LinkedIn Ads best practices for B2B in 2026: where the sources disagree

Cross-checking more than twenty published sources leaves six practices standing, and the most useful finding is how often LinkedIn's own documentation and the agencies that run campaigns on it contradict each other. Everything below comes from those publications; none of it draws on my own campaigns. A practice is on the list only if at least two independent sources assert it, and where they disagree, the article shows the disagreement and leaves it open.

The reading list: LinkedIn's own best-practice, help and specification pages, its Marketing API documentation (hosted on Microsoft Learn), four vendor datasets that publish their samples, trade coverage from Search Engine Land, PPC Land and Social Media Examiner, and blog posts by agencies that sell LinkedIn Ads management, whose commercial interest I point out each time one of their recommendations comes up.

LinkedIn's documentation has a weakness worth keeping in mind throughout: none of its public best-practice pages carries a publication date, and its ad targeting best practices page still tells advertisers to "build a lookalike audience", a feature LinkedIn discontinued on 29 February 2024, as its own help page on the change (last updated in 2026) confirms. Trust that documentation on how delivery works, and check elsewhere before trusting it on what still exists.

If you came for the commercial offer, it's on the LinkedIn Ads service page. The rest of this article stays with what the sources publish, including the places where they argue with each other.

Wide enough to clear LinkedIn's stated floors, and built from two or three targeting facets rather than a stack of them. LinkedIn's ad targeting best practices page sets both rules: "keep the target audience over 50,000 for Sponsored Content and Text Ads, and over 15,000 for Message Ads", and "we recommend not adding more than two to three targeting facets". Factors.ai's targeting guide (28 July 2026) repeats both.

Of all the numbers in this section, these floors are the only ones LinkedIn itself publishes; the others come from someone selling against it.

Source: LinkedIn Marketing Solutions, Best Practices for Ad Targeting, no date on page.

The sources part ways at the top end, where LinkedIn's page gives no ceiling at all. Factors.ai adds one, telling advertisers to "Aim for an audience size between 50,000 and 300,000 members", an upper bound that LinkedIn's own pages don't state. B2Linked leans the same way in its 2026 best practices post (16 July 2026), arguing "bigger isn't always better" and suggesting segmented audiences instead, without putting a number on it. Two sources agree on the floor and the facet count, and two vendors warn against going too broad, but neither produces evidence for a ceiling, which leaves the 300,000 upper bound as one vendor's rule of thumb rather than a benchmark.

No other question splits the sources this sharply, so this section reports the split instead of settling it. LinkedIn recommends both settings: its targeting page tells advertisers to "enable Audience Expansion or build a lookalike audience", and its Audience Network page promises "better return on ad spend and engagement", noting the network is "default enabled for your image, carousel and video format campaigns".

The two agency sources say to switch both off. B2Linked's 2026 post warns that Audience Expansion will "show your ads to people outside the audience you intentionally selected", and GrowthSpree's post on the two settings (last updated 30 September 2026) calls them settings that "can quietly waste 20-30% of your budget", claiming the Audience Network "lowers CPM by 30-40% but reduces lead quality 50%+ on conversion-focused campaigns".

Treat those percentages as claims, not data: neither page states a sample size, a period or a measurement method, and both publishers sell the service of implementing their own advice. The weaker claim underneath does hold up. Two independent practitioners report that the defaults widen delivery beyond the selected audience, and LinkedIn doesn't dispute the mechanism; since switching either setting off takes one click, testing it costs next to nothing. Build the number from your own account rather than adopting anyone's percentage.

Run several creatives per campaign and refresh them every fortnight; on video, style counts for more than production budget. LinkedIn gives the count on two separate pages: its Sponsored Content tips advise advertisers to "Include 4-5 ads in each campaign", its video ad tips repeat "Create 4-5 ads in your campaign", and the first of those pages sets the rhythm: "Every 1-2 weeks, pause the ad with the lowest engagement and replace it with new ad creative." B2Linked's founder, AJ Wilcox, whose agency sells LinkedIn Ads management, argues for fewer: in a Social Media Examiner interview (2 April 2026) he says he runs two ads at a time, three at most, so every creative gets enough impressions to show what's working.

On video, LinkedIn is specific: show what matters "in the first 10 seconds", write for sound-off viewing, keep videos short and test longer ones for demand generation. For what moves engagement, the strongest evidence is a LinkedIn Creative Labs analysis of more than 13,000 B2B video ads, reported by Search Engine Land and PPC Land on 16 July 2025. In the table, one of the four reported lifts is for dwell time and the other three are for engagement.

Source: Search Engine Land and PPC Land, both 16 July 2025, reporting LinkedIn Creative Labs.

PPC Land doesn't link to the Creative Labs study itself, so the figures are credited here to the trade coverage that carries them. LinkedIn's own completion figure calls for more caution still: its video page says videos of 7 to 15 seconds see "up to a 300% lift in completion rates", with "LinkedIn data, 2025" as the only source and no sample or method behind it.

Send them twice, server-side and through the tag, then tell LinkedIn which ones turned out to be qualified. The LinkedIn Marketing API documentation on Microsoft Learn (last updated 12 August 2026) is unambiguous: "we recommend that advertisers send conversions through both the Insight Tag as well as Conversions API for increased reliability." Of the LinkedIn-owned sources cited here, it's also the only one with a full date on it.

The same page sets out the conditions under which qualified-lead optimization works. Read the five-leads row first, because it decides whether the rest applies to you.

Source: Microsoft Learn, LinkedIn Marketing API, Conversions API Use Cases, last updated 12 August 2026.

