Google Ads has fundamentally changed how Lookalike segments function within Demand Gen campaigns. What was once a relatively rigid targeting constraint is becoming a more flexible, AI-driven audience suggestion.
For performance-focused advertisers, this is not a minor interface update. It is a structural shift in how reach, optimisation and scale interact.
If you previously relied on a Narrow Lookalike to keep targeting tightly confined to the people most similar to a seed list, the rules have changed. Your seed no longer necessarily builds a fence around who can see your adverts. It gives Google a direction in which to search for prospective customers.
That creates more potential scale, but it also makes the quality of your first-party data, conversion tracking, bidding and creative considerably more important.
This guide explains:
what Google Ads Lookalike segments are;
how the change from targeting constraint to audience suggestion works;
what Narrow, Balanced and Broad now mean;
why a Lookalike may appear with a “Signal” label in reporting;
how Lookalikes interact with Optimised Targeting;
how to build better seed lists from website, customer and CRM data;
how to measure lead quality rather than accepting Google’s attribution at face value; and
when retaining the previous targeting constraint may be appropriate.
What Are Google Ads Lookalike Segments?
A Lookalike segment helps Google Ads find people who share characteristics with users in one or more of your existing first-party data segments.
The source audience is known as the seed list. Depending on your data and eligibility, a seed might be based on:
previous customers;
Customer Match data;
website visitors;
app users; or
people who have interacted with your YouTube channel or videos.
Google models similarities between the seed audience and other people who may be suitable prospects. The objective is acquisition: reaching potential new customers rather than repeatedly advertising only to people you already know.
Lookalikes are therefore different from remarketing. Remarketing attempts to reach people already contained within a first-party audience. A Lookalike uses that audience as the foundation for finding new people. If you need to create and organise first-party lists first, see our guides to Google Customer Match and Google Ads Audience Manager.
From Hard Boundaries to AI Suggestions
Historically, Demand Gen Lookalike segments used three similarity ranges:
Narrow — approximately 2.5%: the people considered most similar to the seed within the selected location;
Balanced — approximately 5%: a larger pool intended to balance similarity and reach; and
Broad — approximately 10%: the widest pool, providing greater scale but looser similarity.
These percentages acted as targeting constraints. A Narrow setting attempted to restrict delivery to the most similar portion of the eligible population.
Under the newer suggestion-based behaviour, the seed list and selected reach tier are signals rather than absolute limits. Google AI will still use them to prioritise likely prospects, but it may reach people outside the former similarity threshold when it predicts that doing so will help achieve the campaign’s conversion or cost-per-action objective.
Google describes the change in its official Lookalike segment guidance. The rollout is phased, so advertisers should check the behaviour and reporting labels in their own accounts rather than assuming every ad group has changed at exactly the same time.
The simplest way to understand the difference is:
A traditional Lookalike was a fence. A suggestion-based Lookalike is a compass.
This is not merely a new name for Optimised Targeting. Lookalike-as-a-suggestion remains tied to the seed audience and its selected similarity level, while Optimised Targeting is a separate expansion system. However, both can extend delivery beyond the audience an advertiser initially supplied.
Why Google Changed Lookalike Targeting
The change fits the wider direction of Google Ads targeting: advertisers provide objectives, data and controls, while machine learning makes more auction-level decisions.
Several factors explain this direction.
Hard Similarity Thresholds Can Restrict Scale
A person just outside an arbitrary similarity threshold may be as valuable as someone just inside it. A fixed boundary can prevent the system from entering auctions that might produce conversions at an acceptable cost.
Suggestion mode gives Google more freedom to test those opportunities. This can increase reach and conversion volume, particularly when a narrowly defined Lookalike would otherwise exhaust its available audience.
Google Optimises for Outcomes, Not Similarity Purity
Similarity is a means to an end. The commercial objective is normally a conversion, qualified lead, sale or particular return—not the achievement of a mathematically pure audience match.
Google therefore wants to combine a useful seed with auction-time predictions and the campaign’s conversion goal. The model is trying to answer a performance question:
Is this person likely to complete the action the campaign is optimising towards at an acceptable cost or value?
