Influencer marketing measurement: who does your survey select?

A business wants to understand how creator content fits into shopping. One researcher proposes interviewing people who recall buying something after seeing a creator post. Another proposes interviewing people who say creators influence their purchase decisions. Both sound like reasonable ways to find the same audience. The paired survey answers show why the choice deserves a closer look.

Pew Research Center had already reported the two measures separately: about 30% of social-media users said they had purchased after a creator post, while 39% said creators affected their purchase decisions at least a little. Our analysis keeps each person’s two answers together. The resulting overlap adds information that those separate percentages do not contain. ^[2]

The purchase question selected 30.2% of the weighted social-media-user group. The general influence question selected 38.6%. The share selected by both was 24.8%. Another 5.4% were selected by the purchase question alone, and 13.8% by the influence question alone. The remaining 56.0% were selected by neither rule. ^[3]

These four groups describe eligibility under two possible interview rules. Under either rule, only an explicitly positive answer qualifies. The remaining answers include uncertainty and refusal, which stay separate in the underlying table. “Selected by neither” therefore does not mean everyone in that group explicitly rejected both propositions.

The difference has a practical consequence before anyone writes an interview guide: choosing a different screening rule changes who would be invited. A larger eligible group is not simply the smaller group with more people added. This comparison applies two definitions retrospectively to the same responses; it is not a randomized test showing that a wording change caused different answers.

Looking only at the separate totals, a team might say the influence question produces a group 8.4 percentage points larger than the purchase question. That is the net difference between 38.6% and 30.2%. It does not describe every person whose eligibility changes.

Start with the purchase rule, then substitute the influence rule. The share selected by both, 24.8%, stays eligible. The 5.4% selected only by the purchase rule leaves the group. The 13.8% selected only by the influence rule joins it. Added and removed together account for 19.2% of the eligible social-media users in this 2022 US survey on a weighted basis. ^[4]

The arithmetic is easier to see on a scale of 100: retain about 24.8, remove about 5.4 and add about 13.8. These are percentage equivalents, not fractional respondents or a forecast for the next 100 customers. The calculation describes the historical survey, where the two questions were answered by the same people.

Unequal totals make some difference unavoidable. Even if every person selected by the purchase question also passed the influence question, the larger group would still have an extra 8.4 percentage points. The paired answers show 19.2 points of different selection, or 10.8 points beyond that minimum. This is an arithmetic comparison, not a significance test. ^[4]

For a business, that means a report should show both the people added and the people left out. Reporting only the larger total can conceal the loss of respondents who had the particular experience the interview was meant to explore.

If the starting group is people who reported a purchase after a creator post, 82.1% also said creators affected their purchase decisions at least a little. This answers: how much of the purchase group would remain if the team also required reported influence?

Reverse the starting point and the answer is 64.3%. Among people reporting at least some influence, that share also recalled a purchase after a creator post. The same overlap is divided by a larger starting group. Both percentages are correct, and neither is a general measure of influencer marketing effectiveness. ^[3]

Use the first direction when a study already requires purchase experience and the team is considering an additional influence question. Use the second when a study starts with perceived influence and needs to know who can discuss a recalled purchase. The denominator belongs beside the number because it determines the decision that number can support. For some exploratory interviews, an overlap like the 82.1% observed here may be sufficient to use one screening question, as long as the team accepts what that rule leaves out.

There is no reason to expect the two questions to be identical. One asks about an event at any time in the past. The other asks how much creators affect purchasing decisions generally. A past purchase can coexist with little general influence, and perceived influence can exist without a recalled purchase. These are possible interpretations, not explanations established by this analysis.

A follow-up interview can explore the difference without telling someone what their answer is supposed to mean. Ask what purchase they had in mind, what they remember about the post, or what kinds of decisions they considered. Avoid turning a difference between answers into an accusation that someone is confused or concealing influence.

The purchase question offered a “not sure” response. Some respondents also declined to answer one or both questions. Together, uncertain or refused answers on at least one item represent 7.2% of the weighted social-media-user group. They remain in the primary analysis. There were no system-missing answers to these two items among eligible respondents. ^[5]

That choice matters most when starting with the influence group. Of people reporting at least some influence, 64.3% reported a purchase, 28.9% said they had not made one, and 6.7% were unsure or did not answer the purchase question. The last group is neither a confirmed purchase nor a definite no.

We also calculated a defined sensitivity check using only people who answered Yes or No about a purchase and A lot, A little or Not at all about influence. This retains 92.8% of the weighted eligible group. Within that restricted group, the purchase share among people reporting influence rises from 64.3% to 69.0%. The reverse percentage changes from 82.1% to 82.4%. ^[5]

The higher percentage does not show that the restricted calculation is more accurate. It answers a question about a group from which uncertainty and nonresponse have been removed. If a team quietly drops those answers, its summary can look different even though no respondent changed what they said.

For an interview project, decide how to handle uncertainty before sending invitations. Someone who is unsure may be unsuitable for an interview about one specific purchase but useful for understanding what people remember about creator content. That is a choice about the research aim, not a reason to relabel an uncertain answer.

