LinkedIn Predictive Audiences vs. Lookalikes: What Replaced Lookalikes and How to Use It

If you ran LinkedIn ads before 2024, you probably used lookalikes to scale. You uploaded a list of customers, and LinkedIn found people who looked like them on paper. Same job titles. Same industries. Same company sizes.

That tool is gone. Its replacement works in a different way, and many teams still treat it like the old lookalike. That's why some advertisers see great results with predictive audiences while others see little change.

Quick answer: LinkedIn replaced lookalike audiences with predictive audiences in February 2024. Predictive audiences use AI to find LinkedIn members who are likely to take the action you care about, such as filling out a lead form or converting on your website. You build one from a seed of at least 300 people, like a contact list, a lead gen form, or conversion data.

What happened to LinkedIn's lookalike audiences?

LinkedIn retired lookalike audiences in early 2024. After February 29, 2024, advertisers could no longer create new lookalike audiences or edit existing ones, according to Search Engine Journal's report on the change. LinkedIn pointed advertisers to two tools instead: predictive audiences and Audience Expansion.

Here is the timeline in short:

[table]
Date | What changed
2019 | LinkedIn launches lookalike audiences
2023 | Predictive audiences start rolling out
February 29, 2024 | No new lookalikes can be created or edited
After February 2024 | Existing lookalikes become static and stop refreshing
30 days unused | Static lookalikes are archived automatically
[/table]


Two other details matter if you still have old campaigns:

• Old lookalikes froze in place. They no longer update as your source list changes. Over time, they drift away from your real customer base.
• Third-party tools lost access. The API that let platforms like HubSpot create LinkedIn lookalikes was shut off. Any automated sync that built lookalikes stopped working.

Does LinkedIn still have lookalike audiences?

No. LinkedIn no longer lets you create lookalike audiences. If you see an old lookalike in your account, it is a frozen copy from before 2024. The closest replacement is a predictive audience, which you create under Plan > Audiences in Campaign Manager.

Why did LinkedIn replace lookalikes?

LinkedIn has not shared one official reason, but the change follows a clear industry pattern. Ad platforms are moving away from "find people who look similar" toward "find people likely to act." Meta, Google, and TikTok have all shifted toward AI models trained on outcomes like conversions and leads.

For LinkedIn, this shift makes sense. Two people can share the same job title and company size, yet only one is in the market to buy. A model trained on actions can spot that difference. A model trained only on profile traits cannot.

What are LinkedIn predictive audiences?

LinkedIn predictive audiences are AI-built targeting segments that find members most likely to take a specific action, based on a seed of your own data. You give LinkedIn a source, such as a contact list, a lead gen form, or conversion events. LinkedIn studies who in that source took action, then finds other members with a similar chance of doing the same.

Think of it this way:

• A lookalike asks: "Who looks like my customers?"
• A predictive audience asks: "Who is likely to act like my customers?"

That small change in the question leads to a big change in who you reach.

How do LinkedIn’s predictive audiences work?

Predictive audiences work in three stages: LinkedIn matches your seed to member profiles, trains a model on what those members have in common, and then scores the wider LinkedIn network to build an audience of likely converters. The steps look like this:

1. Match. LinkedIn matches the people or companies in your source to LinkedIn members. You need at least 300 matched members for the audience to build.
2. Learn. The model looks at the signals your matched members share. These include professional traits like job function, seniority, industry, and company size. They also include on-platform behavior, such as the content people engage with and how they interact with ads and forms.
3. Score and build. LinkedIn scores members in the locations you choose and keeps the ones most likely to act. The result is a new audience you can target in your campaigns.

LinkedIn's Predictive Audiences API documentation shows two more details that most guides skip:

• A location filter is required. The audience is always built for a geography you set. You cannot build a "global" predictive audience with no location.
• Size is a setting, not a fixed output. LinkedIn suggests a recommended audience size that balances reach and performance. You can accept it or pick your own size. For company-list seeds, a multiplier controls how many similar companies get added. For example, a multiplier of 2 on a 10-company seed returns about 20 companies.

What data sources can you use?