That is different from asking only whether the person falls within a predefined similarity percentage.
First-Party Data Is Becoming More Important
Privacy restrictions, fragmented customer journeys and reduced reliance on third-party identifiers have made consented first-party data more strategically important. Lookalike modelling allows an advertiser to activate patterns from customer, website and CRM data without manually defining every demographic or interest characteristic.
Demand Gen Is Part of an Automation-First Advertising System
The change is consistent with Smart Bidding, audience signals, Performance Max and wider automation across Google Ads. The competitive advantage increasingly comes from the quality of the information supplied to the system. Our guide to training Google Ads AI explains why conversion inputs, commercial outcomes and campaign structure matter as manual controls decrease.
How Suggestion-Based Lookalikes Work in Practice
The exact auction model is proprietary, so advertisers should be careful not to present a speculative list of signals as if Google disclosed every input or weighting.
At a practical level, the system combines:
the characteristics and behaviour represented by the seed list;
the selected Narrow, Balanced or Broad starting point;
the campaign’s conversion goals;
the bidding strategy and efficiency target;
available Google signals and auction context; and
the conversion feedback generated as the campaign runs.
The Lookalike is an input to prediction, not a guarantee that every person reached is inside the former threshold. The system can test beyond it, observe conversion results and adjust delivery towards the campaign goal.
This makes the conversion action critically important. If the campaign optimises for every form submission, Google will seek people likely to submit forms. It cannot automatically know which enquiries were suitable, contactable or profitable unless that information is returned through a reliable measurement system.
For lead generation, the strongest feedback loop often looks like this:
Advert click → Website enquiry → Qualified Lead → Sales Opportunity → Customer → Revenue
Our Google Ads conversion tracking guide explains the initial website measurement, while offline conversion tracking connects later CRM and sales outcomes to the original advertising interaction.
Narrow, Balanced and Broad: What the Reach Tiers Mean Now
The three reach choices remain useful, but their meaning depends on whether suggestion mode or the previous constraint behaviour applies to the ad group.
| Reach tier | Previous constraint | Role in suggestion mode | Practical use |
|---|---|---|---|
| Narrow | Approximately 2.5% | Stronger starting preference for the most similar users, but not necessarily a hard ceiling | Efficiency-led prospecting and higher-intent seeds |
| Balanced | Approximately 5% | A middle starting point between similarity and reach | Controlled scaling where the seed has reasonable volume and quality |
| Broad | Approximately 10% | The widest similarity starting point, with scope for further exploration | Larger budgets, mature data and aggressive acquisition goals |
The slider is no longer the only expansion control. In suggestion mode, the practical reach of an ad group is also influenced by budget, bid targets, conversion volume, creative eligibility, channel inventory, geography and the strength of the campaign’s performance predictions.
A Narrow selection should therefore not be interpreted as a promise that all impressions came from the former 2.5% pool. Likewise, Broad is not automatically better simply because it offers more scale.
Where Google Ads Lookalikes Operate
The suggestion-based change covered in this article applies to Lookalike segments attached to Demand Gen campaigns. Google states that it does not change Lookalike use in other supported campaign environments or Display & Video 360.
Demand Gen itself can serve visual adverts across Google surfaces that may include:
YouTube, including Shorts, in-feed placements and other eligible YouTube inventory;
Discover;
Gmail;
Maps; and
eligible Google Display Network inventory.
Available inventory and controls can change, so review the current account settings rather than relying on an older assumption that Demand Gen means only YouTube, Discover and Gmail.
Performance Max works differently. It uses audience signals across a broader cross-channel system, but it does not use the Demand Gen Narrow, Balanced and Broad Lookalike framework in the same way. See our guide to Performance Max for lead generation for the differences in control, measurement and suitability.
The Strategic Importance of Seed Quality
AI expansion amplifies the meaning of the data supplied to it. A precise, commercially valuable seed gives Google a better direction than a large but poorly defined list.
Strong Seed List Examples
Useful high-intent seeds may include:
recent customers;
Closed Won opportunities;
repeat purchasers;
high-lifetime-value customers;
buyers of a particular profitable product or service;
qualified opportunities that meet agreed sales criteria;
customers with short or successful sales journeys; and
people who have recently completed a high-intent action.
