The honest summary is that peer-reviewed evidence for social selling exists, it is weaker than the marketing claims, and the score everyone optimises has no published formula.

That is a more useful position than either of the two available extremes. It is not true that nothing has been studied: there is a real academic literature in marketing journals, with samples and statistics. It is also not true that the return is established, because every study located shares the same design limitation.

Meanwhile the most quoted statistic in the field could not be found on any current page belonging to the platform it is attributed to.

What the research actually shows, and what it cannot

Three peer-reviewed studies are worth naming, and one design characteristic disqualifies all of them from supporting the claim usually made.

A 2023 study in a business marketing journal. 196 B2B salespeople, cross-sectional survey, analysed with structural equation modelling.

A 2021 study in an industrial marketing journal. 171 B2B salespeople.

And a 2020 study in a marketing science journal, widely cited in the same literature.

What they establish. That a salesperson’s use of social media is associated with their performance, in samples of salespeople, using statistical models appropriate to survey data.

And the limitation they share. All are cross-sectional surveys in which salespeople report their own social media use, and in which performance is generally self-assessed rather than taken from a revenue system.

Which constrains the claim you can make from them. They support a correlation between reported social media use and reported performance. They do not establish that social selling causes revenue, and none of them is an experiment.

One further consequence of self-assessment. Salespeople who invest effort in a channel are being asked to rate both the effort and the outcome. That is not a reason to dismiss the finding, but it is a reason not to convert it into a return figure.

The honest sentence for a plan. There is academic evidence that a salesperson’s social media use correlates with their self-reported performance. There is no experimental evidence of a causal effect on revenue, and the return figures in circulation come from the platform or from tool vendors.

What the peer reviewed literature on social selling establishes and what it cannot establishWhat the peer reviewed academic literature on social selling establishes, and the shared design limitation that constrains what can be claimed from it. Three studies are relevant. A 2023 study published in a business and industrial marketing journal surveyed one hundred and ninety six business to business salespeople in a cross sectional design analysed using structural equation modelling. A 2021 study published in an industrial marketing management journal surveyed one hundred and seventy one business to business salespeople. A 2020 study published in a marketing science journal is widely cited within the same literature. What these studies establish is that a salesperson’s use of social media is statistically associated with that salesperson’s performance, within samples of salespeople, using analytical methods appropriate to survey data. The limitation they share is that all are cross sectional surveys in which the salespeople themselves report their own social media use, and in which the performance measure is generally self assessed by the same respondent rather than extracted from a revenue system. This constrains the permissible claim to a correlation between reported social media use and reported performance. None of the studies establishes that social selling causes revenue, and none is an experiment with random assignment. A further consequence of self assessment is that a salesperson who has invested substantial effort in a channel is being asked to rate both the effort expended and the outcome obtained, which is not grounds for dismissing the finding but is grounds for declining to convert it into a return on investment figure. The defensible formulation for a commercial plan is therefore that academic evidence exists showing a salesperson’s social media use correlates with their self reported performance, that no experimental evidence of a causal effect on revenue exists, and that the return figures in circulation originate with the platform itself or with tool vendors.There is a literature. Here is what it carries.What exists2023, n = 196 B2B salespeople,structural equation modelling2021, n = 171 B2B salespeople2020, widely cited in the same fieldWhat they shareAll cross-sectional surveysSalespeople report their own usePerformance generally self-assessedNone is an experimentWhich fixes what you can sayA correlation between reported use and reported performance. Not a causal effect on revenue, and not a return figure.The sentence to put in a plan”There is academic evidence that a salesperson’s social media use correlates with their self-reported performance.There is no experimental evidence of an effect on revenue.”
Real samples, real statistics, one shared design limitation. Correlation between two self-reported quantities. Source : Franck and Dampérat, Journal of Business and Industrial Marketing 38(8), 2023; Bowen and others, Industrial Marketing Management 96, 2021 (2023)

The index with no formula

The platform publishes a score that a great many sales teams manage against. What it publishes about that score is a list of inputs.

The inputs are named, across four areas. Profile completeness and endorsements. Connections, including “the acceptance rate of your connection requests”. Activity, covering “shares, likes, comments, and reshares”, messages sent and response rates, and group participation. And research behaviour, covering “people searches”, “profile views”, “days active”, plus saved leads and accounts in the paid tool.

What is not published. How the components are weighted. How the four areas combine. How often the score recalculates. And how any amount of raw activity maps onto the hundred-point scale.

