What Psychographic Targeting Can Do
The trade sells psychographic targeting on one famous Facebook experiment, and the first meta-analysis of the field is harsher than the pitch. We read both, along with the two published critiques nobody mentions, and the answer turns on where a category is cut and which comparison was run.
Answer to the question
What can psychographic targeting actually do?
Very little, on the pooled evidence. The first meta-analysis of psychological targeting, published in 2026 by Raphael Perla and five co-authors, found that digital footprints explain about five per cent of the variation in personality, and that under the only design able to identify a matching effect the result cannot be distinguished from zero. Chained end to end, the effect approaches zero again.
Key takeaways
- The first meta-analysis of psychological targeting, published in 2026 by Raphael Perla and five co-authors, found that digital footprints explain about five per cent of the variation in personality once studies with data leakage are removed.
- Under the only design that can identify a matching effect, two of eleven estimates reached significance, against three quarters of the estimates drawn from weaker designs.
- In the 2017 field experiment led by Sandra Matz, the largest campaign produced 390 purchases from 3,129,993 people, a conversion rate of 0.012 per cent, with matched viewers 1.54 times more likely to buy.
- Two published critiques followed in 2018, one of them carrying a correction that discloses its authors' institute funding, and no direct replication has appeared in the nine years since.
Psychographic targeting, the practice of inferring someone's personality from the traces they leave online and then tailoring an advertisement to the inference, does much less than the people selling it believe. It is what the published record shows once you separate the two claims the pitch runs together, which are that a personality can be read off a digital footprint and that a message matched to that personality moves a purchase. Both have now been measured across a whole literature rather than one study at a time. The first works modestly. The second, tested under a design that can identify it, barely registers.
What the seed book settles about types
Benjamin Hardy's Personality Isn't Permanent, published in 2020, opens on personality tests, and on the Myers-Briggs Type Indicator it reaches the right conclusion. Page 28 sets out four standards a theory of personality has to meet, reliability, validity, independence and comprehensiveness, and reports that on the Myers-Briggs "the evidence says no, no, no, and no". Two pages later the five-factor model, the continuous trait framework researchers use, is called "the most scientifically backed theory of personality", with the insistence that "None of these five factors are 'types'".
The pages behind that verdict do not carry it. The notes offer a trade history of the instrument, a quotation, a remark about horoscopes and a podcaster's joke about being one type before coffee and another after. The verdict itself is stated and not cited.
The failure is also not the one usually described. Robert Capraro's reliability generalisation study of 2002, pooling the instrument's record across many administrations, found its scales yielded scores with strong internal consistency and strong test-retest reliability. Robert McCrae's comparison of 1989 had already found its four indices measure aspects of four of the five major dimensions of normal personality, with no support for the view that they measure distinct types.
So the scales hold and the categories do not, which only sounds contradictory until you look at where the cut falls. A continuous spread of scores is sliced near its middle and the slice is reported back as a kind of person. Anyone close to that line crosses it on ordinary noise, their letter changing while their score has barely moved. It is a badly specified measure of something real, which is harder criticism than calling it unreliable.
A reliable continuous scale cut into categories near its middle will move a letter that the score behind it has barely moved.
Type instruments should not decide anything about people. We agree with that.
The chapter never reaches the commercial question. Trait scales survive the criticism that sinks the types, and trait scales are what psychographic targeting runs on. Personality can be measured as a continuum, and our piece on why nobody is born consistent sets out how much it moves. Whether it can be sold on is answered in advertising journals.
What Matz and her colleagues found
The psychographic targeting study the trade runs on was published in 2017 by Sandra Matz and three co-authors, and the strongest version of it is strong. Three field experiments reached more than three and a half million people, with real money behind real Facebook campaigns. In the largest, a beauty retailer's seven-day campaign in the United Kingdom, people shown the advertisement matched to their extraversion were 1.54 times more likely to buy than people shown the mismatched one.
| Campaign | Reached | Clicks | Conversions | Rate |
|---|---|---|---|---|
| Beauty retailer, seven days | 3,129,993 | 10,346 | 390 purchases | 0.012 per cent |
| Crossword application, twelve days | 84,176 | 1,130 | 500 installs | 0.59 per cent |
| Bubble-shooter game, seven days | 534,250 | 3,367 | 1,837 installs | 0.34 per cent |
The authors also mark their own boundaries, which trade summaries rarely do. Targeting used the smallest possible input, one Facebook Like per person, which is why they call their estimate conservative. In the beauty campaign there was no effect on clicks that the study could detect. In the crossword campaign the install effect was carried by audiences low in openness, with nothing the study could detect among people high in openness.
