Traits, reasons and message fit

Moonbrush research note 3

Version

1.0

Published

October 20, 2026

Checked against the product

October 19, 2026

Authors

Moonbrush

Keywordsmeans-end chain; elaboration likelihood model; personalized persuasion; message matching; digital footprints; algorithmic discrimination

Abstract

Two kinds of Gravity model describe people in ways a creative team can use: reasons, which describe why people buy in any category and are what an audience group’s angle is built on, and traits, which are stable leans between two ends of a scale. stage gives each audience group an angle and an objection, and matches a client’s own ads to each group. This note sets out the research on those ideas and how strong it is.

Means-end research gives a long-standing account of why people buy: product attributes matter for the consequences they bring and the values those serve. Persuasion research explains why a message that fits its audience can work better, and a rating study and field studies report that messages matched to recipients’ personalities persuade more. The largest field evidence has been challenged on methodological grounds, and this note grades it as contested. The same research shows that behavioral records can reveal attributes people consider private, and audits of ad delivery show that targeting systems can produce discriminatory outcomes. The note ends with the limits on use that are consistent with that research.

1. Introduction

Knowing that a person is likely to act is not the same as knowing what to say to them. An audience group in stage carries an angle, the message direction the evidence supports, and an objection, what the group needs answered before it acts. The creative plan then pairs each group with the client’s own ads that speak to it, each pairing with a fit word and a one-line reason. See Audience groups and Creative plan.

Two scientific questions sit underneath. Is there a stable, readable structure to why people buy and how they lean? And does a message that fits that structure persuade more than one that does not?

2. Why people buy: reasons

Means-end chain theory holds that people value product attributes for the consequences those attributes bring, and value consequences for the personal values they serve (Gutman, 1982). Laddering interviews, which ask people repeatedly why an attribute matters to them, are the method developed to trace those chains (Reynolds and Gutman, 1988).

The theory has two properties that matter here. The chains end in a small set of values that recur across categories, so the same reason can drive purchases of very different products. And the chain links what a product is to why a person wants it, which is the link an angle has to make. This is consistent with Gravity reading reasons the same way in any category. See The model catalog.

3. Stable leans: traits

Personality psychology treats some dispositions as stable across time and situations. Research with digital records shows that such dispositions can be estimated from behavior. In one study, judgments of personality computed from Facebook Likes were more accurate than those made by humans (Youyou et al., 2015). In another, Likes from 58,466 volunteers distinguished a wide range of personal attributes, for example with an AUC of 0.93 for gender, and also predicted attributes such as political and religious views and sexual orientation (Kosinski et al., 2013).

A Gravity trait names both ends of its scale, so a person who leans one way is described by that end, not by the absence of the other. Trait names are given in the product. Section 6 sets out what Moonbrush does not model or offer.

4. Why fit can persuade

The elaboration likelihood model distinguishes persuasion that runs through careful thought about a message’s arguments from persuasion that runs through simpler cues, and describes the conditions under which people think carefully (Petty and Cacioppo, 1986). A review of personalized matching effects in persuasion sets out a framework for why messages matched to a recipient’s characteristics can be more persuasive (Teeny et al., 2021).

Table 1. Studies of matching messages to personality, with the published critique.

Study	Design	Result
Hirsh et al. (2012)	Rating study. 324 participants rated five versions of an ad for one phone, each framed for one personality trait.	Ads were rated more positively the more they matched the reader’s own motives.
Matz et al. (2017)	Three field studies on Facebook, reaching over 3.5 million people, with ad content matched to personality inferred from digital records.	Matched ads produced up to 40% more clicks and up to 50% more purchases than mismatched or unpersonalized ads.
Eckles et al. (2018)	Published critique of the field studies above.	The ad platform’s delivery optimization decides who in each audience sees an ad, so differences between conditions may reflect delivery rather than fit. The original authors replied that this was unlikely to explain their results.

The rating study cited finds that fit improves how a message is received. The field evidence that fit changes behavior at scale is reported but contested, because the platforms that deliver ads also choose who sees them. This is the same problem of comparing groups that were not formed at random that note 7 discusses for measuring results.

5. How Moonbrush applies this, and what it claims

• Angles are written from evidence. An angle and an objection are drawn from the behaviors that set a group apart from its audience, and group explanations describe the group, never a single member.
• Matching uses the client’s own ads, and may decline. When no ad in the library speaks to a group, the matcher says so instead of placing a misfit. An empty row tells the creative team what to make next.
• No persuasion effect is claimed. Moonbrush does not publish a measured effect of matching. A group’s response is read from the campaign’s Outcome, counted under rules fixed before launch; whether the ads changed what the group did would need a held-back comparison. See note 7.

