Personalization：证据证明了什么，又止步于何处

Personalization is the practice of tailoring marketing content or ads to an individual's data, such as past behavior, stated interests, or profile traits, instead of showing everyone the same message. It is a practitioner term with no single official definition, so its scope varies by team.

Does it work? Modestly, and in a specific way. A 2025 meta-analysis of 53 experiments found that personalized ads beat generic ones on overall persuasion, consumer attitudes, and behavioral intentions. The effects were small. And the research measured what people thought and intended, not what they bought, which is where this entry draws the line.

That line matters. Many pitches jump from "people like relevant ads" to "relevant ads drive revenue," and the evidence reviewed here supports the first half far better than the second.

Personalization, Segmentation, and the Terms Around It

Personalization sits in a family of terms that often get used interchangeably. They are not the same thing, and mixing them up leads to muddled briefs and mismatched tests.

Customer segmentation groups people who share traits, such as age band, purchase history, or region, and then speaks to each group. Personalization goes one level further and adapts the message to the individual, using data about that person. In practice, segmentation is often the groundwork that personalization is built on, since you need clean customer data and sensible groupings before individual tailoring makes sense.

One-to-one marketing is an older strategic idea about managing each customer relationship individually. Personalization is narrower. It describes what happens to a piece of content or an ad, not the whole relationship.

Then there is the more granular, data-heavy end of the spectrum. If you are working with real-time signals and many data points per person, the separate entry on hyperpersonalization covers that ground. This entry stays with the broader concept and with a narrower question: what does the evidence actually show?

Customization is different again. When customers choose their own settings or preferences, they do the tailoring. With personalization, the marketer or the system does it.

What the Research Actually Measures

The strongest single body of evidence on personalized advertising available for this entry is a meta-analysis published in the Journal of Advertising Research in 2025 by Tien Ee Dominic Yeo, Tsz Hang Chu, and Qiqi Li. A meta-analysis pools the results of many separate studies to estimate an average effect, which makes it more reliable than any one experiment on its own.

Persuasion, Attitudes, and Intentions

The review covered 53 experimental studies. According to the paper's reported results, those studies included 11,786 participants and produced 114 effect sizes, with the literature search updated through November 2024. The authors frame their question around what they call the personalization paradox: earlier research had found that personalization could produce both positive and negative persuasion outcomes, and nobody had a clear average.

Their answer was that personalized advertising is generally more effective than generic, non-personalized advertising. The reported effect sizes, measured as Cohen's d, were:

• Overall persuasion: d = 0.16 (95% confidence interval 0.05 to 0.27)
• Consumer attitudes: d = 0.13 (95% confidence interval 0.01 to 0.25)
• Behavioral intentions: d = 0.18 (95% confidence interval 0.06 to 0.30)

Cohen's d expresses the gap between two groups in standard deviation units. By the common rule of thumb, a d around 0.2 counts as small. So all three effects are real, in the sense that their confidence intervals stay above zero, and all three are modest.

A d of 0.16 is not a 16% increase in anything. It cannot be converted into a conversion lift, a revenue figure, or a percentage gain without data the meta-analysis does not provide. Any deck that turns these numbers into "personalization lifts results by 16%" is misreading the statistic.

Where the Evidence Stops: Purchases and Revenue

Look closely at the three outcomes. Persuasion, attitudes, and intentions are all things people report, typically in an experimental setting where they see an ad and answer questions about it. None of them is a completed purchase.

That distinction is not pedantic. The gap between "I would consider buying this" and a paid order is familiar to anyone who has read a cart abandonment report. Across the sources reviewed for this entry, no verified field evidence tied personalization to a measured change in purchases or revenue. That does not mean the effect is absent. It means the evidence here does not establish it.

For a marketer, the practical reading is this. Personalization has decent support as a way to make ads feel more relevant and to nudge attitudes and intentions upward. Whether that carries through to your conversion rate is something you have to test in your own data, not assume from the literature.

What Makes Personalization Work

An average effect tells you whether something works, not why. The same meta-analysis went further.

Relevance, Not Intrusiveness

Using meta-analytic structural equation modeling, the authors tested two competing explanations. One says personalization works because the ad feels relevant. The other worries that personalization backfires because it feels intrusive.