The third campaign was not a matching test. It set the company's standard copy against one tailored message, on an audience the authors profiled as introverted.
Now read the famous numbers against the table. "Up to 40 per cent more clicks" and "up to 50 per cent more purchases" are its most favourable cells, and the biggest campaign's rate is 390 purchases from 3,129,993 people.
Both of those statements are in the same paper.
"Up to 40 per cent more clicks" and a conversion rate of 0.012 per cent come from the same study, two pages apart.
What Eckles and Sharp each objected to
Two published critiques of psychographic targeting followed, attacking different things, so answering one does not dispose of the other.
Dean Eckles and two co-authors, writing in 2018, took the design. People were never randomly assigned to advertisements; a platform decides who sees which creative, and it decides on expected response, so each advertisement already has a different audience. They reanalysed the authors' own data and found the age and gender imbalances this predicts. Their sharpest move is arithmetic: the reported rises in conversions would be accounted for by confounding biases of 25, 9 and 5 per cent, and a documented Facebook study had found a bias near 6 per cent. A bias that size would cover the 5, and it leaves the 25 and the 9 standing.
Byron Sharp and two co-authors, also in 2018, took the comparison instead. In two of the three campaigns the matched advertisement was tested against a deliberately mismatched one rather than against ordinary untargeted advertising, which is the only baseline a marketer can act on. They also argued creative quality was never ruled out, since the advertisements were pretested on small student groups rather than on the audience that saw them.
That letter carries a published correction disclosing that its three authors are employed by an institute funded by Facebook and by its competitors. We print it because the same institute's work sits behind our piece on why brand loyalty is mostly brand size. The 2017 paper carries a declared interest of its own, and we print both: David Stillwell received revenue as owner of the myPersonality application, the source of its Likes data, until 2012. These are disclosures, not findings about anybody.
Matz and her co-authors replied to both critiques in print. To Eckles they granted that platform algorithms can threaten a field study, then argued that the demographic differences were tiny, the estimates almost identical once controls were added, and an interaction harder to produce by accident than a main effect. To Sharp they argued that five of six tests came out significant with none against their hypothesis.
They also invited replication, naming the newer Facebook tools that allow a clean split test. Nine years on, we looked and did not find one.
What the first meta-analysis reports
In 2026 Raphael Perla and five co-authors published the first meta-analysis of psychological targeting, the research literature's name for what the trade calls psychographic targeting. It takes the pitch apart in the order the pitch is assembled.
1. Read the personality. Digital footprints predicted about five per cent of the variation in personality once the pool held only studies free of data leakage, which is any bleed between a model's training data and its test data. Leakage was confirmed in about 43 per cent of studies and nearly doubled the accuracy they reported. The typical study had six observations per feature, and almost 60 per cent failed the minimum requirements for their own regression.
2. Match the message. Only a third of the targeting studies used the design that can identify a personalisation effect at all, which tests the trait and the message together so a good advertisement cannot be mistaken for a well-aimed one. Under the weaker designs, three quarters of the estimates came out significant. Under the identifying design, two of eleven did, and the authors say the true effect cannot credibly be distinguished from zero.
3. Chain the two together. Discount that small targeting effect by the accuracy with which a personality can actually be read, and the end-to-end effect approaches zero again.
Their reading of the gap between the two sets of designs should worry a buyer. Earlier effects may have been picking up the quality of the message, or the personality of the recipient, rather than any benefit from matching one to the other. Hence their first recommendation: put the money into the advertisement.
Two honest things about that paper. It leaves the beauty-retailer campaign out of its pooled estimate and says why, which is that converting an odds ratio into a correlation misbehaves when the outcome is as rare as 390 cases in three million. That is defensible, and means the best-known result in the field sits outside the headline figure. Its reported heterogeneity does not match between the results and the discussion, and we could not reconcile the two.
What we think the record supports
Set the strongest case for psychographic targeting beside the strongest case against it. For: a reach of millions, a real purchase effect in the largest campaign, authors who flagged their own nulls, and a political study by Brahim Zarouali's team in which people were more persuaded by advertisements matched to their traits. Against: no randomisation, a comparison against mistargeting, creative quality uncontrolled, no replication, and a pooled effect surviving only under designs that cannot tell matching apart from message quality.
