

Now let’s take it further: what if we didn’t just customize models for clients, but for specific campaigns?

One model could generate compelling ad copy for a high-profile product launch, while another focuses solely on extracting actionable insights from sentiment analysis on social media.

For an agency, modularity means running multiple, highly focused models in parallel — a product-description model for e-commerce clients alongside a social-listening model for real-time trend tracking.

Campaigns wouldn’t just feel tailored; they’d be tailored, at every step.

You can already do this with Custom GPTs or projects in Claude, but deeper customization adds more control, sovereignty, and data protection.

If you pitch to creating a company or client some marketing chatbots tailored to them in Claude or ChatGPT — you can see how that conversation might go.

Questions like, “Will our data be trained on?” or “Can you guarantee this data won’t be used elsewhere?” are almost inevitable.

Even if the client understands the process, the idea of sensitive customer information or proprietary strategies being fed into a general-purpose model can feel like a risk they’re not willing to take.

This is currently what’s holding back enterprise gen AI.

SLMs and other more customizable, sovereign models are set to change turn the tables.

Hope you enjoyed this week’s issue. If you missed it last week, you can read it here.

If you found it useful, please recommend it to a friend or colleague.

Until next time. Happy marketing.

—The AI Marketer