As of March 15, 2026, Google Demand Gen Lookalike segments stopped being a hard targeting boundary and became a signal that guides Google’s AI. Where a Lookalike segment once capped your reach to the 2.5%, 5%, or 10% similarity tier you chose, the AI can now expand beyond that tier whenever it predicts a better conversion rate or lower CPA — the reach level you set is a hint, not a ceiling (PPC Land). Google re-surfaced the change in its October 2026 Demand Gen guidance. This post covers what a B2B SaaS account loses, why it matters more here than for consumer advertisers, and how to file the opt-out that restores strict targeting.
The direct advice: if your Demand Gen seed lists are narrow — closed-won accounts, high-intent converters — file Google’s strict-targeting opt-out form, or at minimum watch CPA and lead quality closely, because AI expansion off a tight B2B seed is exactly where budget leaks into cheap, unqualified leads.
What actually changed
A Lookalike segment is built from a seed — a customer list or high-value audience — and a reach tier that controls how similar the targeted users must be. Until March 2026 that tier was a hard constraint: pick the 2.5% tier and your ads reached only the users the model judged most similar to your seed, full stop. The update converts that constraint into a directional signal. The AI still prioritizes users like your seed, but “the system may go beyond those parameters if its models predict better conversion performance or lower cost-per-action” (Search Engine Land).
In plain terms: you no longer decide the outer edge of who gets reached — the algorithm does, and it will cross your chosen boundary whenever its predictions favor doing so. For high-volume consumer campaigns with broad seeds and transactional conversions, that expansion often helps, which is the case Google optimizes for. It is the same hand-more-to-the-AI direction as the broader shift toward audience signals over hard targeting across the platform.
Why narrow B2B seeds lose the most
B2B SaaS is the worst-fit case for uncapped expansion because the value of a Lookalike here lives in its precision. A strong B2B seed is small and deliberate — a few hundred closed-won accounts, or the converters who became paying customers. That list is a tight definition of the exact buyer you want more of. Let the AI expand past it to chase predicted conversions and the definition erodes, because the conversion it is expanding toward is usually the cheapest one available.
In a lead-gen account that cheapest conversion is typically a form fill, not a sales-qualified opportunity. So expansion pulls in volume that reports as success in Google Ads while quietly lowering the share that reaches pipeline — the classic gap between a lead and a real buyer we unpack in optimizing for SQLs, not leads. A consumer advertiser with a large seed and a transactional purchase absorbs this easily; a SaaS selling a $20k ACV product paying for the wrong expansion does not.
How to keep strict targeting: the opt-out
Google provides a dedicated opt-out form for advertisers who want Lookalike segments to keep working as a strict constraint. Submit the form requesting continued access to strict Lookalike targeting and the original hard-ceiling behavior is restored; reports put the turnaround at about one week (Herdl). Opted-out segments enforce your chosen similarity tier instead of letting the AI expand past it. For most narrow-seed B2B SaaS accounts this is the right default — file it deliberately rather than discovering three weeks later that CPA drifted up as the audience widened.
If you leave expansion on by choice — some accounts with broader top-of-funnel seeds benefit — then treat CPA and downstream lead quality as the control instead, and review them weekly rather than monthly so a quality slide surfaces early. Either way, the setting is a decision to make on purpose, not a default to accept. Pair the review with disciplined budget control, as covered in demand-led budget pacing, so expansion cannot quietly outrun the spend you intended.
The objective is the more durable control
Fighting the targeting setting is necessary but not sufficient. The more durable constraint on AI expansion is the conversion you ask it to optimize toward. If Google is optimizing for raw form fills, expansion that finds cheap form fills looks correct to the algorithm even when the leads are worthless. Feed it a conversion that reflects pipeline — an SQL, a qualified opportunity, revenue imported from your CRM — and the same expansion is judged against a signal that actually maps to your business, so it self-limits toward the users who become customers.
That is why this change is really an argument for getting your conversion definitions and offline data right before you scale any audience-driven channel, Demand Gen included — the same foundation behind scaling Demand Gen campaigns without wasting spend. If you are running Demand Gen off narrow seeds and are not sure whether the AI is widening your audience past what converts, a structured Google Ads audit will check your Lookalike settings and conversion objectives together, and the free 10-point audit checklist is a quick first pass before you change anything.