We come down with the meta-analysis, with two qualifications. The Zarouali result is real, and it is political rather than commercial, measured on persuasion scales rather than money spent. The prior is also unkind to personalisation: when Alexander Coppock's team ran 59 randomised experiments testing 49 political advertisements on 34,000 people, the effects were small and did not vary much by who received the message, which leaves matching on the receiver little to work with. Those were political advertisements, and the carry-over to commercial targeting is our inference.
### What this cannot settle
An interval that includes zero says what the evidence can resolve, not that the effect is absent, and Perla's team say as much, that small positive effects could in principle accumulate at scale. Nor can we say Matz and her co-authors were wrong. Two sets of critics found two different problems, the authors answered both, and the largest arm was later set aside on statistical grounds rather than refuted.
Some sources here are thin in ways that matter, the problem our piece on what forty-one participants buys describes. The lab precursor the field cites used 324 survey respondents and measured ad ratings. We also declined to print the most-quoted statistic on Myers-Briggs retest instability, because no published record we could reach states it.
The practical position is narrower than either side's. Psychographic targeting at the type level should not be bought, because the categories are an artefact of the cut. At the trait level it is not nothing, merely much smaller than the accuracies make it sound, since a categorical classification at 88 per cent and a modest trait correlation get quoted in the same breath. If aiming buys less than making a better advertisement does, what exactly is the segment being sold to the advertiser, and who ever checks whether it did anything?
Common questions
What can psychographic targeting actually do?
It reads personality from digital traces at a modest level and moves behaviour very little. The field's first meta-analysis, published in 2026 by Raphael Perla and colleagues, found that digital footprints predict about five per cent of the variation in personality once the pool held only studies free of data leakage, and that under the design that can identify a matching effect the result cannot be distinguished from zero.
Does personality-targeted advertising increase sales?
In the best-known psychographic targeting field test, by Sandra Matz and co-authors in 2017, people shown a matched advertisement were 1.54 times more likely to buy, on a conversion rate of 0.012 per cent. Two published critiques questioned the randomisation and the comparison, and no direct replication has appeared.
Should a marketer buy psychographic segments?
Treat them as a low priority rather than a lever. Perla and colleagues recommend putting the budget into the advertisement's quality and fit, because differences there dominate any plausible targeting gain at today's accuracies. Ask a vendor which comparison their case study used, and whether anybody replicated it.
Sources
Perla, R., Maran, T., Bagci, B., Kraus, S., Kanbach, D. K., & Bouncken, R. B. (2026). 'The (In)Effectiveness of Psychological Targeting: A Meta-Analytic Review.' Psychology & Marketing, 43(3), 575-590. doi:10.1002/mar.70073. [FULL TEXT READ, open access CC-BY at onlinelibrary.wiley.com; author order confirmed at Crossref and OpenAlex, Perla first, Heidelberg University; Kraus corresponding. First published online 28 October 2025, print issue March 2026]
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:10.1073/pnas.1710966114. [FULL TEXT READ including Table 1, Methods and declared interests, from the author's copy at Columbia Business School; author order confirmed at Crossref and OpenAlex, Matz first, Columbia; Kosinski at Stanford GSB; Nave at Wharton; Stillwell at Cambridge Judge last; the paper itself states Kosinski and Stillwell 'share senior authorship']
Eckles, D., Gordon, B. R., & Johnson, G. A. (2018). 'Field studies of psychologically targeted ads face threats to internal validity.' PNAS, 115(23), E5254-E5255. doi:10.1073/pnas.1805363115. [FULL TEXT READ at pnas.org, free access; author order confirmed at OpenAlex, Eckles first, MIT IDSS and MIT Sloan; Gordon and Johnson at Kellogg, Northwestern]
Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2018). 'Reply to Eckles et al.: Facebook's optimization algorithms are highly unlikely to explain the effects of psychological targeting.' PNAS, 115(23), E5256-E5257. doi:10.1073/pnas.1806854115. [FULL TEXT READ at pnas.org]
Sharp, B., Danenberg, N., & Bellman, S. (2018). 'Psychological targeting.' PNAS, 115(34), E7890. doi:10.1073/pnas.1810436115. [FULL TEXT READ at pnas.org; author order confirmed at OpenAlex, Sharp first, all three at the Ehrenberg-Bass Institute, University of South Australia]
Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2018). 'Reply to Sharp et al.: Psychological targeting produces robust effects.' PNAS, 115(34), E7891. doi:10.1073/pnas.1811106115. [FULL TEXT READ at pnas.org]
Kosinski, M., Stillwell, D. J., & Graepel, T. (2013). 'Private traits and attributes are predictable from digital records of human behavior.' PNAS, 110(15), 5802-5805. doi:10.1073/pnas.1218772110. [Abstract read in full; author order confirmed at Crossref and OpenAlex, Kosinski first, University of Cambridge; Graepel at Microsoft Research UK, which is a declared-interest fact; the vault separately holds a full-text read from its pass of 2026-10-05]
Youyou, W., Kosinski, M., & Stillwell, D. (2015). 'Computer-based personality judgments are more accurate than those made by humans.' PNAS, 112(4), 1036-1040. doi:10.1073/pnas.1418680112. [Abstract read in full; author order confirmed at Crossref and OpenAlex, Wu Youyou FIRST, University of Cambridge; credit Youyou and not Kosinski]
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:10.1177/0956797611436349. [Abstract read in full; author order confirmed at OpenAlex, Hirsh first, University of Toronto; Kang at Toronto; Bodenhausen at Northwestern; OpenAlex records the title as 'Personalized Persuasion' alone]
Zarouali, B., Dobber, T., De Pauw, G., & de Vreese, C. H. (2022). 'Using a Personality-Profiling Algorithm to Investigate Political Microtargeting: Assessing the Persuasion Effects of Personality-Tailored Ads on Social Media.' Communication Research, 49(8), 1066-1091. doi:10.1177/0093650220961965. [Abstract read in full; author order confirmed at Crossref and OpenAlex, Zarouali first, University of Amsterdam; online 20 October 2020, print issue December 2022]
Winter, S., Maslowska, E., & Vos, A. L. (2021). 'The effects of trait-based personalization in social media advertising.' Computers in Human Behavior, 114, 106525. doi:10.1016/j.chb.2020.106525. [RECORD ONLY. ABSTRACT NOT AVAILABLE and not read. Author order confirmed at Crossref and OpenAlex, Winter first, University of Koblenz and Landau; Maslowska at Illinois Urbana-Champaign; Vos at Amsterdam. Crossref print issue January 2021; OpenAlex dates it 2020. Tried OpenAlex by DOI, the Crossref item record, and the ScienceDirect landing page in the built-in browser, which refused the request]
Coppock, A., Hill, S. J., & Vavreck, L. (2020). 'The small effects of political advertising are small regardless of context, message, sender, or receiver: Evidence from 59 real-time randomized experiments.' Science Advances, 6(36). doi:10.1126/sciadv.abc4046. [Abstract read in full; gold open access; author order confirmed, Coppock first, Yale; Hill at UC San Diego; Vavreck at UCLA]
McCrae, R. R., & Costa, P. T. (1989). 'Reinterpreting the Myers-Briggs Type Indicator From the Perspective of the Five-Factor Model of Personality.' Journal of Personality, 57(1), 17-40. doi:10.1111/j.1467-6494.1989.tb00759.x. [Abstract read in full; author order confirmed at Crossref and OpenAlex, McCrae first, National Institute on Aging; Costa second]
Stein, R., & Swan, A. B. (2019). 'Evaluating the validity of Myers-Briggs Type Indicator theory: A teaching tool and window into intuitive psychology.' Social and Personality Psychology Compass, 13(2), e12434. doi:10.1111/spc3.12434. [Abstract read in full; author order confirmed at Crossref and OpenAlex, Stein first, California State Polytechnic University; Swan at Eureka College; item number e12434, not a page range]
Capraro, R. M., & Capraro, M. M. (2002). 'Myers-Briggs Type Indicator Score Reliability Across Studies: A Meta-Analytic Reliability Generalization Study.' Educational and Psychological Measurement, 62(4), 590. doi:10.1177/001316402128775030. [Abstract read in full; author order confirmed at Crossref and OpenAlex, Robert M. Capraro first, both authors Texas A&M. The DOI is NOT 10.1177/0013164402062004004, which redirects; the deposited DOI is 10.1177/001316402128775030. Crossref deposits first page 590 only and the commonly printed range 590-602 is secondary]
Pittenger, D. J. (1993). 'The Utility of the Myers-Briggs Type Indicator.' Review of Educational Research, 63(4), 467-488. doi:10.2307/1170497. [Abstract read in full; sole author, Marietta College at publication; Crossref deposits first page 467 only and the range 467-488 is secondary]
Pittenger, D. J. (2005). 'Cautionary comments regarding the Myers-Briggs Type Indicator.' Consulting Psychology Journal: Practice and Research, 57(3), 210-221. doi:10.1037/1065-9293.57.3.210. [Abstract read in full; sole author, University of Tennessee at Chattanooga at publication]