4 ways to connect your ads data to generative AI for smarter PPC

In the era of Google’s smart bidding, we’ve come to appreciate the power of feeding solid data to AI. 

But that data can do more than just improve bids. 

When connected to a large language model (LLM), it opens up entirely new ways to manage and optimize your PPC accounts.

As generative AI becomes more embedded in our daily workflows, it’s worth exploring the latest methods for connecting it with your performance data.

This way, it can deliver insights that are not just creative but also grounded and actionable.

Sure, you could manually upload a CSV or paste metrics into each prompt, but that defeats the very promise of AI: faster, smarter, and more efficient work.

Thankfully, new tools and frameworks are making it easier than ever to plug your PPC data directly into generative AI – no more disconnected reports or tedious exports.

And when AI has access to the right data, it stops guessing and starts acting like a strategist.

This article outlines four practical ways to connect your ads data to generative AI and start getting truly useful, data-driven results.

Why your own data is the secret weapon

Imagine this: You ask ChatGPT how to improve your PPC account. 

It tells you to “adjust bids,” “test creatives,” or “exclude low-performing locations.” 

Not bad – just vague.

Now imagine the same conversation, but GPT sees your actual campaigns. 

It notices that your CPA spiked last month because a new campaign started spending in Bangladesh, where your conversion rate is almost zero. 

Now the LLM gets more specific and might tell you to “exclude Bangladesh to reduce wasted spend and bring CPA back down.”

Thanks to specific data, it stops guessing and starts guiding.

Or consider seasonality. Everyone expects a Black Friday bump, but your own data might tell a very different story. 

Microsoft’s U.S. Retail Holiday Preview – June 2024 showed that Q4 shopping often starts much earlier than expected, sometimes as early as September. 

Their research found that upper-funnel Audience Network ads begin influencing conversions up to nine days after exposure.

By October, over 67% of November conversions and nearly half of December conversions are already being driven by clicks that month.

When generative AI analyzes your own historical performance data, it can uncover early buying patterns you might otherwise miss.

From there, it can recommend proactive shifts in budget, bidding, or creative so your campaigns align with how your customers actually behave, not just with calendar assumptions.

This is why the real AI advantage starts with integration, not inspiration.

Below are low-friction ways to get your data into an LLM.

1. Use Google Ads scripts to feed data to GPT

Google Ads scripts have always been a goldmine for automation. 

However, as powerful as scripts are, they only handle the scenarios the developer covered in their code. 

For example, a script explaining account performance must consider every possible combination of changes in all important KPIs.

This means the developer must consider hundreds of scenarios to frame the analysis in a written report. 

That’s where a more flexible, nuanced approach – like what LLMs offer – can make a real difference.

By feeding ads data to GPT, it can construct the narrative and do a far better job than any piece of deterministic code written by a script programmer.

I’ve set up weekly scripts that automatically pull key metrics like impressions, conversions, ROAS, and CPA across all campaigns. 

That structured data goes straight into GPT using a custom prompt. 

From there, the model doesn’t just summarize what happened. It:

• Flags problems.
• Highlights trends.
• Proposes strategic next steps.

I’ve published scripts on Search Engine Land that let advertisers automate weekly account reviews, flag anomalies, and surface underperforming keywords.

The beauty of using a script is that it’s free to install, and since you can see the code, you can modify the logic. 

Or you can ask an LLM to help you customize the code if you’re uncomfortable making the changes yourself. 

For example, do you want ad suggestions in line with your geographic target market? 

Customize the script’s prompt to tell the LLM what region you’re targeting and what language should be used for new ads and keywords. 

Want better creative suggestions? Feed in ad-level performance rather than just campaign-level data. 

However, the fact that you still need to consider what data to feed the LLM hints at a problem and our next solution for getting data to the AI.

2. Use Anthropic’s Model Context Protocol for smart data access

If you want to go beyond structured exports, Anthropic’s Model Context Protocol (MCP) is a glimpse into the AI-native future.

Think of MCP as an intelligent middleware layer. 

It lets generative AI models ask for the data they need in real time rather than waiting for static reports. 

MCP is designed to “bridge AI assistants with various data sources and tools, enabling models to retrieve and act upon real-time information beyond their static training data.”

Let’s look at that in the context of something we’ve been using for a long time: APIs. 

They allow computer systems to communicate in a standardized way. 

It’s how one website can talk to another and do things with the underlying data of another system.

And while APIs can work with generative AI, they rely on structured, deterministic programming. 

As with scripts, the developer must decide which API to call and when. 

However, the real power of generative AI lies in its flexibility. 

What if it could choose the right API on its own at the moment? 

That is exactly what the new MCP model is designed to enable.

When you ask for account optimization advice, MCP might begin by retrieving high-level campaign performance data. 

If it detects an underperforming campaign, it could request a more detailed ad report to analyze specific elements, such as headlines, targeting, or settings, that might be driving poor results.

It decides in real time what data is relevant and pulls exactly what it needs to fulfill the task at hand. 

Hugging Face describes this dynamic discovery capability as allowing “AI agents to automatically detect and utilize available MCP servers without hard-coded integrations.”