Which means the score cannot be reasoned about. You can raise it by doing more of the listed things, without knowing which one moved it or by how much, or whether the movement corresponds to anything outside the score.

Note also what half the inputs measure. Connection acceptance rate, people searches, profile views and days active are measures of platform usage. A score that rewards time on the platform is not a neutral instrument for deciding how much time to spend on the platform.

And the performance claims published alongside it carry no methodology. Claims include that users of the paid tool “save 65 hours annually”, that “sellers with at least four LinkedIn connections at a target account are 16% more likely to close a deal”, that personalised messages “can increase acceptance rates by 40%”, and a “312% ROI over 3 years”. None states a sample size, a study period or a method.

One statistic deserves a specific flag. The best-known social selling figure, asserting 45 percent more opportunities and a 51 percent greater likelihood of reaching quota, was not found on any current page belonging to the platform. It circulates through agency material. Ask for the original before repeating it.

Published inputs and undisclosed methodology of the social selling indexThe inputs published for the social selling index published by the principal professional network, and the methodological elements that remain undisclosed, together with the character of the inputs themselves. The published inputs fall across four areas. The first area covers profile completeness and endorsements received. The second covers connections, explicitly including the acceptance rate of connection requests sent. The third covers activity, comprising shares, likes, comments and reshares, messages sent together with response rates to those messages, and participation in groups. The fourth covers research behaviour, comprising searches for people, profile views performed, days active on the platform, and within the paid product the leads and accounts saved. What is not published anywhere is how the individual components are weighted relative to one another, how the four area scores combine into a single figure, how frequently the score is recalculated, and how any given quantity of raw activity maps onto the hundred point scale. The hundred point scale and the four pillar structure appear exclusively on marketing pages rather than in any methodological documentation. The consequence is that the score cannot be reasoned about, since it can be raised by performing more of the listed activities without any means of establishing which activity moved it, by what magnitude, or whether that movement corresponds to any quantity outside the score itself. A further observation concerns the character of the inputs, since connection acceptance rate, people searches, profile views and days active are all measures of platform usage rather than of commercial outcome, meaning a score which rewards time spent on the platform is not a neutral instrument for deciding how much time to spend on the platform. Separately, the performance claims published alongside the index carry no sample size, study period or method, including claims of hours saved annually, a percentage increase in likelihood of closing associated with a number of connections at a target account, a percentage increase in message acceptance rates, and a three year return on investment percentage.The ingredients, without the recipePublished: the inputsProfile completeness, endorsementsConnections, and request acceptance rateShares, likes, comments, resharesMessages sent, and response ratesPeople searches, profile views, days activeNot published: any of the methodHow components are weightedHow the four areas combineHow often it recalculatesHow activity maps to the 0 to 100 scaleThe scale itself appears only on marketing pages.And look at what half the inputs measureAcceptance rate, people searches, profile views, days active. These are measures of time on the platform.Which is the problem in one sentenceA score that rewards time on the platform is not a neutral instrument for deciding how much time to spend on it.And without weights, you cannot tell which action moved it or whether the movement means anything.
Naming the ingredients is not publishing a recipe. And half the ingredients measure time spent on the platform. Source : LinkedIn Help, Social Selling Index, and LinkedIn Sales Solutions pages (2026)
Published social selling performance claims and the methodological information accompanying eachFour performance claims published by the platform in connection with its social selling product, together with the methodological information accompanying each of them, and one further claim whose source could not be located. The first claim states that users of the paid sales product save sixty five hours annually. The second states that sellers holding at least four connections at a target account are sixteen percent more likely to close a deal. The third states that personalised direct messages can increase acceptance rates by forty percent. The fourth states a return on investment of three hundred and twelve percent over three years. None of these four claims is accompanied by a sample size, a study period, a description of method, or a link to an underlying research document, and several link only to other marketing pages belonging to the same platform. A fifth claim, and the most widely circulated statistic in the field, asserts that social selling leaders create forty five percent more opportunities and are fifty one percent more likely to reach quota. That claim could not be located on any current page belonging to the platform, and circulates through agency and consultancy material rather than through platform documentation, and should accordingly be treated as unsourced until an original publication is produced. The general pattern is that the quantitative case for the practice is made through figures that carry no methodological accompaniment, in contrast with the peer reviewed academic literature which publishes samples and statistical methods but which supports only a correlation between self reported activity and self reported performance rather than any causal effect on revenue.Four claims, no methods”Save 65 hours annually”no sample, no period”16% more likely to close a deal”no sample, no period”Increase acceptance rates by 40%“no sample, no period”312% ROI over 3 years”no sample, no periodAnd the most famous one is not on the site”45% more opportunities, 51% more likely to reach quota” was not found on any current page belongingto the platform. It circulates through agency material.The pattern: the quantitative case comes with no methodology, while the academic literature comes withmethodology and supports only a correlation.
Four figures, no sample sizes, no periods, no methods. And one famous statistic that is not on the site at all. Source : LinkedIn Sales Solutions pages, and a search of current LinkedIn-owned pages for the quota statistic (2026)