You can build a predictive audience from one of these source types:

[table]
Source type | What it is | Best used for
Contact list | Emails or contact details you upload or sync from a CRM | Finding people similar to your best customers or SQLs
Company list | A list of accounts | Finding new accounts that fit your ICP
Lead gen form | People who submitted a LinkedIn Lead Gen Form | Scaling lead volume from forms that already convert
Conversions | Actions tracked by the Insight Tag or Conversions API | Finding people likely to complete a website action, like a demo request
[/table]


Each predictive audience uses one source type. You can combine several sources of the same type, such as two contact lists, but you cannot mix a contact list with conversion data in one audience.

If you are new to list uploads, start with our guide to LinkedIn matched audiences. Matched audiences are the lists and retargeting segments that often become your predictive seed.

Predictive audiences vs lookalikes: what is the difference?

The main difference is what each model tries to match. LinkedIn lookalike audiences match profile traits, so they find people who resemble your list. Predictive audiences model likely actions, so they find people who are likely to convert, fill out a form, or engage. Predictive audiences also use more signals, give you more control over size, and stay current as your data changes.

[table]
Feature | LinkedIn lookalike audiences (retired) | LinkedIn predictive audiences
Status | No new audiences after Feb 29, 2024 | Active, the main replacement
Core question | Who looks like my list? | Who is likely to act like my list?
Main signals | Profile traits: title, industry, company size | Profile traits plus engagement and conversion behavior
Seed sources | Contact lists, company lists, website audiences | Contact lists, company lists, lead gen forms, conversions
Minimum seed | 300 matched members | 300 matched members
Source mixing | Not applicable | One source type per audience
Size control | Limited | Recommended size or your own chosen size
Location | Set at campaign level | Required when building the audience
Updates | Now frozen (static) | Stays in sync with your source data
Audience Expansion | Could be layered | Cannot be used with predictive audiences
Funnel fit | Awareness and reach | Consideration and conversion
[/table]


Are predictive audiences better than lookalikes?

For most B2B lead generation, yes. Predictive audiences are built to find people who are likely to convert, which is what lead gen campaigns need. Early data shared across the industry pointed to roughly 21% lower cost per lead for predictive audiences. One agency, Workshop Digital, reported a 39% lower CPL than its earlier campaigns in its own tests.

Treat these numbers as directional. Your results depend on seed quality, offer, and creative. A predictive audience built from a weak seed will still find weak leads, just more efficiently.

Predictive audiences vs Audience Expansion vs matched audiences

LinkedIn offers three tools that people often mix up. Here is the short version:

• Matched audiences reach people you already know. These are your uploaded lists and retargeting segments.
• Predictive audiences reach new people who are likely to act like your best contacts or converters.
• Audience Expansion loosens your targeting so LinkedIn can show ads to people with similar attributes to your chosen criteria.

[table]
Tool | Who it reaches | Data it uses | Best for | Typical size
Matched audiences | People already in your lists or who visited or engaged | Your first-party data | Retargeting, ABM, nurture | Small to medium
Predictive audiences | New people likely to convert | Your seed plus LinkedIn's AI model | Prospecting for leads and conversions | Medium, adjustable
Audience Expansion | People similar to your targeting criteria | LinkedIn's demographic data | Adding reach to a narrow audience | Varies by campaign
[/table]


What is the difference between predictive audiences and Audience Expansion?

Predictive audiences build a new audience from your own seed data and focus on people likely to act. Audience Expansion is a checkbox that widens the targeting you already set, using profile attributes. 

Predictive is about intent. Expansion is about reach.

The two do not stack. LinkedIn does not let you apply Audience Expansion to a predictive audience. That is by design. Expansion would add people the model did not pick, which would undo the point of prediction.

Should you use matched audiences or predictive audiences?

Use both, for different jobs. Matched audiences are for people who already know you. Predictive audiences are for finding new people who behave like your best buyers. Most strong LinkedIn accounts run matched audiences for retargeting and ABM, and predictive audiences for net-new prospecting. Your matched audiences also become the seeds that make predictive audiences work. Our matched audiences glossary entry covers list types and match rates in more detail.