6. Sensitive attributes and discrimination

The research that shows behavior can reveal traits also shows it can reveal what people regard as private (Kosinski et al., 2013). And research on data-driven systems shows they can discriminate even without intent. Legal scholarship has set out how decisions driven by data can have a disparate impact on protected groups (Barocas and Selbst, 2016), and studies of advertising platforms have examined the potential for discrimination in targeting and in ad settings (Speicher et al., 2018; Datta et al., 2015). In one audit, ad delivery optimization alone skewed who saw real job and housing ads by gender and race, even when the advertiser’s targeting was neutral (Ali et al., 2019).

Moonbrush’s rules on use are consistent with this research. Some are product rules and some are terms of the Data Use Policy.

• Permitted use (Data Use Policy). Moonbrush data and models may be used for marketing and engagement only, and never to decide a person’s eligibility for credit, insurance, employment, housing, education or similar. See Privacy and permitted use.
• Modeled ethnicity and political ideology (Data Use Policy) are never selectable, never used as targeting inputs and never exported.
• Age. Every age band the product offers starts at 18.
• Demographics are context, not evidence. Age band or income can describe an audience group, but never count as the reason it fits. See note 4.
• A language gate. A model’s name and card must describe what it measures and nothing more, so a model cannot be named for an attribute it does not measure. See Quality, coverage and freshness.

These controls govern what Moonbrush offers. They do not govern how an ad platform delivers an ad once a list reaches it, which an audience provider does not control.

7. Questions an auditor should ask

1. For each trait and reason model, what evidence gate did it pass, and what does its card say it does not mean?
2. How is the language gate applied, and has it blocked a model name?
3. What checks show that no selectable model is a close proxy for a field that is never selectable?

8. Conclusion

There is good research for the idea that why people buy has a readable structure, and peer-reviewed reports that stable dispositions show in behavior. The evidence that matching messages to those dispositions changes behavior at scale is reported but contested, and Moonbrush claims no effect of its own. The same research shows why limits on use matter, and Moonbrush’s rules on use are consistent with it.

References

1. Ali, M., Sapiezynski, P., Bogen, M., Korolova, A., Mislove, A., & Rieke, A. (2019). Discrimination through optimization: How Facebook’s ad delivery can lead to biased outcomes. Proceedings of the ACM on Human-Computer Interaction, 3(CSCW), Article 199. doi.org/10.1145/3359301
2. Barocas, S., & Selbst, A. D. (2016). Big data’s disparate impact. California Law Review, 104, 671–732. doi.org/10.15779/Z38BG31
3. Datta, A., Tschantz, M. C., & Datta, A. (2015). Automated experiments on ad privacy settings. Proceedings on Privacy Enhancing Technologies, 2015(1), 92–112. doi.org/10.1515/popets-2015-0007
4. Eckles, D., Gordon, B. R., & Johnson, G. A. (2018). Field studies of psychologically targeted ads face threats to internal validity. Proceedings of the National Academy of Sciences, 115(23), E5254–E5255. doi.org/10.1073/pnas.1805363115
5. Gutman, J. (1982). A means-end chain model based on consumer categorization processes. Journal of Marketing, 46(2), 60–72. doi.org/10.1177/002224298204600207
6. Hirsh, J. B., Kang, S. K., & Bodenhausen, G. V. (2012). Personalized persuasion: Tailoring persuasive appeals to recipients’ personality traits. Psychological Science, 23(6), 578–581. doi.org/10.1177/0956797611436349
7. Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits and attributes are predictable from digital records of human behavior. Proceedings of the National Academy of Sciences, 110(15), 5802–5805. doi.org/10.1073/pnas.1218772110
8. Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2017). Psychological targeting as an effective approach to digital mass persuasion. Proceedings of the National Academy of Sciences, 114(48), 12714–12719. doi.org/10.1073/pnas.1710966114
9. Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of persuasion. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 19, pp. 123–205). Academic Press. doi.org/10.1016/S0065-2601(08)60214-2
10. Reynolds, T. J., & Gutman, J. (1988). Laddering theory, method, analysis, and interpretation. Journal of Advertising Research, 28(1), 11–31. doi.org/10.1080/00218499.1988.12467766
11. Speicher, T., Ali, M., Venkatadri, G., Ribeiro, F. N., Arvanitakis, G., Benevenuto, F., Gummadi, K. P., Loiseau, P., & Mislove, A. (2018). Potential for discrimination in online targeted advertising. In Proceedings of the 1st Conference on Fairness, Accountability and Transparency, Proceedings of Machine Learning Research 81 (pp. 5–19). proceedings.mlr.press/v81/speicher18a.html
12. Teeny, J. D., Siev, J. J., Briñol, P., & Petty, R. E. (2021). A review and conceptual framework for understanding personalized matching effects in persuasion. Journal of Consumer Psychology, 31(2), 382–414. doi.org/10.1002/jcpy.1198
13. Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. Proceedings of the National Academy of Sciences, 112(4), 1036–1040. doi.org/10.1073/pnas.1418680112

How to cite this note

Moonbrush. (2026). Traits, reasons and message fit (Research note 3, version 1.0). https://docs.moonbrush.com/research/traits-reasons-and-fit/