In a LinkedIn demo, Mike Rhodes showcased this concept in action. 

Using Anthropic’s Claude and a custom-built Google Ads inspector, the AI requested live performance data on demand.

It didn’t just read a report but asked for what it needed based on the conversation. 

The result? 

Claude acted like a strategist by: 

• Identifying low-performing campaigns.
• Recommending budget shifts.
• Offering optimization suggestions with zero upfront data input. 

It’s one of the clearest examples of MCP-like workflows already taking shape in PPC.

---

3. Use OpenAI’s Custom GPTs and GPT Actions

While MCP is an open protocol that any LLM could use, OpenAI uses its own variation, called GPT Actions. 

Because they are the leading generative AI provider, it makes sense to understand how those work, even if they may not be the solution that wins in the long run.

OpenAI’s Custom GPTs let anyone customize models for their own needs. 

I’ve long recommended that every agency create a custom GPT for each client – so it can respond in a way that’s tailored to that client’s unique needs and preferences.

Besides using custom instructions and static files added to its knowledge, Custom GPTs can also specify actions that let the assistant interact directly with data sources, CRMs, or anything with an API, like Google Ads. 

GPT Actions are essentially API calls wrapped in natural language instructions. 

They allow the model to trigger specific tasks or retrieve live data during a conversation.

Here’s an illustration from OpenAI of how actions work:

Since Actions let the LLM craft API calls, they can be used to: 

• Request data (e.g., give me the last 30 days’ campaign performance).
• Make changes (e.g., pause the campaign that has exceeded its target budget). 

Today, most GPTs I see only pull data. 

But soon, they’ll use the data to generate insights that lead to actions that help advertisers hit their goals.

Picture this: Your GPT sees a campaign overspending without hitting ROAS goals. 

It flags it, providing a written rationale.

If you approve, it executes the budget reduction or pauses the campaign. 

No logging into the platform. No delays.

These capabilities are already live in Custom GPTs.

They are actively being used to link workflows with campaign performance data, analytics tools, and reporting dashboards. No early access is needed. 

While some advanced features, like scheduled tasks, are still in beta, the core Actions functionality is fully available and ready to use today.

4. Use tools with built-in AI

I started my career as a programmer, so I’m excited about all three options covered above. 

But not everyone wants to build their own workflows from scratch. Frankly, they shouldn’t have to. 

Most marketers prefer working within the platforms they already know. 

While trying something new in a chatbot is cool, it’s usually not going to provide the scale and efficiency advertisers crave. 

Most people think the ideal solution will be an AI-enhanced tool, not a detour. 

That’s why AI is showing up inside tools like Google Sheets, Docs, Slack, and Notion. 

It’s faster, smoother, and less disruptive than bouncing between separate chatbots and dashboards.

The same shift is happening in PPC. 

Instead of exporting campaign data to plug into an external AI tool, some advertisers are turning to platforms that bring generative AI directly into the workflow.

Speaking about a tool I’m most familiar with, my company developed a solution designed to bridge the gap between AI and account data. 

This assistant functions similarly to an MCP, pulling any relevant data – on the fly – and using it to help advertisers with whatever questions they have.

The tool connects directly to your Google Ads account, allowing you to ask natural language questions like:

• “Which campaigns dropped in ROAS week over week?”
• “What are my top-performing headlines from the last 90 days?”
• “Where should I reallocate budget this week?”

Because it’s fully integrated into the platform, there’s no need to export data, configure APIs, or clean up spreadsheets. 

You simply ask, and the assistant provides data-backed answers in seconds, right where you’re already working.

This type of built-in AI makes the next generation of PPC tools not just smarter but truly scalable.

Here’s a roundup of the four ways generative AI can be connected with ads data to produce better results:

The next phase: Predictive PPC that plans for you

We’re now entering the era where AI doesn’t just optimize based on what happened, and it’s helping shape what happens next.

What’s emerging isn’t just automation but anticipation.

Generative AI tools are evolving into strategic engines that can:

• Forecast performance trends.
• Shift budget priorities preemptively.
• Shape creative based on predictive signals from your own data.

As Dario Amodei, CEO of Anthropic, put it in his essay “Machines of Loving Grace,” we’re heading toward a world where AI systems have “intellectual capabilities matching or exceeding that of Nobel Prize winners across most disciplines – including biology, computer science, mathematics, and engineering.” 

In PPC terms, that means tools that don’t just analyze campaign performance but also anticipate what’s likely to happen and recommend what to do next before your metrics take a hit.

It’s early, but the building blocks are already here. 

Once these systems are connected to real performance data, they’ll transform from reactive assistants to proactive strategists.

The takeaway: Your data is your differentiator

We’ve moved beyond the hype of generative AI. 

What was once a theoretical concept is now a reality, and those who succeed will be the ones who move from idea to execution. 

The true advantage in PPC today lies in how effectively your data connects with AI.

Integrating your performance data with generative AI is no longer optional – it’s essential for staying ahead. 

The methods I’ve outlined here offer a clear starting point.