Who is actually there, which is the question worth answering

If the return cannot be established, the audience composition can, and it is a better basis for the decision.

The best available measurement, with its methodology published. A national survey of 5,733 US adults, fielded 19 May to 5 September 2023, margin of error plus or minus 1.8 percentage points. 31 percent of US adults reported using the platform.

And the finding the researchers flagged themselves. “53% of Americans with at least a bachelor’s degree report using the platform, far higher than among those who have some college education (28%) and those who have a high school degree or less education (10%). This is the largest educational difference measured across any of the platforms asked about.”

Which is the useful sentence for a channel decision. Not how far a post travels, but who is present. If your buyers hold degrees and work in offices, they are disproportionately present. If they do not, they are disproportionately absent, and no amount of posting changes that.

One caveat on currency. The same organisation removed the platform from its 2025 survey entirely. There is no 2025 figure, and its interactive summary page can produce apparently current numbers that are artefacts. The last real measurement ends with fieldwork in September 2023.

And a related absence worth knowing. The platform has never published an audited user count. Its parent company’s most recent annual report mentions it eighteen times, all concerning revenue, segments or litigation. Its public “one billion members” is a cumulative registration total.

The measured composition of the professional network’s United States audience by education levelThe measured composition of the professional network’s United States audience by level of formal education, drawn from the most methodologically sound source available on the subject. The survey was conducted with a sample of five thousand seven hundred and thirty three United States adults, with fieldwork running from the nineteenth of May to the fifth of September 2023, and a stated margin of error of plus or minus one point eight percentage points at fifty percent incorporating the design effect. Thirty one percent of United States adults reported using the platform. The composition by education is markedly uneven. Fifty three percent of Americans holding at least a bachelor’s degree reported using the platform. Twenty eight percent of those with some college education reported using it. Ten percent of those with a high school degree or less education reported using it. The research organisation itself flagged this as exceptional, stating that this is the largest educational difference measured across any of the platforms asked about. The consequence for a channel decision is that the relevant question is not how far a post travels but who is present to see it, so that where a company’s buyers hold degrees and work in office environments they are disproportionately present on the platform, and where they do not they are disproportionately absent, a condition no amount of posting activity can alter. Two caveats attach. The same research organisation removed the platform from its 2025 survey battery entirely, so no 2025 figure exists and its interactive summary page can generate apparently current figures that are artefacts of stale labels paired with newer values, meaning the most recent genuine measurement concluded with fieldwork in September 2023. Separately, the platform has never published an audited user count, its parent company’s most recent annual report referring to it exclusively in connection with revenue, segment reporting and litigation, and its public figure of one billion members being a cumulative count of registered accounts.Who is present, measured properlyBachelor’s degree or more53%Some college28%High school or less10%n = 5,733 US adults, fielded 19 May to 5 September 2023, margin of error ±1.8 points.Flagged as exceptional by the researchers themselves”This is the largest educational difference measured across any of the platforms asked about.”The useful questionNot how far a post travels. Who is thereto see it.One currency caveatThe platform was dropped from the 2025survey. There is no 2025 figure.And its “one billion members” is a cumulative registration count, never audited by anyone.
A probability sample with a published margin of error. And the widest education gap the researchers had measured on any platform. Source : Pew Research Center, Americans' Social Media Use, published 31 January 2024 (2024)

Where the cost actually sits

Social selling is usually presented as free. It is not free, it is unbudgeted, and the difference is the point.

The recurring cost is salesperson hours. Writing, commenting, connecting and messaging are not residual activities. Whatever time they take is time not spent on calls, proposals or existing accounts.

And that time is the most expensive input available. It is the same hour that would have gone into a conversation with a buyer already in a process.

The tooling cost is visible and usually the smaller one. A paid seat, a scheduling tool, sometimes a writing service. Easy to see, easy to approve, and rarely the largest line.

