In the proverbial home of paid media, generative AI hasn’t knocked politely - it’s kicked the door down, barged in, and is now bashing pots and pans around in the kitchen, demanding marketers’ attention.

Some have locked themselves in their rooms, refusing to acknowledge its presence entirely. More are cautiously side-eying it from the corner of the room. But an increasing number of paid media professionals are actively interacting with it, understanding it, and figuring out exactly what it can do for them. 

And few were quicker to hand generative AI a mop and put it to work than Frederick Vallaeys.

Recently voted the #1 most influential PPC expert by PPCsurvey.com, Fred has decades of experience in the PPC industry. He even helped shape it, spending a decade as a Product Specialist / Ads Evangelist at Google, before becoming the Cofounder and CEO at Optmyzr - a company that specializes in making PPC teams more effective and efficient.

In a world where PPC effectiveness and efficiency is largely divided by those who use GenAI well vs. those who don’t, we took the opportunity to sit down with Fred to get his insights on how marketers can use tools such as ChatGPT to their full potential. 

Watch the full episode below, where Fred revealed how he’s harnessing the power of ChatGPT and other large language models to streamline campaign management and save thousands of hours on PPC tasks. Alternatively, read on for our full written breakdown of the session:

Timestamps:

0:00 - Intro: PPCSurvey results, PerformanceMCR event

5:10 - Which tasks should you automate with ChatGPT?

7:54 - How to write better prompts for GenAI

12:24 - Using GenAI for ad copy, keywords, socials

18:51 - Feeding ChatGPT the right data

23:40 - What tasks should you NOT use ChatGPT for?

28:07 - Final thoughts

Routine PPC tasks that can be automated by ChatGPT

AI can’t do everything… Yet. 

But as Fred pointed out, there’s a wide range of use cases that can streamline your workflows, sharpen your strategy, and save countless hours previously lost to mundane, repetitive tasks.

If you’re not already, consider making sure your generative AI strategy is put to use in the following areas:

Keyword development

GenAI significantly speeds up keyword research by generating comprehensive keyword lists from initial seed terms. 

By quickly brainstorming synonyms, variations, long-tail keywords, and industry-specific terminology, it can ensure thorough keyword coverage, helping marketers efficiently expand their reach and identify valuable opportunities they might otherwise overlook.

Ad group structuring

Effective ad group structuring is important for achieving high Quality Scores and enhancing ad relevance. 

GenAI automates this process by semantically grouping keywords, ensuring that ad groups are tightly focused around specific themes. This can improve ad targeting and overall campaign performance, by both minimizing wasted spend and boosting ROI.

Data analysis and seasonality

LLMs simplify complex data analysis tasks such as performance reviews and seasonality trends. It can automatically analyze historical PPC data to identify patterns, forecast seasonal demand shifts, and recommend budget adjustments proactively. 

Frederick highlights how marketers, without extensive statistical training, can easily use these detailed analytics to make more informed decisions.

Script automation

Frederick noted that scripting, enabled by ChatGPT, automates repetitive PPC tasks such as bid adjustments, campaign pauses, and budget reallocations. 

By developing scripts through simple natural language prompts, marketers reduce human error, improve consistency in their PPC accounts, and potentially reclaim thousands of hours previously spent on mundane tasks.

Feed optimization and cleanup

ChatGPT can help automate the important but extremely time-consuming process of optimizing product data feeds. 

This includes tasks such as standardizing feed entries, correcting data inconsistencies, and enhancing product descriptions for shopping and dynamic ads. Automating these improvements can result in higher feed quality, and ultimately more effective campaigns.

Competitive keyword gap analysis

Frederick also highlights how GenAI can perform competitive keyword analysis by comparing your keyword lists with those of competitors. This can reveal potential keyword opportunities and gaps, allowing marketers to adjust strategies proactively and stay ahead of competitors.

Advanced statistical modeling

Beyond basic data analysis, ChatGPT facilitates advanced statistical modeling, including predictive analytics and detailed performance forecasting. 

Marketers can easily generate sophisticated statistical models to understand potential outcomes and strategize accordingly, without needing deep analytical expertise.

Enhancing ad copy with ChatGPT

While LLMs won’t create organic, high-quality ad copy that truly understands the nuance of your product/service, they can certainly help you do it much faster.

Frederick emphasizes the importance of using ChatGPT as a tool for refining and enhancing your existing ideas rather than relying exclusively on AI-generated content. This approach maintains authenticity, originality, and creativity - critical factors for grabbing audience attention and differentiating your ads from competitors.

By verbally recording initial ad ideas or concepts using transcription apps (such as TalkNotes or Whisper), initial ideas and concepts can be refined and polished with ChatGPT. 

It can then help incorporate persuasive elements, optimize phrasing, and ensure clear alignment with your established brand voice.

Though ChatGPT isn’t a substitute for your unique brand personality, when combined with precise, thoughtfully structured prompts, it can rapidly generate multiple ad headlines and description variations for testing.

Leveraging ChatGPT for advanced data analysis

ChatGPT offers robust capabilities for complex data-driven analysis tasks. Frederick recommends:

• Advanced seasonality analysis to predict fluctuations and inform budget allocation.
• Identification of underperforming campaign segments for quick rectification.
• Automated generation of periodic performance reports for continuous improvement.

Frederick particularly highlights Optmyzr’s Sidekick feature, which dynamically integrates ChatGPT technology to provide real-time insights and recommendations that can be actioned directly from the Optmyzr interface. 