For an interview about a past purchase, begin by asking for the experience the person will discuss. A general statement that creators influence their decisions does not establish that they can recall buying after a post. The purchase question is closer to that recruitment aim, although the survey’s “ever” time frame may still be too broad for a project about a recent launch.

For an interview about how creator content shapes product consideration, a purchase requirement may be unnecessary. People who report influence without recalling a purchase may have relevant experiences to describe. The survey does not reveal whether they discovered a brand, compared products, delayed a decision or did something else. Those are possible interview topics, not findings we can assign to this group.

For a comparison of experiences, asking both questions lets the team intentionally include different combinations. The recruitment brief can distinguish people who report both, only one, or uncertainty. It need not assume that one group is more truthful, more valuable or more likely to buy in the future.

Requiring both positive answers would select 24.8% of the weighted survey group. Accepting either would select 44.0%. The first rule suits a project that needs both reported experiences; the second allows a broader comparison. Those are different research aims. Neither percentage tells a team which choice is best without knowing what the interviews need to address.

An influencer preparing an audience summary faces the same wording choice. A statement that some audience members recall a purchase after a post should remain a purchase-recall statement. A statement about perceived influence should retain its own question and denominator. Neither can be converted into the creator’s sales contribution simply by placing it in a media kit.

These distinctions help keep audience research separate from campaign reporting. A question about any creator at any time does not identify the creator, campaign, spend or sales that a business would need for an ROI calculation. If the job is to evaluate a specific campaign, its scope and outcome evidence must be defined separately.

Imagine a skincare business preparing interviews before its next creator partnership. The team wants to hear how customers moved from seeing a product demonstration to deciding whether to try the product. This is a hypothetical example of using the distinction, not a business or campaign observed in the Pew survey.

If the team specifically needs accounts of completed purchases, it could ask whether someone recalls buying after seeing a creator discuss the product category, then invite them to describe the episode. It should specify a relevant recall period instead of assuming the survey’s “ever” wording meets its needs. Changing the category or time frame creates a new question, so the Pew percentages do not predict its recruitment yield.

If the team instead wants to understand consideration, it could invite people who say creator content affected what they considered, including those who did not buy. The discussion would then explore what affected their decision and what happened next. It would not begin by treating every participant as a purchaser.

The important choice is which conversation the business needs. Asking both questions can preserve the distinction while allowing the interviewer to follow an unexpected answer. It does not require the team to manufacture equal-sized groups or copy the survey’s proportions into its own sample.

A short brief can capture the decision: “We want to understand [the experience or decision]. We will invite people who [have the required experience or report the relevant influence], within [the relevant category and period]. We will retain uncertain answers separately and use them for [a stated interview purpose, if any]. We will describe the findings in terms of [the questions actually asked], without treating them as attributed campaign sales.”

In the hypothetical skincare project, the final report should say whether it describes recalled purchase experiences, perceived influence or both. That small piece of wording makes the interviews easier for the next person to interpret, whether they are a business owner choosing a brief or a creator presenting audience research.

We analyzed Wave 111 of Pew Research Center’s American Trends Panel. Fieldwork ran July 5–17, 2022, through self-administered online surveys in English and Spanish. The target population was noninstitutionalized US adults aged 18 and older, including Alaska and Hawaii. The released file contains 6,034 respondents, of whom 4,628 reported using social media and were eligible for both questions. All 4,628 remained in the primary denominator. ^{[1] [6]}

The wave combined 4,996 American Trends Panel respondents with 1,038 lesbian, gay or bisexual respondents from Ipsos’ KnowledgePanel. Its sampling design also oversampled specified groups within the American Trends Panel. We used the supplied final wave weight, WEIGHT_W111, which accounts for selection and nonresponse through several adjustments and population benchmarks. An unweighted headcount would not reproduce this survey’s weighted estimates. ^[6]

Before calculating the joint results, the method fixed the two questions, all original answer combinations, the two positive-answer rules, both conditional directions and the sensitivity check. The purchase rule uses Yes. The influence rule combines A lot and A little. Not sure, refusal and any system-missing values remain distinct in the underlying table; they never become a substantive No.

The source checks confirmed the released row count, eligible group, allowed codes, routing and positive finite weights. The separate weighted margins reproduced Pew’s published whole-percentage results. All respondents outside the social-media-user group had structural blanks on the two items and were excluded from its denominator. The public release describes limited changes to certain values for confidentiality; we made no demographic or individual-level claims.

Each result is a sum of respondent weights in the relevant category divided by the sum of weights in its stated group. We report descriptive weighted percentages. The public file and accompanying instructions do not establish an exact variance procedure for these joint estimates, so we do not add confidence intervals, significance tests or the full survey’s margin of error to them. Sampling uncertainty, recall and question interpretation still matter. ^[6]

Independent review examined the original analysis code and independently invented test records, including unequal weights, uncertain answers and reversed denominators. It did not independently reanalyze the confidential respondent data. The download contains the limited paired aggregates, definitions, source hashes and reproduction code; it does not redistribute respondent records, identifiers or weights. ^[7]

The opinions expressed herein, including any implications for policy, are those of the author and not of Pew Research Center.