LinkedIn predictive audiences requirements

To create a LinkedIn predictive audience, you need a Campaign Manager ad account, a role above Viewer, and a source with at least 300 matched LinkedIn members. Here is the full checklist:

[table]
Requirement | Detail
Minimum seed size | 300 matched members (not 300 uploaded rows)
Source types | Contact list, company list, lead gen form, or conversions
Sources per audience | One source type; several sources of the same type can be combined
Location | Required; the audience is built for the geography you choose
Account access | Any ad account role except Viewer
Build time | Usually 24 to 48 hours, sometimes up to about 3 days
Audiences per account | LinkedIn launched with a cap of 30; some newer guides report 100, so check your account
Sharing | Built for one ad account; not shared across accounts
Use in campaigns | Can be used to include or exclude
Audience Expansion | Not available on predictive audiences
[/table]


Why 300 uploaded contacts is not enough

The 300 minimum counts matched members, not rows in your file. Contact lists often match only 30% to 50% of the time, because people sign up with personal emails that are not on their LinkedIn profile. So a list of 500 emails may only produce 200 matches, which is below the limit.

What you can do is upload at least 1,000 contacts if you want a seed that clears 300 matches with room to spare. Company lists match at higher rates, often 60% to 80%, so they are easier to qualify.

How to create a predictive audience on LinkedIn (step by step)

You create a predictive audience in Campaign Manager under Plan > Audiences > Create audience > Predictive audience. The full process takes about five minutes, plus one to three days for LinkedIn to build it.

1. Prepare your seed. Upload a contact or company list, or confirm that your lead gen form or conversion has enough results. Wait until the source shows at least 300 matched members.
2. Open Audiences. In Campaign Manager, select your ad account and go to Plan, then Audiences.
3. Click Create audience and choose Predictive audience.
4. Name it clearly. Use a format your team will understand later, such as "PA | Closed-won 2026 | US".
5. Pick the source type and select the specific list, form, or conversion to use as the seed.
6. Set the location. Choose the countries or regions you plan to target.
7. Choose the size. Accept LinkedIn's recommended size or set your own. For company lists, set how many similar companies to add.
8. Create the audience. LinkedIn will show it as building. Plan for up to three days before launch.
9. Add it to a campaign. In the campaign's audience section, select the predictive audience under your matched audiences.
10. Leave Audience Expansion off, and add exclusions such as current customers, open opportunities, and employees.

Which campaign objectives work best?

Predictive audiences fit best with Lead Generation and Website Conversions objectives, because the model is built to find people likely to act. They can also work for Website Visits or Engagement when you want cheaper warm-up traffic. For pure brand awareness, a broad firmographic audience is often cheaper per impression.

What is the best seed for a LinkedIn predictive audience?

The best seed is a list of people who reached the outcome you want more of, such as closed-won customers or sales-qualified leads. The model copies whatever pattern your seed holds. If the seed is full of students and job seekers who downloaded an ebook, the model will find more of them.

Rank your seed options from strongest to weakest:

[table]
Seed | Strength | Why
Closed-won customers (last 12 to 18 months) | Strongest | Shows who actually buys
Sales-qualified leads or opportunities | Strong | Close to revenue, more volume than customers
Demo or trial sign-ups via Conversions API | Strong | High-intent action, refreshes on its own
Lead gen form submits, filtered for quality | Medium | Volume is high, quality varies
All website visitors or all form fills | Weak | Mixes buyers with researchers and job seekers
Old or purchased lists | Avoid | Teaches the model the wrong pattern
[/table]


Three rules make any seed better:

1. Keep it recent. Buyers change. A seed from three years ago reflects an older market. Refresh list-based seeds at least every quarter.
2. Keep it pure. One seed per outcome. Do not mix customers with webinar sign-ups.
3. Clean it first. Remove competitors, partners, students, and your own employees before you upload.

Can you use the Insight Tag and Conversions API as a seed?

