AI is everywhere in PPC right now.
Open LinkedIn and you’ll see claims that Claude, ChatGPT or some custom AI workflow has transformed an agency overnight. You’ll also see predictions that AI will replace account managers, automate optimisation and completely change how agencies operate.
The reality is far less dramatic.
During a recent PPC Live Slack AMA, PPC influencer and automation specialist Benjamin Wenner shared his experiences integrating Claude with Google Ads. The discussion quickly moved beyond the hype and into the practical realities of using large language models in PPC workflows.
The biggest takeaway?
AI can be incredibly useful, but we’re still a long way from handing over the keys to our ad accounts.
Most people haven’t connected Claude yet
One of the first questions posed to the PPC Live community was simple:
Who has actually connected Claude to Google Ads?
The response was telling.
Very few people had tried it, and the one attendee who had attempted a setup had already run into technical issues.
This probably reflects the wider PPC industry. While everyone is talking about AI, relatively few practitioners have moved beyond prompting tools and into direct platform integrations.
Part of the challenge is that there are now multiple ways to connect AI systems to advertising platforms, ranging from Google Sheets workarounds to third-party tools and official MCP (Model Context Protocol) servers.
According to Wenner, simplicity matters.
Every additional layer creates another point of failure. While using spreadsheets as a middleman can work, a direct connection via an official MCP server is generally the cleaner and more reliable approach.
What PPC professionals are actually using Claude for
One of the most common questions during the AMA was how Claude is being used in day-to-day PPC management.
Interestingly, the answer wasn’t campaign creation or automated bid management.
Instead, Wenner described using Claude primarily for:
- Daily account briefings
- Performance summaries
- Reporting workflows
- Data analysis
- Building custom PPC agents and automation systems
He also revealed that earlier experiments allowing Claude to make account changes directly were less successful than expected.
“I used to play around with doing changes via Claude, but things didn’t work so well.”
That experience led him towards more structured agentic systems rather than allowing an AI model to make direct account modifications.
For many PPC professionals, this is probably the safest place to start. Use AI to interpret information and accelerate analysis before trusting it with account changes.
The failure rate is higher than most people admit
Perhaps the most refreshing part of the conversation was the honesty.
When asked how he prevents Claude from doing anything “crazy”, Wenner’s response was simple:
“It does tons of crazy stuff.”
In fact, he estimated that failures currently outnumber successes by roughly three to one.
That’s an important reality check for an industry often dominated by AI success stories.
Many of the workflows being built today are still experimental. Even systems that work reliably can occasionally break due to API changes, connector issues or unexpected behaviour.
Human oversight remains essential.
The lesson is not that AI doesn’t work.
The lesson is that AI currently works best when supervised.
The biggest mistake: Building AI for the sake of AI
One of the most interesting observations came from the audience.
Many businesses become excited about AI and start building complex systems to solve problems that already have simpler solutions.
As one attendee put it, it’s like building a parcel-carrying robot when what you actually need is a conveyor belt.
This resonated strongly throughout the discussion.
Not every reporting problem requires Claude.
Not every workflow needs an agent.
In many situations, established tools such as Supermetrics, Looker Studio or traditional reporting systems remain the more practical option.
The challenge isn’t finding somewhere to use AI.
The challenge is finding a use case where AI genuinely creates value.
Skills and MCPs: The real productivity multipliers
While AI models receive most of the attention, Wenner argued that the real power comes from combining Skills and MCP servers.
Skills allow users to define structured workflows, preferred outputs and repeatable best practices.
The catch?
Most public Skills are generic.
After experimenting extensively, Wenner found himself using mostly custom-built Skills tailored to his own processes, communication style and client requirements.
For PPC practitioners starting out, his advice was straightforward:
Begin with public Skills from trusted sources, learn how they work, then gradually build your own versions around your specific workflows.
The more personalised the Skill becomes, the more useful it tends to be.
The data privacy question isn’t going away
No discussion about AI would be complete without addressing privacy and compliance.
Several attendees raised concerns about client confidentiality, GDPR and whether performance data should be shared with third-party AI systems.
The answer, unsurprisingly, was nuanced.
For most agencies and businesses, enterprise privacy controls offered by providers like Claude may be sufficient. Features such as disabling model training and turning off memory generation can reduce risk.
However, organisations requiring complete control over data may need to host AI models in isolated environments through services such as AWS or Microsoft Azure.
The trade-off is complexity.
More security generally means more infrastructure, more maintenance and more technical expertise.
As Wenner noted, there is rarely a simple answer.
Should junior PPC professionals be learning this?
Towards the end of the session, the conversation shifted to skills development.
Should junior PPC professionals be learning how to connect LLMs to advertising platforms?
The consensus was yes—but with an important caveat.
AI should supplement PPC knowledge, not replace it.
Understanding campaign structure, keyword strategy, audience targeting and measurement remains essential. Without those foundations, AI simply accelerates mistakes.
Early exposure to AI tools makes sense.
Heavy reliance on AI probably does not.
At least not until someone has developed a solid understanding of how PPC actually works.
The future will probably be more boring than people think
When asked how agencies will use AI in the future, Wenner offered a prediction that may disappoint the more enthusiastic futurists.
Most agencies are unlikely to build and maintain their own AI infrastructure.
Instead, they will probably adopt official connectors, MCP servers and supported integrations offered by major platforms.
Why?
Because maintaining custom AI infrastructure is difficult.
Version control becomes messy.
Updates become a burden.
Technical debt accumulates quickly.
For many agencies, the simplest solution will be the best one: use officially supported tools and focus on solving business problems rather than managing technology stacks.
Final thoughts
The PPC industry’s relationship with AI is entering a more practical phase.
The conversation is slowly moving away from grand claims about replacing account managers and towards more useful discussions about reporting, analysis, workflow automation and operational efficiency.
Claude, ChatGPT and other large language models clearly have a role to play.
But the agencies seeing the most success aren’t treating AI as a replacement for expertise.
They’re treating it as an amplifier.
And for now, that’s probably the most realistic expectation anyone should have.
