Separating Brand Lift from Click-Through Optimization: A Protocol for Measuring Attitude Change and Purchase Intention in Advertising Exposure

1. Problem statement and scope

Advertising systems are often judged by click-through rate because it is easy to observe, aggregate, and optimize in real time. That practice is inadequate for brand research. Click-through measures a discrete response to a prompt; brand lift concerns a change in evaluation, preference, or intention. A campaign may maximize clicks through curiosity, friction, or incentive design while leaving brand attitude unchanged. It may also improve brand perception in contexts where clicking is unlikely, unnecessary, or socially discouraged. This protocol therefore separates click optimization from brand-lift estimation and limits inference to brand attitude, sponsor-brand response, and purchase intention. The scope excludes claims about sales incrementality, long-run loyalty, or conversion unless they are separately measured. [1] [2] [3]

The research question is: how can a study estimate brand lift as evaluative change while preventing click-through from functioning as a proxy? The answer proposed here is a measurement architecture that combines pre-exposure and post-exposure brand judgments with weighted sentiment indicators and an exposure model that accounts for differential information behavior. The protocol is intended for advertising contexts in which format, appeal, and audience predispositions affect response. It is not a systematic review; it is a targeted narrative synthesis of directly relevant evidence and a transferable measurement design. [4] [5] [1]

2. Targeted narrative evidence base

The most direct application evidence comes from experimental and survey-based work showing that response to advertising depends on format, message content, and audience predispositions. One randomized experiment found that brand placement and conventional television advertising do not produce identical brand-attitude patterns: the placement advantage over a 30-second advertisement appeared among viewers with low attitude toward advertising, while no statistically significant difference between formats was detected at mean or high levels of that moderator. Another line of research reported that advertising appeal interacts with brand maturity in shaping purchase intention, with the direction of the effect depending on whether the brand is new or mature. A third study linked perceived social-media advertising characteristics to brand loyalty, purchase intention, and sharing intention through psychological-need satisfaction, with aspects of the platform environment acting as moderators. In sponsorship settings, a field experiment showed that counter-advertising can change sponsor-brand attitudes, purchase intentions, and perceived sponsor-event fit, while also shifting some alcohol-in-general outcomes. Taken together, these studies support the premise that brand lift must be measured as a brand-specific evaluative outcome rather than inferred from engagement volume or click behavior. [1] [6] [3] [2]

Methodological foundations transferable to this topic come from two additional sources. One proposes a Brand Polarity Score weighted by tweet engagement and author influence, together with event-sensitivity, convergence, and stability checks. The other develops a duopoly model in which consumer information search and experience sharing affect targeted advertising outcomes through heterogeneous information behavior. These works support two design choices used here: first, brand perception can be operationalized as a weighted polarity construct rather than a raw mention count; second, targeted advertising effects should be modeled under differentiated information behavior rather than assumed homogeneous across consumers. For the present protocol, these sources are used only for those bounded methodological elements: polarity construction and heterogeneity in information behavior. The separation between click-through and evaluative change is therefore a design decision of this manuscript, not a direct finding from those two studies. [4] [5]

3. Conceptual distinction between brand lift and click-through optimization

Brand lift is defined in this protocol as the change in measured brand evaluation attributable to exposure. The construct has three primary components: brand attitude, sponsor-brand response, and purchase intention. Click-through is defined as a behavioral reaction to an ad unit that indicates contact with a landing environment, not a change in brand evaluation. A campaign can increase clicks by using frictionless calls to action, curiosity gaps, or incentive framing without producing a positive shift in brand attitude. Conversely, a campaign can strengthen brand attitude through placement, appeal fit, or authenticity cues even when the audience does not click. The protocol therefore treats click-through as a possible mediator of further information exposure, not as an outcome that can stand in for brand lift. [1] [2] [3]

This distinction matters because the same exposure can generate different response layers. Audience-level advertising attitude moderated the relative effect of placement versus conventional advertising on brand attitude in one experiment, which implies that an engagement metric alone cannot characterize persuasion. Appeal matching with brand maturity altered purchase intention in another experiment, which implies that effectiveness depends on message-brand fit rather than on one generic response indicator. Sponsorship counter-advertising changed sponsor-brand attitudes and purchase intentions independently of broader alcohol attitudes, showing that brand-specific and category-level reactions may diverge. A valid brand-lift framework must therefore estimate change at the level of the brand construct itself. [1] [6] [2]

4. Proposed variables and operational definitions

The protocol uses four variable blocks. Exposure variables: format (e.g., placement, conventional ad, sponsored post, or sponsorship cue), appeal type (functional, lifestyle, or other pre-specified appeal classes), disclosure status, and repetition count. Audience variables: baseline attitude toward advertising, baseline brand familiarity, prior purchase experience, and information behavior profile. Brand-lift variables: post-exposure brand attitude, sponsor-brand fit or similarity where relevant, purchase intention, and weighted brand polarity from social text. Behavioral variables: click occurrence, dwell time, and subsequent information search. Click occurrence is recorded but not included in the brand-lift index. [1] [6] [4] [5]