Yes. Conversions tracked by the LinkedIn Insight Tag or the Conversions API (CAPI) can be a predictive seed. This is often the best long-term option, because the seed grows on its own as new conversions come in. CAPI usually captures more conversions than the tag alone, since it is not blocked by browsers or cookie settings. If you can send offline events, like SQLs or closed deals from your CRM, the model learns from revenue, not just form fills.

How to use predictive audiences: 7 strategies that work

The teams that win with predictive audiences treat them as one layer in a full-funnel setup, not as a single magic audience. These seven strategies come from what works across B2B accounts.

Build one audience per outcome. 

Create separate predictive audiences for customers, SQLs, and demo requests. Test them against each other in separate campaigns.

Start with the recommended size, then test smaller. 

Smaller audiences are usually more precise. Larger ones give more reach but drift toward average users. Workshop Digital found its best results at 50,000 to 100,000 members, with larger audiences performing worse.

Do not stack too many filters on top. 

Adding job title, industry, and seniority filters on top of a predictive audience shrinks it and fights the model. Use location plus exclusions, and let the AI do the rest. If you must add a layer, use broad firmographics like company size, not narrow titles.

Always exclude your seed. 

People in your seed are already customers or leads. Exclude them, plus open opportunities and employees, so you pay only for new reach.

Pair predictive prospecting with matched retargeting. 

Use the predictive audience to reach new people with a value-first offer, such as a guide or LinkedIn document ad. Then retarget engagers with a matched audience and a demo or trial offer.

Use predictive audiences inside ABM. 

Build a predictive audience from your best-fit accounts to find similar companies, then reach the whole buying committee at those accounts with personal posts boosted as Thought Leader Ads.

Give it time. 

Let each campaign run two to four weeks before you judge it. Early CPL swings are normal while delivery settles.

Common mistakes with predictive audiences

Most predictive audiences fail for the same few reasons.

• Treating it like a lookalike. 
• Using a seed that barely clears 300. 
• Stacking narrow filters. 
• Forgetting exclusions. 
• Letting the seed go stale.
• Judging on CPL alone. 
• Running one audience for everything. 

How to measure predictive audience performance

Measure predictive audiences on pipeline, not just on platform metrics. A lower CPL is a good sign, but the real test is whether those leads become SQLs and revenue at the same rate as your other channels.

Track these metrics for each predictive audience:

[table]
Metric | What it tells you | Where to find it
CTR | Whether the creative fits the audience | Campaign Manager
Cost per lead | Top-of-funnel efficiency | Campaign Manager
Lead-to-MQL rate | Whether the model finds the right people | CRM
MQL-to-SQL rate | Whether the leads are real buyers | CRM
Cost per SQL | True efficiency compared with other audiences | CRM plus ad spend
Pipeline and revenue influenced | Business impact | CRM attribution
[/table]


Run a simple test to prove value: put the same offer and creative in two campaigns. Target one at a predictive audience and the other at your best manual firmographic audience. Compare cost per SQL after four weeks. 

Use our free CTR calculator to check whether your creative is pulling its weight.

When not to use predictive audiences

Predictive audiences are not the right tool for every campaign. Skip them, or use them with care, when:

• Your seed is small or weak. Under a few hundred good matches, a tight firmographic audience or a matched company list will often win.
• Your total market is tiny. If you sell to 200 named accounts, use account targeting with a matched company list instead.
• You need pure awareness at the lowest CPM. Broad targeting is usually cheaper per impression.
• You are in a regulated space with strict targeting rules. Check whether AI-built audiences fit your compliance needs.
• You have almost no conversions. If your conversion seed has only a handful of events each month, start with lists until volume grows.

If you want a second set of eyes on your setup, our LinkedIn ads agency team builds and tests predictive audience programs for B2B SaaS and AI companies. You can also see results in our case studies.

The Bottom Line 

LinkedIn predictive audiences are more than a replacement for lookalikes. They give B2B teams a way to find people who are more likely to take a specific action, making seed quality and conversion data more important than simply building a large audience.

If you're not sure whether your LinkedIn audiences are reaching the right buyers, book a free audit with ScalixAI to find the gaps and opportunities in your targeting.