Google also recommends high-intent seeds and recent converters. For some businesses, people who engaged through external media channels can add useful first-party evidence when that data is eligible and collected correctly.
Weak or Ambiguous Seed List Examples
Less useful seeds can include:
every website visitor regardless of intent;
all leads mixed together, including spam and unsuitable enquiries;
old, inactive or poorly maintained contact databases;
job applicants mixed with sales enquiries;
customers from unrelated products or markets; and
lists containing inconsistent consent, formatting or identity data.
A larger list is not automatically a better list. A mixed seed can teach the system that several very different types of person are equally valuable.
Do Not Mix Lead Volume With Lead Value
For service providers, a submitted form is rarely the final commercial outcome. Separate leads that were qualified, quoted, converted and lost. Where volume permits, create seeds based on the lifecycle stages and customer groups that genuinely represent the next customers you want to acquire.
A CRM integration with Google Ads makes this segmentation easier because the seed lists, exclusions and offline conversion signals can reflect the actual sales pipeline rather than the contents of a generic marketing database.
Google’s current guidance also states that the former minimum seed-size requirement of 100 users is no longer applicable to creating a Lookalike segment. That does not mean extremely small or sparse seeds will necessarily perform well. Eligibility and statistical usefulness are separate questions: a segment can be technically available while still providing limited modelling evidence.
What the “Signal” Label Means in Reporting
When a Lookalike is operating as a suggestion, Google can display a Signal label beside the Lookalike segment row in reporting.
The label indicates that:
the Lookalike is being used as a directional input rather than the previous strict similarity constraint;
delivery may extend outside the selected 2.5%, 5% or 10% starting range; and
performance attributed to this expansion can continue to appear against the Lookalike row.
That final point requires care. A Lookalike row can include performance generated by users outside the former similarity boundary. The row should not be read as a transparent list-membership report showing that every impression matched the chosen percentage.
You may observe greater reach, increased impression volume and short-term CPA movement while the system explores more inventory. These changes are not automatically positive or negative. They must be judged against qualified leads, customers, revenue and incremental performance.
Lookalikes and Optimised Targeting
An ad group can use both Lookalike-as-a-suggestion and Optimised Targeting. When both are enabled, two expansion mechanisms are active.
Google explains that more of the performance gain may be reported against the suggestion-based Lookalike, while Optimised Targeting can still find additional users beyond the Lookalike system’s reach.
This creates scale potential, but it also reduces the value of judging performance from the selected audience names alone. Advertisers should understand which systems are enabled and compare results at ad-group and campaign level.
For a cleaner test, consider separating strategies:
one ad group using the Lookalike approach you want to evaluate;
another using a broader audience or Optimised Targeting approach; and
distinct creative where the audience proposition genuinely differs.
Avoid unnecessary fragmentation. If each ad group receives too little budget or too few conversions, the test may create noise rather than insight.
Will Lookalike Expansion Increase CPA?
It can. Expansion allows Google to enter more auctions, including auctions outside the previously selected similarity boundary. Whether that improves or weakens efficiency depends on the whole system.
CPA volatility is more likely when:
the seed is broad, mixed or low intent;
website and CRM tracking are incomplete;
the campaign optimises for low-quality micro-conversions;
conversion volume is too low for stable learning;
the target CPA is unrealistic;
creative attracts people who are interested but commercially unsuitable; or
sales outcomes are not returned to Google Ads.
Expansion is more defensible when:
the seed represents valuable customers or Qualified Leads;
the conversion goal matches the commercial objective;
Enhanced Conversions for Leads or another reliable offline import method improves matching;
the bid strategy has enough accurate data;
exclusions remove audiences the business does not want to acquire; and
lead quality and revenue are measured outside Google Ads as well as inside it.
Do not assume that an increase in conversions reported by the platform proves incremental growth. Compare CRM outcomes, sales quality, assisted conversions and, where budgets allow, controlled experiments.
Bidding Becomes an Expansion Control
When audience boundaries become softer, the bid strategy and its target take on more responsibility.