The finding was clear. Perceived relevance explained the persuasiveness of personalized advertising. Perceived intrusiveness did not. This fits two established ideas in consumer psychology, congruity theory and self-referencing theory, both of which predict that people respond more to messages that connect to who they are and what they care about.

The implication is useful. What does the work is how relevant the ad feels, not how much data sits behind it. Heavy data use that still produces generic-feeling ads gains little.

Real Data Versus Imagined Scenarios

Many personalization experiments ask participants to imagine that an ad was tailored to them. Others actually tailor the ad using the participant's real information. According to reported findings from the meta-analysis, personalization based on participants' actual data was more effective than scenario-based, imagined personalization.

So studies built on hypothetical scenarios may say little about how real personalization performs. When you read a personalization study, check which design it used before leaning on its results.

Covert Versus Overt

Coverage of the meta-analysis suggests that personalization based on behavioral traces may do better than explicit cues such as inserting a first name. The subgroup statistics behind that comparison could not be confirmed from the open sources reviewed here, so this entry makes no claim in either direction.

Generative AI and Personalization at Scale

For most of marketing history, writing a different message for every person was too expensive to do at scale. Generative AI changes that cost equation, and research has started to test whether the output is any good.

A 2024 paper in Scientific Reports by Sandra Matz, Jacob Teeny, and colleagues found that messages written by ChatGPT and tailored to people's psychological profiles were more persuasive than messages that were not personalized. In other words, the machine-written personalized copy did what personalized copy is supposed to do.

The same caution applies. The study measured persuasion, not sales. What it does suggest is that producing tailored variants is no longer the bottleneck. The harder questions are whether the underlying data is accurate and consented, whether the result feels relevant, and whether the uplift survives a real business test.

How Consumers See It

Experiments measure how people respond to a single ad in a controlled setting. Surveys measure how people feel about personalization in general. The two give different pictures, and both are worth knowing.

The European Commission's 2025 Consumer Conditions Scoreboard, based on a survey carried out by Ipsos in November 2024 and published in March 2025, found that 93% of online shoppers in the EU expressed worries over online targeted advertising. The specific worries included the collection of personal data without consent (71%), excessive advertising (67%), and unavoidable personalisation (63%).

Separately, Eurostat reported in January 2026 that 58.8% of EU internet users chose not to allow their personal data to be used for advertising in 2025, up 4.5 percentage points from 2023. The figure covers people who had used the internet in the three months before the survey. It comes from a different survey with a different population from the Commission scoreboard, so the two numbers should be read side by side, not combined.

Do these surveys contradict the finding on intrusiveness? Not really. The meta-analysis measured responses to a specific ad; the surveys capture broad concern and data choices. A person can find a well-targeted ad persuasive and still worry about data collection.

Where this becomes a business issue is reach. If more users refuse to let their data be used for advertising, the audience you can personalize to shrinks. That is analysis, not a measured outcome, but it argues for building on consented first-hand data.

Evidence From Indonesia

Indonesia-specific evidence on personalization is thin. One relevant study, published in October 2025 in Image: Jurnal Riset Manajemen by Mendrika Ny Hanitra Randrianavo, Vanessa Gaffar, and Tutik Inayati of Universitas Pendidikan Indonesia, surveyed 114 Shopee users selected through purposive sampling.

The study found that personalized advertising had a significant direct effect on purchase intention, and a significant indirect effect through attitude toward the ads. The authors describe this as partial mediation: part of personalization's effect on intention runs through how people feel about the ad itself, and part goes direct.

Two limits apply. The sample is small and not randomly drawn, so the results describe this group of users, not Indonesian shoppers as a whole. And like the international research, the outcome is intention, not purchase. Nationally representative data on personalization in Indonesia was not found in the sources reviewed for this entry.

How to Judge a Personalization Program

The evidence above suggests a handful of practical checks. These are recommendations drawn from the research, not findings in themselves.

Define the outcome before you launch. If success means more purchases, measure purchases. Click-through or survey-based lift is a reasonable early signal, but the literature shows those measures can move without proving a sales effect.

Use a holdout. Keep a randomly selected group on the generic experience and compare. Structured A/B testing is the simplest way to see whether personalization adds anything beyond what would have happened anyway.

Judge relevance, not data volume. The research points to perceived relevance as the mechanism. Ask whether a person would recognize the message as meant for them.

