Trust AI, but check what it is actually doing
Google Ads automation can be a powerful growth engine, but industry practitioners warned this week that advertisers risk confusing optimisation with commercial strategy. Speaking on the PPC Live podcast, Mike Ryan, head of e-commerce insights at Smarter Ecommerce, described a familiar pattern: AI-driven expansion can deliver headline performance while quietly driving traffic the business never wanted. The discussion was first reported by Anu Adegbola for Search Engine Land. (See the full report on Search Engine Land.)
When automation meets business context gap
Ryan outlined a case where AI Max substantially grew clicks for a client by moving into queries associated with a much larger rival. On surface metrics the results looked impressive. Behind the scenes, however, the advertiser had deliberately been steering clear of that competitor to avoid a costly bidding war. The algorithm had no way to understand that strategic context, and the campaign began to deliver outcomes at odds with the client’s commercial plan.
That experience underpins Ryan’s central maxim: trust AI, but verify. New features and strong top-line KPIs demand close monitoring from day one. In practice that means checking search terms, match type reporting and match source data rather than relying on aggregate impression or conversion numbers alone.
Early conclusions can mislead the market
Ryan also shared one of those industry mistakes that spreads rapidly: after seeing certain patterns he concluded AI Max over-indexed the Search Partners network and posted the finding on LinkedIn. With more data, that hypothesis did not hold up. The episode is a useful reminder that initial observations from limited samples can be wrong, and once an assertion goes viral it is almost impossible to fully retract. As Adegbola notes in her coverage, the stakes are high because agencies and in-house teams often make immediate budget and bidding decisions based on widely read commentary. Read the original piece by Anu Adegbola on Search Engine Land.
Guardrails exist but must be used properly
Thankfully Google’s automation is not entirely freewheeling. Ryan pointed to practical levers advertisers can use to prevent unwanted expansion: negative keywords, brand safety inclusions and exclusions, and Search Partner settings. Performance Max also exposes controls such as match source and match type reporting so advertisers can see where expansion is coming from and act accordingly.
The problem is not the lack of controls but the inconsistent use of them. Advertisers who treat AI as a black box often only notice issues after campaign budgets have moved in unintended directions.
Strategic segmentation trumps tactical fragmentation
A recurring theme in Ryan’s research is the harm caused by over-segmentation. Highly granular account structures that once made sense for manual bidding can starve automated strategies of the conversion volume they need. According to Ryan, Smart Bidding performs reliably only with sufficient data; he advises aiming for at least 30 conversions per month per bidding unit and preferably 60 or more where possible.
That means campaign splits should be strategic, not reflexive. Every additional campaign or split must have a clear commercial rationale rather than reproducing legacy structures designed for manual optimisation.
When CFOs and algorithms collide
Ryan described a particularly instructive example: an account split into many tiny buckets created by a finance-driven demand for separate margin and ROAS targets. On paper the segmentation looked analytically neat, but in practice it multiplied the number of bidding units and left each with insufficient signals for the algorithm to learn. The result was a fragmented account that could not meet the very targets the CFO had set.
Custom labels as a pragmatic source of business context
One constructive workaround Ryan recommends is feeding business data into Google via custom labels: margin tiers, sell-through speed, return rates and other product-level signals that Google does not natively possess. These labels can help align automated bidding with true profitability rather than a narrow conversion metric.
Channel importance and unintended tolerance
Performance Max has introduced channel importance settings that let advertisers shift how Google balances asset distribution. Ryan warned that increasing a channel’s importance can have an unintended side effect: it may give the algorithm more leeway on CPA or ROAS to achieve presence in that channel. In short, toggling importance is not a neutral action and can produce outcomes that differ from advertiser expectations.
The August 17 reaction and a possible market opportunity
Ryan’s research looked directly at the market response to Google’s August 17 Smart Bidding update. He found many advertisers reacted by raising target ROAS and pulling back spend, often in accounts that were not constrained by budget and therefore had no need for conservative changes. In other words, fear of an algorithm change induced conservative behaviour rather than data-driven adjustments. Ryan suggested this collective pullback could create an opportunistic opening for more aggressive advertisers to capture market share. Search Engine Land carried coverage of the Smart Bidding update and how the industry responded.
Patience, not panic: let the cycles complete
One practical point Ryan emphasised was Google’s own guidance: allow one to two conversion cycles after a material change before drawing conclusions. Immediate claims about the impact of platform updates are frequently premature. Advertisers and commentators should give algorithms time to stabilise and gather sufficient signal before making sweeping changes.
Curiosity as a core competency
Beyond techniques and controls, Ryan framed a cultural shift: curiosity is becoming a core PPC skill. In an era where automation handles many tactical decisions, the human advantage lies in scepticism, experimentation and the discipline to investigate what performance metrics actually mean for the business. Good practitioners will treat AI as a partner that needs oversight, not as a substitute for strategic thinking.
Practical takeaways for Irish advertisers
– Monitor search terms and match source reporting from day one when enabling new automation features.
– Use negative keywords, brand exclusions and Search Partner settings proactively.
– Consolidate segments where necessary so Smart Bidding has sufficient conversions to learn.
– Feed business-critical signals into Google using custom labels to align bids with profitability.
– Resist kneejerk changes after platform updates; wait one to two conversion cycles before revising targets.
– Encourage curiosity and rigorous testing in teams so that automation amplifies strategy rather than obscuring it.
The conversation between Ryan and the PPC Live team is available on the PPC Live show pages and podcast feed. For a deeper read, see Anu Adegbola’s report on Search Engine Land and the site’s coverage of Google’s Smart Bidding expansion.