Common conversion-focused approaches include:
Maximise Conversions: seeks as many conversions as possible within budget, without an explicit CPA target;
Target CPA: aims to generate conversions around a desired average acquisition cost;
Maximise Conversion Value: prioritises total reported value within budget; and
Target ROAS: aims to generate conversion value around a desired return.
An unconstrained Maximise Conversions strategy may be appropriate during some tests, but it should not be treated as automatically safe when cost control matters. Target CPA can place an efficiency objective around lead acquisition, while value-based bidding becomes more useful when conversion values reliably represent customer value or revenue.
Targets that are too aggressive can restrict delivery; targets that are too loose can allow expensive expansion. Review our guides to Google Ads bidding strategies, Target CPA bidding and value-based bidding before using a target as a substitute for good measurement.
Creative Is Now Part of Audience Qualification
As Google gains more freedom to expand, creative must do more than attract attention. It should also help unsuitable people recognise that the offer is not for them.
Strong qualification can include:
a clearly named service or product;
location coverage;
relevant industry language;
the customer type the offer is designed for;
meaningful pricing or eligibility information where appropriate;
a specific problem and outcome; and
a landing page that continues the same proposition.
Broad, generic creative may generate cheap engagement without producing valuable enquiries. This is especially dangerous in lead generation, where a form completion can look successful in Google Ads even when the sales team rejects the lead.
Demand Gen is visual and often reaches people before they actively search. Use the correct image and video formats, test materially different concepts and review performance by creative—not merely by audience setting.
A Practical Lookalike Strategy Framework
1. Start With a Commercially Meaningful Seed
Choose the customer or pipeline segment that best represents what you want more of. Do not begin with “all contacts” simply because it is the largest available list.
2. Confirm Data Quality and Eligibility
Check consent, formatting, recency, country coverage and whether contacts belong to the intended product or service. Customer data should be used in accordance with Google’s policies and applicable privacy requirements.
3. Separate Meaningfully Different Customer Groups
Do not mix high-value and low-value customers if the difference matters to profitability. Separate B2B and B2C customers, different services, different locations or materially different product categories where there is enough data and budget to support the structure.
4. Select a Sensible Reach Starting Point
Narrow is a reasonable efficiency-led starting point for a high-intent seed, but it is no longer a guaranteed boundary in suggestion mode. Balanced can support measured expansion. Broad is more appropriate when the advertiser has mature tracking, sufficient creative and a genuine need for additional scale.
5. Align the Conversion Goal
Use the action that best reflects the campaign’s purpose. For e-commerce, this may be a purchase with accurate value. For lead generation, start with the website enquiry but develop the system towards Qualified Leads, Sales Opportunities, customers and revenue.
6. Use Clear Exclusions
Depending on the objective, exclusions may include:
existing customers in a new-customer campaign;
employees and internal traffic;
job seekers;
unsuitable locations; and
first-party audiences reserved for a separate retention or remarketing strategy.
Lookalike segments automatically exclude people in their seed lists, but that does not remove the need for a wider exclusion strategy. A customer list used elsewhere, website visitors outside the seed and other known audiences may still require attention.
7. Give the Test Enough Time and Budget
Avoid making decisions from a few conversions or several days of volatile delivery. The required duration depends on spend, conversion rate and sales-cycle length. A lead-generation campaign cannot be judged properly before a meaningful portion of its leads has had time to qualify or progress.
8. Measure Downstream Performance
Evaluate:
cost per enquiry;
cost per Qualified Lead;
qualification rate;
cost per Sales Opportunity;
customer acquisition cost;
revenue per acquisition;
return on advertising spend; and
lifetime value where the data is mature enough.
CTR, impressions and video engagement can help diagnose creative, but they are not substitutes for commercial outcomes.
Value-Based Lookalike Expansion
If accurate conversion values are returned to Google Ads, the campaign can optimise for value as well as the probability of a conversion.
For e-commerce, this may mean using transaction revenue or margin-informed values. For lead generation, values can represent expected commercial worth at selected CRM stages, provided they are defined consistently and do not double-count the same customer journey.