The hidden cost is attribution. Social activity produces effects that arrive months later through channels nobody instruments, which means the programme is defended on faith or on a vendor’s figure. Both are unstable positions in a budget conversation.

And there is a cost to the wrong scoreboard. A team managed on an index that rewards platform activity will produce platform activity. That is what a score does.

One framing worth knowing, and worth citing correctly. The observation that a small share of buyers is in market at any moment is an arithmetic deduction rather than a measurement: if companies change supplier roughly every five years, a fifth are in market annually, so a twentieth in a quarter. It is a useful way to think about patience. It is not a count of anybody.

What to measure, given all that

The absence of a reliable external benchmark makes the internal measurement more important, not less.

Hours, honestly recorded, for four weeks. Per person, per week, including writing time. Most teams discover the number is larger than they assumed and smaller than it needs to be to matter.

Conversations that name it. Whether a prospect mentions a post, an article or a profile. Rare, countable, and the only direct evidence available.

Inbound messages received rather than sent. A message you did not initiate is a different quantity from a connection request that was accepted.

And meetings sourced, with the source recorded at the meeting rather than reconstructed later. Asking “how did you come across us” at the start of a call produces better attribution than any model.

What not to measure. The index, because you cannot interpret a movement in it. And engagement rate against any published average, because no comparable average exists.

One decision this makes easier. If four weeks of honest hours produce no conversation that names the channel, that is not proof it does not work. It is proof you cannot yet defend it, which is a different and more actionable problem.

The cost structure of a social selling programme and the corresponding measurementsThe cost structure of a social selling programme and the measurements that correspond to it, given the absence of any reliable external benchmark. The recurring cost consists of salesperson hours spent writing, commenting, connecting and messaging, which are not residual activities and which displace time that would otherwise be spent on calls, proposals or existing accounts, making them the most expensive input available since the same hour would otherwise have gone into a conversation with a buyer already in a purchase process. The tooling cost, comprising a paid seat, a scheduling tool and occasionally a writing service, is visible, easily approved and rarely the largest line item. The hidden cost is attribution, since social activity produces effects arriving months later through channels that are not instrumented, meaning the programme is defended either on faith or on a vendor supplied figure, both of which are unstable positions in a budget discussion. A further cost arises from managing to the wrong scoreboard, since a team managed against an index that rewards platform activity will reliably produce platform activity, that being the function of a score. The corresponding measurements are four. First, hours honestly recorded per person per week over four weeks including writing time, which most teams find to be larger than assumed and smaller than necessary to matter. Second, conversations in which a prospect names a post, an article or a profile, which are rare, countable and constitute the only direct evidence available. Third, inbound messages received rather than sent, since a message the seller did not initiate is a materially different quantity from an accepted connection request. Fourth, meetings sourced with the source recorded during the meeting rather than reconstructed afterwards, since asking how the person came across the organisation at the start of a call produces better attribution than any model. What should not be measured is the index itself, because a movement in it cannot be interpreted, and engagement rate against a published average, because no comparable average exists.The cost is hours. The output is conversations.Where the cost sitsSalesperson hours, the expensive inputTooling, visible and usually smallerAttribution, defended on faithAnd managing to the wrong scoreboardWhat to count against itHours, recorded honestly for four weeksConversations that name the channelInbound messages received, not sentMeetings sourced, asked at the meetingWhat not to measureThe index, because a movement in it cannot be interpreted. And engagement rate against an average that does not exist.And what a null result actually meansFour weeks of hours with no conversation naming the channel is not proof it does not work. It is proof you cannotyet defend it, which is a different and more solvable problem.
The expensive input is salesperson hours. The measurable output is conversations that name the channel. Source : Method (2026)

What to do with this

Record the hours for four weeks before deciding anything. Per person, per week, writing time included. That number is the entire cost side of the argument and almost nobody has it. Once it is on the table it can be set against what comparable reach costs to buy, and our page on B2B paid acquisition gives the other side of that comparison in cost per lead.

Then add one question to the start of every discovery call: how did you come across us. Recorded at the meeting rather than reconstructed from a model, it produces attribution nobody can dispute.

Stop managing to the index. Half its inputs measure time on the platform, and none of its weights are published, so a rising score tells you that somebody was busy.

And when a proposal quotes the 45 percent figure or the 51 percent figure, ask where it was published. It is not on any current page belonging to the platform it is attributed to.

The related pieces are organic social benchmarks do not exist and does a white paper still work.