Tip:

• Combine ChatGPT’s analysis capabilities with visualization tools like Google Data Studio or Power BI to enhance report clarity and ease of stakeholder communication. (Learn how to do this here)

Best practices for prompt engineering

At the end of the day, ChatGPT (or any LLM for that matter) is only as good as the prompts you give it. Here’s Frederick’s key methodology for creating prompts that get the best from LLMs:

Create highly specific and restrictive prompts

Formulate prompts that are clearly specific about your desired outcome. The more detailed and restrictive the prompt, the more creatively and precisely the AI will respond. 

For example:

• Bad prompt: “Refine my ad copy, aim it towards young people.”
• Good prompt: “Refine the ad copy below, adopting a friendly and persuasive tone aimed at young professionals aged 25-35. Emphasize our sustainable practices and affordability. Keep it within 90 characters and avoid overly formal language or technical jargon.”

Use customized GPT models

Building custom GPT models tailored to specific contexts greatly improves output relevance. 

For agencies, Frederick suggests creating individualized GPT models for each client, incorporating essential details such as brand guidelines, historical performance data, the client’s PPC knowledge level, past marketing strategies, and preferred language style. 

These customized models ensure you don’t need to re-enter all the necessary pre-qualifying information for each client each time you need an output. It also ensures that outputs align more closely with client expectations and brand voice.

Employ a multi-persona approach

"Why limit myself to having one AI assistant, when there's really nothing holding you back from having a thousand custom GPTs, each with a different persona"
‍
Frederick Vallaeys

Even though a thousand is somewhat excessive, Frederick advocates for using multiple GPT personas to review and action feedback on things like performance reports or client side decks.

For example, you might use:

• A skeptical client to critically evaluate and question your decks, allowing you to prepare accordingly
• A fact-checker to ensure all statements and data are accurate.
• A CMO persona to assess strategic alignment and high-level messaging.
• A junior marketer to verify practical applicability.

You can then consolidate feedback from these personas to enhance your final output. And you know what can help you do this? Another customGPT. Use one to review the outputs from all the others, and deliver the most actionable suggestions to improve the report/slide deck/etc. 

This ensures that, rather than having to sift through an entire essay's worth of feedback, you’re spending your time on the most important aspects.

Use iterative refinement

Refining prompts iteratively based on how closely the output matches your expectations works extremely well with GenAI. The more you continually adjust your prompts, the more you enhance the AI’s understanding and accuracy over time. 

This iterative approach helps refine the outputs progressively, increasing their quality and reducing manual intervention.

Continuously monitor and adjust to combat drift

Regular monitoring and adjustment of your GPT prompts are crucial. Stay vigilant to detect any changes in output quality or accuracy over time due to AI model "drift." 

Fred mentioned that even the best GenAI models only get things right around 80% of the time, and most models perform much worse than that. As such, LLM outputs must be reviewed and double checked for accuracy.

What’s more, ensuring prompts remain precise and effective by regularly revisiting and updating them is vital for maintaining consistent, high-quality results.

Tip:

• Once you’ve vetted outputs for accuracy, save them for future use to navigate the AI drift issue. This means you won’t have to ask the model to create similar outputs from scratch every time you need something similar - which gives the LLM an additional chance to “drift”. 

The risks and limitations of AI in PPC

Generative AI + PPC marketers = more efficiency, more effectiveness, and more time saved. 

But, as marketers, we can’t use the excuse of AI to kick back and put our feet up. If anything, integrating AI into our workflows requires us to be sharper than ever, and keep the following potential risks in mind:

• Content genericization: Over-reliance on AI-generated copy risks diluting distinct brand messaging. This is why using AI to refine, rather than create, is key.
• Accuracy challenges: As mentioned, AI models may exhibit "drift," providing inconsistent analytical outcomes over time, requiring continuous human validation.
• Dependence on AI-generated insights: While beneficial, overreliance on AI-generated insights can potentially erode human strategic thinking. PPC marketers need to ensure they maintain an active role in strategy development, using AI as a supportive tool rather than a complete replacement. 

Practical tools and resources for PPC automation

To seamlessly integrate ChatGPT into PPC operations, consider:

• Automated data integration tools: Use platforms like Optmyzr or SEMrush for seamless PPC data integration with GPT capabilities. For more information on AI integration, check out Fred’s detailed Search Engine Land article on the 4 ways to connect your ads data to generative AI here.
• Tailored GPT models: Develop client-specific GPT models incorporating historical data, unique brand voices, and campaign-specific information.
• Voice transcription for initial concepts: Adopt transcription tools (Whisper, TalkNotes) for capturing initial strategic ideas that ChatGPT can later refine.
• For more advanced data integration, take a look at Anthropic’s Model Context Protocol, which Fred mentioned in the podcast. This works by creating a bridge between AI models and business tools. 
• Fred also called out Will Reynolds as a key thought leader in this space that people should be paying attention to. 

Tip:

• Read our blog on the best Custom GPTs for marketing to discover some of the best PPC GPTs currently available. 

Final thoughts

According to GetResponse, 71% of marketers are using ChatGPT for work. But our discussion with Fred made us realize that only a very small percentage of those are likely using it to its full potential.

The insights shared by Fred in this episode offer a practical guide for any business looking to optimize its paid media efforts - and save hundreds of hours that would otherwise have been spent on the important (yet mind-numbing) repetitive tasks that are forever put off, or passed on to junior marketers.

The discussion also served as a stark reminder of the importance of balancing AI automation with human input. Relying too heavily on AI can undermine campaign authenticity, strategic depth, and performance - proving just as inefficient as not using AI at all.

Make sure to watch the full episode with Fred above, and subscribe to the Paid Media Lab on YouTube (or listen on Spotify, Apple Music, or Amazon) to ensure you don’t miss an episode.

Big thanks to Frederick Vallaeys for taking the time to join us on this one. If you’re not already, make sure you’re following Fred on LinkedIn for his in-depth takes on the very latest in PPC news.