This can help distinguish between:
a low-value and high-value order;
an unqualified enquiry and a genuine Sales Opportunity; or
a small contract and a strategically valuable customer.
Value-based bidding is not improved simply by assigning arbitrary numbers. The values should reflect observed economics closely enough to influence bidding in the right direction. If every lead is assigned the same value, the system still sees every lead as equally valuable.
When Retaining a Targeting Constraint May Make Sense
Google provides a route for advertisers who need the previous constraint-based Lookalike behaviour. The current process and account controls can be checked in Google’s targeting-constraint guidance.
Retaining stricter behaviour may be worth considering when:
the campaign operates under sensitive-category or compliance restrictions;
the addressable market is extremely specialised;
broader delivery creates a documented brand-safety or lead-quality problem;
the campaign is designed as a controlled audience test; or
the business cannot yet measure the downstream quality of expanded traffic.
Do not opt out simply because automation feels uncomfortable. First determine whether expansion is producing worse commercial outcomes. Equally, do not accept expansion merely because Google predicts improvement. The correct decision should be based on evidence from your account.
Common Questions About Google Ads Lookalikes
Are Google Ads Lookalike Segments the Same as Meta Lookalike Audiences?
They serve a similar strategic purpose: both use an existing audience to help find new prospects. However, the platforms use different identity systems, inventory, modelling, controls and reporting. Google’s suggestion-based direction resembles Meta’s wider move towards algorithm-led audience expansion, but performance on one platform does not predict performance on the other.
Does Narrow Still Mean Only the Closest 2.5%?
Not when the ad group is using the newer suggestion behaviour. Narrow remains a stronger similarity starting signal, but Google may deliver outside the former threshold. If the previous constraint behaviour is retained, the reach slider continues to apply the original range.
Does the Signal Label Mean Optimised Targeting Is Enabled?
Not necessarily. The Signal label can indicate that the Lookalike itself is being treated as a suggestion. Optimised Targeting is a separate setting, although both systems may be active in the same ad group.
Are Lookalikes Suitable for Small Lead-Generation Businesses?
Potentially, but they are rarely the first priority when budget and conversion volume are limited. Google Search campaigns normally capture clearer bottom-of-funnel intent. Demand Gen Lookalikes become more compelling when the advertiser has strong creative, reliable tracking, suitable first-party data and enough budget to test prospecting beyond existing demand.
Should I Use All Website Visitors as the Seed?
Usually not as the first choice. An all-visitors seed can contain customers, researchers, job seekers, irrelevant traffic and people who bounced immediately. A smaller high-intent segment—such as customers, qualified opportunities or visitors to commercially important pages—may provide a clearer direction.
Can Lookalikes Replace Remarketing?
No. They perform different roles. Remarketing re-engages known visitors or customers, while Lookalikes seek new people who resemble a seed. A mature strategy may use both, with separate messages, exclusions and performance expectations.
Final Thoughts: Fence Versus Compass
Traditional Lookalikes were fences. Suggestion-based Lookalikes are compasses. Google may now look beyond the former fence when its system predicts that wider delivery can achieve the campaign objective.
That can unlock valuable inventory which strict similarity ranges might have missed. It can also amplify weak data, attract low-quality leads and obscure where delivery actually occurred if advertisers rely too heavily on audience-row labels.
The practical controls have shifted towards:
seed quality;
conversion tracking accuracy;
CRM and offline sales feedback;
bidding targets;
audience exclusions;
creative qualification; and
downstream measurement.
The strategic conclusion is not that targeting no longer matters. It is that targeting now includes the data, goals and feedback used to guide the algorithm—not only the audience option selected in the interface.
If the business supplies Google Ads with high-quality first-party data and measures real commercial outcomes, a Lookalike can become a useful scaling tool. If it supplies mixed leads and optimises for superficial conversions, automation may simply find more of the wrong people.
Need Help With Demand Gen and First-Party Data?
One PPC can review your Demand Gen structure, Lookalike seeds, bidding, conversion tracking and CRM feedback loop. The objective is not expansion for its own sake, but profitable reach supported by measurable lead quality or revenue.
Contact One PPC to discuss a Google Ads audit or campaign-management strategy.
Related Google Ads Guides