Use real data, gathered with consent. Personalization based on genuine information outperformed imagined scenarios in the research, and refusals to share data for advertising are rising in the EU. Both point to well-consented, accurate first-hand data.

Report effect sizes honestly. If your test shows a small lift, say so. Do not convert academic effect sizes into percentage claims, and do not borrow published averages as forecasts for your own campaign.

Frequently Asked Questions

Does personalization increase sales?

The research reviewed here shows that personalized ads improve persuasion, attitudes, and purchase intentions compared with generic ads. It does not establish a measured increase in actual purchases or revenue. That has to be tested in your own campaigns with a holdout group.

How much more effective are personalized ads than generic ones?

A 2025 meta-analysis of 53 experiments reported small effects: Cohen's d of 0.16 for overall persuasion, 0.13 for attitudes, and 0.18 for behavioral intentions. These are standardized effect sizes, not percentage lifts.

Why does personalization work?

The same meta-analysis found that perceived relevance explained the effect. People respond better when an ad feels meant for them. Perceived intrusiveness did not explain the results.

Do consumers worry about personalized advertising?

Many do. In the European Commission's 2025 scoreboard, 93% of online shoppers in the EU expressed worries over online targeted advertising, and 63% cited unavoidable personalisation. Separately, Eurostat found that 58.8% of EU internet users chose not to allow their data to be used for advertising in 2025.

Can generative AI make personalization more effective?

A 2024 study in Scientific Reports found that ChatGPT-written messages tailored to people's psychological profiles were more persuasive than non-personalized messages. The study measured persuasion, not sales.

Is there research on personalization in Indonesia?

A 2025 study of 114 Shopee users found that personalized advertising affected purchase intention both directly and through attitude toward the ads. The sample was small and purposive. Nationally representative Indonesian data was not found in the sources reviewed.

Related Terms

• Customer Segmentation: grouping customers by shared traits, often the groundwork for personalization.
• Marketing Automation: software-driven workflows that can deliver personalized messages at scale.
• Email Marketing: one of the most common channels for personalized content.

Sources

Yeo, T. E. D., Chu, T. H., and Li, Q. (2025). How persuasive is personalized advertising? A meta-analytic review of experimental evidence of the effects of personalization on ad effectiveness. Journal of Advertising Research, 65(4), 616-631. DOI: 10.1080/00218499.2025.2467763. Metric: meta-analysis of experiments, Cohen's d. Geography: pooled studies, no country breakdown. Confidence: VERIFIED for the abstract findings (53 studies, direction of effect, relevance as mechanism); REPORTED for effect sizes, participant and effect-size counts, search date, and the real versus imagined comparison. https://doi.org/10.1080/00218499.2025.2467763.

European Commission, Directorate-General for Justice and Consumers (2025). 2025 Consumer Conditions Scoreboard, published 14 March 2025. Consumer Conditions Survey fieldwork by Ipsos, November 2024. Metric: share of online shoppers reporting worries over online targeted advertising. Geography: EU. Confidence: VERIFIED. https://cyprus.representation.ec.europa.eu/news/new-data-shows-strong-levels-consumer-trust-online-threats-persist-2025-03-14_en.

Eurostat (2026). Data Protection Day news release, 28 January 2026. Metric: share of internet users (used the internet in the three months before the survey) who chose not to allow personal data to be used for advertising, 2025. Geography: EU. Confidence: VERIFIED. https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20260128-1.

Matz, S. C., Teeny, J. D., Vaid, S. S., Peters, H., Harari, G. M., and Cerf, M. (2024). The potential of generative AI for personalized persuasion at scale. Scientific Reports, 14, 4692. DOI: 10.1038/s41598-024-53755-0. Metric: experimental persuasion outcomes. Geography: not specified in the sources reviewed. Confidence: VERIFIED for the study and direction of finding. https://doi.org/10.1038/s41598-024-53755-0.

Randrianavo, M. N. H., Gaffar, V., and Inayati, T. (2025). The effect of personalized advertising on purchase intention: The mediating role of attitude toward ads among Shopee users. Image: Jurnal Riset Manajemen, 13(2), 168-176. DOI: 10.17509/image.2025.164. Metric: survey, regression and mediation analysis. Geography: Indonesia, 114 Shopee users, purposive sampling. Confidence: VERIFIED for direction and design; coefficients not used. https://ejournal-economy.upi.edu/image/article/view/164.