Brand polarity is operationalized with a weighted score that combines the sign of sentiment toward the brand and a weight for source influence. One exact operational rule is: BPS = (sum over i of w_i * s_i) / (sum over i of w_i), where s_i is the sentiment polarity of mention i on a normalized scale from -1 to 1, and w_i is a positive weight reflecting engagement and author influence. In survey data, brand attitude is measured with a multi-item semantic differential scale, and purchase intention is measured with a multi-item intention scale. If a campaign uses sponsorship or embedded placement, sponsor-brand fit should be assessed separately from brand attitude because fit may move independently of general favorability. [4] [2] [1]

5. Proposed estimation model and decision sequence

The estimation sequence is as follows. Step 1: define the unit of assignment before data collection. The default rule is individual-level random assignment when exposure can be controlled at the respondent level; if delivery occurs by cluster, the cluster becomes the assignment unit and inference must account for that clustering. Step 2: measure baseline values before any study exposure: brand attitude, purchase intention, brand familiarity, prior purchase experience, attitude toward advertising, and information-behavior profile. Search intensity must be measured at baseline or during a pre-exposure run-in window, not inferred from post-exposure behavior alone, because exposure may itself alter search. Step 3: classify exposures by format and appeal type and record click occurrence only as a behavioral covariate or mediator. Step 4: estimate post-exposure outcomes in two pre-registered windows, an immediate window and a delayed window, using the same attitude and intention instruments plus the polarity score for the corresponding observation period. Step 5: estimate adjusted change with a repeated-measures model that includes baseline outcome, exposure condition, time, and their interaction, plus pre-specified covariates for brand familiarity, prior purchase, attitude toward advertising, and baseline search intensity; if assignment is clustered, use a mixed-effects specification with random intercepts for cluster. Step 6: test heterogeneity by advertising attitude, brand maturity where applicable, and information behavior. The primary comparator is a randomized neutral-control group; if randomization is infeasible, the fallback comparator is a pre-registered matched unexposed group constructed on baseline covariates. [1] [5] [2]

A campaign is classified as producing brand lift only if a pre-registered primary estimand is positive and corroborated by secondary measures under explicit tolerance rules. The default primary estimand is the adjusted pre/post change in a composite survey index formed by averaging standardized brand-attitude and purchase-intention scores. Let Delta_S denote the adjusted difference-in-differences estimate for that index between exposed and comparator units. Let epsilon be a pre-registered equivalence margin for measure disagreement, selected before analysis by a documented sensitivity plan. The default decision rule is: classify brand lift only if Delta_S is greater than zero with its pre-specified uncertainty interval excluding zero, and the polarity change Delta_P is not directionally contradictory, defined as Delta_P being either positive or within the interval [-epsilon, epsilon]. If Delta_P is less than -epsilon while Delta_S is positive, the result is classified as discordant rather than positive. If Delta_S does not clear the uncertainty threshold, the campaign is not classified as producing brand lift even when clicks rise. This rule makes survey-based evaluative change the primary criterion while allowing polarity to function as corroboration or falsification instead of as an unrestricted substitute. [4] [1]

6. Comparator structure and boundary conditions

The comparator should reflect the mechanism under test. If the question concerns format, the comparison is placement versus conventional advertising, with audience attitude toward advertising treated as a moderator. If the question concerns message content, the comparison is functional versus lifestyle appeal, with brand maturity or developmental stage treated as a boundary condition. If the question concerns sponsorship, the comparison should include sponsor-exposing or sponsor-critical conditions where feasible, because sponsorship can alter sponsor-brand image and intentions. These comparisons are not interchangeable. Combining them into one undifferentiated benchmark obscures whether brand lift is driven by format, message appeal, or audience receptivity. [1] [6] [2]

Two boundary conditions are especially important. First, the brand-lift effect may be weak or absent among audiences with positive baseline attitudes toward advertising, even when the same ad works for less receptive viewers. Second, the effect of appeal type may depend on brand maturity or stage, so a single appeal rule should not be generalized across all brands. A further boundary is information behavior: when consumers actively search for product information, targeted advertising may work through a different pathway than when consumers rely on passive exposure. These are not limitations of the measurement framework; they are features the framework should model explicitly. [1] [6] [5]

7. Validation design and prospective tests

Validation should proceed in a fixed order because the three brand-lift measures need not agree. First, measurement validation assesses whether the survey items form reliable attitude and purchase-intention scales and whether the polarity score remains stable under pre-registered data-processing variants. Second, convergence validation evaluates whether attitude, purchase intention, and polarity move in compatible directions after exposure. Compatibility is defined by the same tolerance margin epsilon used in the classification rule: small polarity changes inside the equivalence band count as non-contradictory, whereas movement beyond that band in the opposite direction counts as disagreement. Third, construct validation tests whether the measures respond to manipulations for which the underlying literature gives directional expectations, such as format contrasts, appeal-by-maturity contrasts, or sponsorship-related contrasts. Fourth, external validation examines whether the direction of the estimated effect replicates in a different category or media context without assuming equal magnitude. If the survey index and polarity diverge beyond the tolerance rule, the protocol records the case as inconclusive or discordant rather than forcing a single positive classification. [4] [1] [6]