Three more sources back the practice, though none adds to the mechanics. Search Engine Land's B2B LinkedIn tests piece (11 March 2026) says to "sync your CRM data and provide the platform with your definition of a qualified lead", and PPC Land reported Qualified Leads Optimization arriving on 11 January 2025. Dreamdata's 2025 benchmarks summary (5 March 2025) relays a LinkedIn figure: CAPI users see "a 20% lower CPA and a 31% increase in attributed conversions, compared to non-integrated users". It is LinkedIn's own number, published without a method. Plan around the five-lead threshold: an account that cannot produce five qualified leads a fortnight has a volume problem before it has an optimization problem.

The buying group, not the single decision-maker, and mostly people who are not buying yet. Edelman's 2025 B2B Thought Leadership Impact Report, fielded among 1,934 global executives between 17 March and 3 April 2025 with a stated margin of error of 2.0%, is the best-documented survey cited here. It finds "64% of hidden decision-makers spend more than an hour on average per week consuming thought leadership" while "71% say they have relatively little or no interaction with sales". LinkedIn co-sponsors it and sells the channel it favors.

6sense's 2025 B2B Buyer Experience Report, built on close to 4,000 buyer responses, points the same way: "94% of buying groups put their shortlist in order of preference before engaging with sellers". Hootsuite's LinkedIn ads guide (21 July 2026), quoting a LinkedIn product marketing manager, gives the operational version: "Focus on targeting the whole buyer group that will influence a purchase, not just executive decision makers."

This usually gets framed with the 95-5 rule, which deserves more care than it gets. LinkedIn's B2B Institute page states that "95% of your potential buyers aren't ready to buy today", and the Ehrenberg-Bass summary restates it, but neither names the originating paper or its year. It is also a different claim from 6sense's shortlist finding, though the two get merged constantly. The evidence does support the practical conclusion: most of the audience is not in market, so asking a cold audience for a sales meeting asks the wrong thing of the wrong people.

Longer than a quarter, and Dreamdata's data shows why. Its 2026 LinkedIn Ads benchmarks announcement, released on 10 March 2026 per its PR Newswire release and built on more than 66 million sessions across more than 3.5 million customer journeys, puts numbers on the lag. The gap between the two timings tells you more than either one on its own.

Source: Dreamdata, LinkedIn Ads 2026 Benchmarks announcement, March 2026.

HockeyStack's benchmark report, drawn from three years of data across more than 70 B2B SaaS companies, finds the same shape seasonally: Q1 has the shortest MQL conversion cycle at 24 days on average, while Q3 delivers 6.01x spend-to-pipeline return. The 121% figure is the one our LinkedIn Ads benchmarks article sets against the competing platforms. Both vendors report their own customers' data rather than a random sample, and both sell software that benefits from advertisers measuring longer.

Every source here states the resulting practice in some form: measure cost per qualified opportunity and pipeline contribution rather than cost per lead, and match the evaluation window to the buying cycle instead of the reporting month. That is the premise of our methodology, and no other practice in this article has as little disagreement behind it.

Start with measurement, then creative, and leave the settings for last, because that is the order in which the evidence gets weaker. LinkedIn documents the Conversions API setup and the qualified-lead feedback loop on a dated page, and both trade reporting and a vendor dataset back it up. Creative volume, refresh cadence and video construction appear on several LinkedIn pages, and LinkedIn's own Creative Labs analysis of more than 13,000 video ads points the same way, so most of that evidence comes from the platform itself. The default settings sit at the bottom, supported only by two commercially interested agencies whose figures carry no method.

The audience practices sit in between, and they're where following the platform costs least, since LinkedIn's floors are mechanical facts about delivery, not opinions. Where a practice rests on one publisher, the article says so; where LinkedIn and its practitioners contradict each other, you get both positions.

Whether LinkedIn is the right channel for a given program is a comparison rather than a practice, and nothing here settles it. For the demand-capture side of a B2B plan, read our Google Ads best practices article; the Google Ads service page covers why the two work best as a pair.

How many ads should a LinkedIn campaign contain?

LinkedIn states 4 to 5 on both its Sponsored Content and video best-practice pages, while B2Linked's founder runs two at a time and three at most (Social Media Examiner 2026). LinkedIn repeating its own number on two pages is still a single source, so neither figure has independent backing. Start from LinkedIn's range and treat the smaller set as a test.

Should the LinkedIn Audience Network be turned off for B2B?

LinkedIn and the agencies disagree: the platform recommends leaving it on and enables it by default for image, carousel and video campaigns, while two agencies that sell LinkedIn Ads management recommend disabling it. Neither agency publishes a method behind its numbers, so measure the difference in your own account.

What does LinkedIn document about the Conversions API?

That it should run alongside the Insight Tag rather than instead of it, and that sharing five or more qualified leads within two weeks speeds up a two-week learning phase, with a 30-day window for leads to enter optimization (Microsoft Learn 2026). Check your lead volume against that threshold first.

How long should a LinkedIn program run before it is judged?

Longer than most reporting cycles allow. The published figures put 281 days between first ad impression and revenue and 214 days between first ad conversion and revenue (Dreamdata 2026), so a monthly cost-per-lead review measures the wrong thing. Match the review window to the length of the buying cycle.

Is there a reliable minimum audience size for LinkedIn Ads?

There's a documented floor but no documented ceiling. LinkedIn states 50,000 for Sponsored Content and Text Ads and 15,000 for Message Ads, while the 300,000 upper bound circulating in vendor guides doesn't appear on LinkedIn's own pages. Use the floor and ignore the ceiling.