Picture this: you’ve got a client meeting in an hour and they are expecting a performance breakdown. To save time, you feed the raw data into Claude and ask for a quick MoM revenue calculation.
Trouble is, the AI hallucinates. Instead of showing the true 10% MoM growth, it confidently throws out a 30% surge. Because you were rushing, you skip the QA and drop the number into your report.
Mid-meeting, the client asks why your numbers don’t match their internal database. In an instant, they become skeptical of your attention to detail and start questioning your entire way of working.
In 2026, PPC marketers have a wealth of automated tools, metrics, and platform features at their disposal. While they’re helpful, they’re engineered by companies that benefit financially from us adopting them (whether they work perfectly for our specific goals or not).
Because these platforms have a built-in financial incentive, the burden falls on us to dig into the documentation, understand how the tools work, and identify potential risks. During this AI boom, critical thinking is exactly what will set strong advertisers who drive real results apart from those who adopt everything blindly.
Data Analysis: Examining Performance Metrics (QFC & ABS As Examples)
Google tells us we can finally evaluate upper-funnel activity using predictive metrics like Qualified Future Conversions (QFC) and Attributed Brand Searches (ABS). This sounds great on paper because we all want to protect demand generation spend, but we shouldn’t blindly trust the chart just because Google calls it a great metric & we’re excited.
As stewards of client budgets, we have to look past a romanticised version of the truth. For example, if Google Ads claims a Demand Gen campaign drove a 20% lift in “future demand” via QFCs over a 60-day period, cross-reference that model with raw business reality. Look at your total direct, organic, and brand search volumes outside of Google Ads during that exact same window.
If the platform shows a massive spike in attributed future value, but total backend revenue and baseline brand searches remain completely flat, that modeled demand may be a phantom, not actual business growth. The point of skepticism isn’t to yell “Google is bad!”, it’s just to ensure we are auditing the full picture.
The risk of never auditing this is that you continue to pour money into campaigns that don’t drive the demand you want because they look good in UI. The way we examine our platform ROAS is the way we should examine new metrics.
Feature Adoption: Evaluating New Features (UCP As An Example)
Google’s Universal Commerce Platform (UCP) is another shiny new feature generating buzz. It allows us to intercept conversational search queries in real time and present products as immediate solutions. It completely removes checkout friction, which is great for quick sales, but it introduces huge structural shifts that require a critical review.
First, consider the loss of brand identity. If users buy right inside the AI chat interface, your unique brand site gets stripped away. Traditional e-commerce brands risk turning into generic marketplace products, making you look less like an independent retailer and more like a commodity seller on Amazon or Shein. Second, you lose a mountain of web behavior data. If native checkout means users are no longer coming to your site, you have to seriously consider what you will substitute those lost behavioral insights with before you blindly adopt the feature.
This might be more applicable to small/mid-sized businesses who’re still building brands and audiences, but still worth considering even if you have a strong brand. Critical thinking is less about finding problems in every feature and more about ensuring you’ve done your due diligence to assess its role within your advertising strategy & the potential risks + benefits.
AI Workflows: Guardrails
AI and custom agents have been discussed to death, so I won’t drone on. Instead, I’ll focus on the precise ways AI can fail and how we can check in on them. To keep AI from becoming an account liability, you need clear operational guardrails across your workflows:
- Inputs: High-quality output requires clean inputs. Use strict prompt guidelines from the platforms.
- Compliance: Ensure you’ve read up on security & privacy guidelines. Don’t risk inputting sensitive or private data just to save half an hour on an analysis.
- The Right Model for the Job: Match the tool to the task. Is a standard conversational LLM actually the best option for a massive N-gram analysis, or should you be using a dedicated script instead?
- Outputs: Even with perfect inputs, AI is just making statistical predictions about the most probable next word or number. You have to manually review the response to make sure it’s accurate and makes real-world sense.
- Training: This is absolutely paramount for team members who lack deep experience with these models.
If you aren’t putting guardrails in place, things can go wrong quickly. For example, Claude recently confidently claimed that it’s not possible to set keyword-level URLs in Google Ads. It stated it as a hard technical fact. If a junior had trusted that hallucination blindly, they would have defaulted to a broader, inappropriate ad-level URL, completely destroying the account’s landing page precision and tracking.
This is a relatively minor hallucination, but try and imagine if it had hallucinated something more serious like ad copy, inserting inappropriate snippets of text in or mixing languages up. You could end up with German ad copy in an English language campaign without even knowing. Hence the importance of checking.
The Takeaway
Ultimately, our value as modern paid experts isn’t in how fast we can generate a report or spin up an asset, especially since the systems can handle the manual heavy lifting now. Our value rests entirely on our judgment. Efficiency means nothing if you are driving your campaigns down the wrong road. Automation is a powerful copilot, but never let a tight deadline tempt you to skip the human check.