A prospective evaluation design should include one exposed arm and one neutral-control arm, with randomization at the individual level by default and cluster randomization used only when delivery logistics require it. Baseline measures are collected before assignment. Immediate follow-up should occur soon after the exposure session, and delayed follow-up should occur in a second pre-registered window chosen to test whether any shift persists beyond immediate reaction; the exact intervals must be fixed in advance for the study context. For social-media settings, segmented analysis by baseline search intensity may be added because information behavior can condition targeted-advertising effects, but that variable should be measured before treatment. For weighted polarity, the protocol should pre-register normalization and trimming rules: scale engagement and follower-based components to a common range, cap extreme weights at a fixed upper percentile, label or exclude automated or spam-like accounts by stated criteria, and report both weighted and unweighted polarity as a sensitivity pair. Failure cases should be tracked explicitly: clicks without attitude change, attitude change without clicks, and polarity shifts driven by a small number of amplified posts. [5] [4] [2]

8. Implications for measurement and analysis

The main analytical implication is that ad systems should maintain separate objective functions for acquisition behavior and brand evaluation. The click objective may remain useful for traffic generation, retargeting, and funnel entry. The brand-lift objective should instead be estimated from change in attitude, purchase intention, and polarity, with clicks modeled as an exposure pathway or a secondary behavior. This separation reduces the risk of rewarding ads that optimize navigation rather than perception. It also allows researchers to identify cases where embedded formats, appeal fit, or sponsor criticism affect brand outcomes even when click propensity is low. [1] [2] [6]

Generalization is limited by brand category, audience familiarity, and platform context. The evidence base used here includes familiar consumer brands, social-media environments, and sponsorship settings; it does not justify universal claims about all products or all media. The protocol is therefore deliberately bounded. It is suitable for studies that need a clean estimate of evaluative change under controlled exposure and for cases where click-through is available but theoretically insufficient. It is not a substitute for sales modeling, nor does it imply that every brand-lift effect will be visible in social text or survey responses alone. [1] [3] [4]

9. Conclusion

This protocol establishes a conceptual and methodological separation between brand lift and click-through optimization. Brand lift is defined here as change in brand attitude, sponsor-brand response where relevant, and purchase intention; clicks are treated as behavioral responses that may accompany exposure or mediate further information contact but do not by themselves establish evaluative change. Methodologically, the protocol specifies a survey-first primary estimand, a weighted polarity measure used as corroborating evidence, a pre/post exposure decision sequence, explicit comparators, and a prespecified rule for discordant evidence across measures. The cited evidence supports the narrower claim that format, appeal type, attitude toward advertising, sponsorship context, and consumer information behavior can condition brand-related outcomes, so these factors should be modeled directly rather than collapsed into click metrics. [1] [6] [3] [2] [4] [5]

What remains untested is the full protocol itself: the pre-specified estimands, the combined survey-and-polarity measurement system, and the boundary conditions under which click-through and brand-lift estimates diverge. The next step is a prospective evaluation in which the proposed decision rules are applied to real campaign data without collapsing the two outcomes into one metric. Until that is done, the present contribution should be treated as a bounded measurement proposal for separating persuasion from traffic. [4] [5] [1]

References

1. Li, Linshang (2026). From entertainment to persuasion: how attitudes toward advertising shape responses to brand placement and television advertising. Frontiers in Psychology. 17, 1963379. doi:10.3389/fpsyg.2026.1963379.
2. Dixon, Helen; Scully, Maree; Niederdeppe, Jeff; Brennan, Emily; O'Brien, Kerry; Vandenberg, Brian; Pettigrew, Simone; Wakefield, Melanie (2023). Can counter‐advertising dilute marketing effects of alcohol sponsorship of elite sport: A field experiment. Addiction. 118(12), 2360-2373. doi:10.1111/add.16317.
3. Han, Yu (2026). The relationship between the social media advertising content characteristics and the advertising effectiveness of female users: a moderated mediation model based on basic psychological needs. Frontiers in Psychology. 17, 1831184. doi:10.3389/fpsyg.2026.1831184.
4. Mathew, Sujith Samuel; Hayawi, Kadhim; Venugopal, Neethu; El Barachi, May (2026). Quantifying customer sentiment for automobile brand perception analysis using machine learning on Twitter. Scientific Reports. 16(1), 5703. doi:10.1038/s41598-026-35637-9.
5. Dan, Wu; Jiang, Zhao (2026). The complementary-competitive interplay of consumer information search and experience sharing: Implications for targeted advertising and firm performance. PLOS One. 21(10), e0340849. doi:10.1371/journal.pone.0340849.
6. Xu, Xinliang; Ji, Nannan; Liu, Peipei; Kang, Yixin; Yao, Lin; Lai, Shuxian; He, Xianyou; Zhang, Wei (2026). The Impact of Advertising Appeal Types on Purchase Intentions for Brands in Different Maturity Levels. Behavioral Sciences. 16(10), 1926. doi:10.3390/bs16101926.