On September 1, 2026, Google began automatically upgrading certain Search campaigns to AI Max. If your campaigns use campaign-level broad match or standalone Automatically Created Assets, yours may already be in the queue. The question isn't whether AI Max can find new traffic. It can. The question is whether the traffic it finds aligns with decisions your business already made.

Good metrics, wrong strategy

Mike Ryan, head of e-commerce insights at Smarter Ecommerce, shared a case that should make any B2B operator pause. AI Max expanded a client's traffic into searches involving a much larger competitor. The dashboard looked strong: volume up, cost metrics in range. But the advertiser had deliberately avoided those auctions to prevent a bidding war. The AI couldn't know that. It saw opportunity in the data. The business saw a strategic violation.

Ryan's framing is blunt: "Optimization isn't strategy." Automation improves metrics against the goals you configure. It doesn't understand which fights you've chosen to avoid, which segments carry bad unit economics, or which competitor overlap would trigger an escalation your margin can't absorb.

The guardrails that actually exist

Google's documentation lists real controls. For AI Max: negative keywords, brand inclusions and exclusions, Search Partner Network settings, and match type/match source reporting. For Performance Max: campaign-level negative keywords (applying to Search and Shopping inventory), brand exclusions, URL exclusion rules, page feeds for landing-page control, and first-party audience exclusions added in 2026. These aren't cosmetic. They're the difference between steering the system and hoping it steers itself.

The catch: PMax negative keywords don't apply across all channels. Reporting shows channel-level performance but doesn't give you channel-level budget control. You can see where spend went after the fact. You can't pre-allocate it. Governance shifts from "where does the money go" to "what inputs and exclusions constrain the system, and how fast do we catch drift."

Conversion configuration is the strongest guardrail

This is the part most teams underinvest in. Smart Bidding optimizes to whatever conversion actions you've configured. If those actions are form fills or MQLs with no downstream quality signal, the AI will produce exactly what you asked for: more low-quality leads, faster, at a lower cost per lead. It'll look great in the weekly report and terrible in the pipeline review.

Ryan's operational guidance: use custom labels to feed the platform information it can't infer on its own, like margins, sell-through rates, and return rates. For B2B, the equivalent is importing offline conversions (qualified meetings, opportunity creation, closed revenue) so bidding aligns with pipeline outcomes instead of proxy metrics.

Don't starve the algorithm, but don't over-segment either

Ryan's rule of thumb: Smart Bidding needs at least 30 monthly conversions per campaign to work consistently, with 60+ being the target. Over-segment into hyper-granular structures and individual campaigns won't accumulate enough data for the algorithm to learn. He described one account where a CFO had dictated extremely granular margin-based campaign splits, each with its own ROAS target. Sounded rigorous. In practice, it fragmented the data so badly the algorithm couldn't hit any of the targets. The corrective isn't to collapse everything into one campaign. It's to segment only when there's a strategic or measurement reason.

The pre-migration checklist

If you're running Search campaigns that might be auto-upgraded to AI Max (Dynamic Search Ads migration is now scheduled for February 2027), lock down these items before the switch:

Early conclusions are expensive

Ryan revised his own initial finding about AI Max's Search Partner behavior after collecting more data, and he's candid about the cost of premature claims: a correction rarely gets the reach of the original post. Google recommends waiting one to two conversion cycles before drawing conclusions about changes. In B2B with longer sales cycles, that window might be even wider.

The hypothesis (make it falsifiable): if we implement negative keywords, brand exclusions, and offline conversion imports before AI Max migration, then post-migration qualified pipeline volume will hold within 10% of baseline because the system's expansion will be constrained to strategically aligned queries. Success = qualified pipeline volume within 10% of pre-migration baseline at 4 weeks. Guardrails = weekly search-term review, match-source reporting. Stop-loss = if CPA on qualified opportunities rises 25%+ in the first two conversion cycles, pause expansion and audit conversion configuration.

Ryan's operating principle, "trust but verify," sounds measured. In practice it means something more demanding: every default the platform sets is a decision you didn't make, and the cost of not reviewing it compounds quietly until the pipeline review where someone asks why lead quality fell off a cliff while the dashboard showed green across the board.