Native Advertising and the Inversion
James Williams raises native advertising in order to set it aside, and we think the part he sets aside is the part worth keeping. Nine readers in ten never see the label, and the one field experiment at real scale pays the advertiser to keep it quiet.
Answer to the question
Why do readers miss native advertising labels?
Because they never look at the label. In two nationally representative United States samples, of 738 and 800 adults, fewer than one in ten readers recognised sponsored news as paid advertising. A meta-analysis of 61 papers then found that recognising a label on its own did not move evaluation, while understanding what the sponsorship meant did.
Key takeaways
- In two nationally representative United States samples, of 738 and 800 adults, fewer than one in ten readers recognised sponsored news as paid advertising.
- The University of Georgia's account of the founding experiment by Bartosz Wojdynski reports 17 of 242 viewers, 7 per cent, identifying the content as advertising, and readers seven times likelier to spot it when the label said advertising or sponsored content.
- A meta-analysis led by Martin Eisend, pooling 61 papers using 57 distinct data sets, found that recognition of sponsored content rose without moving evaluation, while understanding what it meant did move evaluation.
- In a randomised field experiment delivered to 138,436 newsletter subscribers, led by Anocha Aribarg, display advertising beat native on visual attention, brand recognition and site trust, while a less prominent disclosure earned more clicks.
The usual question about native advertising is whether the label works. That question assumes a reader who sees the label and then decides what to think about it. In two nationally representative American samples, meaning samples built to look like the adult population rather than a university class, fewer than one in ten readers got that far.
What should you check before reading on?
Three habits hold whatever else turns out to be true, so they come first. Read the small type above a headline before you read the headline, because that is where the word sits if it sits anywhere. Treat a phrase like brand voice or presented by as the weakest signal on the page, since the wordings that most raise recognition are the blunt ones, advertising and sponsored. And notice the device in your hand, because a phone makes the whole job harder.
Those three cost nothing. The rest is why they are needed.
This is a different failure from the one in what an AI label does to a product, where the label is read and then discounted. Readers who would not trust online ratings without a second look still take a sponsored story as editorial, because the sponsorship never registered.
The founding experiment is a good place to see the size of the problem. Bartosz Wojdynski and Nathaniel Evans, both at the University of Georgia, built twelve versions of a single sponsored news story, crossing four label wordings with three positions on the page. The university's own published account of the study reports that in the first experiment 17 of 242 viewers, 7 per cent, identified what they were reading as advertising. The same account reports readers were seven times more likely to identify it when the label used advertising or sponsored content rather than brand voice or presented by.
Position turned out to matter more than anyone had assumed. In the eye-tracking half of that work, the release reports that a label at the top of the page was looked at by 40 per cent of viewers, a label in the middle by 90 per cent, and a label at the bottom by 60 per cent. A label nobody reaches is a label that does nothing, and the top of the page is where regulators have tended to want it.
Those figures come from the university, not from the journal article, whose own summary gives no percentages.
Why is the label so rarely seen?
Because the finding survives the move out of a student pool, which is where most of these effects go to die. Michelle Amazeen at Boston University, working with Wojdynski, ran the question past a representative sample of 800 American adults and found that fewer than one in ten could recognise sponsored news as paid advertising. A second study of theirs, with a nationally representative sample of 738 adults, landed in the same place. Recognition was likeliest among younger and more educated readers who came to news for information.
Two different samples, two different years, the same answer.
The obvious objection is that readers have simply got better at this. Amazeen tested the opposite case, comparing phones with computers across two experiments, and found that readers with more motivation to process what they were reading did recognise native advertising more often, and that it was not enough to make up for the extra effort a mobile screen demands. Wanting to pay attention does not rescue you on a phone.
There is a sharper version still. Eva van Reijmersdal at the University of Amsterdam recruited 206 people in the Netherlands and analysed 133 of them, the 133 being those who correctly remembered whether a label had been present. In the full sample the disclosure did not activate the sense that this was advertising at all. Everything the experiment found, it found among the minority who had noticed.
Does Williams describe a blur or an inversion?
The seed for this piece is James Williams, whose Stand Out of Our Light makes the best short argument we know about what advertising did to the design of digital media. His claim sits in chapter 5, "Empires of the Mind", pages 26 to 40.
> "In previous media, advertising had largely been an exception to the rule of information delivery, but in digital media, it seemed to have broken down some essential boundary; it seemed now to have become the rule."
He puts product placement in the same bracket. The mechanism he gives is instrumentation rather than camouflage. Analytics platforms built to manage advertising, and he names Omniture, Coremetrics and Google Analytics, promoted advertising's own numbers, clicks and impressions and time on site, into the operating metrics of websites themselves. That is a claim about what sits on a dashboard, and it is a good one.
And then he raises native advertising in order to set it aside. The same paragraph says it "wasn't just that the line between advertising and nonadvertising was getting blurry, as with 'native advertisements'". For him the blur is the small version of the story.
We disagree, and the disagreement is the point of this article rather than a quibble about emphasis. The blur is the only part of the story anyone has run an experiment on.
Williams names native advertising in order to set it aside. We think the part he sets aside is the only part anybody has measured.
He also asks not to be read as evidence, twice, unprompted. The preface says the book "will be more exploration than argument" and quotes Emerson that "sometimes a scream is better than a thesis". Chapter 6 adds that he is "not making any sort of scientific claim or argument with these distinctions". Take him at his word: the framework is his, the findings have to come from elsewhere.
So the fair comparison is not Williams against the researchers. It is a strong argument about dashboards against a weaker argument about labels that happens to have been measured.
What changes when a reader does notice?
Something that costs more parties than you would expect. Wojdynski's own single-author experiment, with 343 participants, tested four features of a disclosure: how close it sat to the content, how visually prominent it was, how clearly it was worded, and whether it carried a logo. Recognition lowered judgements of the article's quality, attitudes to the sponsor, and the intention to share it. The logo raised recognition and also raised the odds that readers mistook the label itself for an ordinary banner advert.
The best evidence here is a meta-analysis, which is a study that pools the results of earlier studies to see what they say together. Martin Eisend at European University Viadrina and three co-authors pooled 61 papers using 57 distinct data sets. Disclosing sponsored content lowered brand attitudes, credibility and evaluations of the source, and raised recognition, resistance and what the field calls persuasion knowledge, meaning a reader's sense that they are being sold to.
Then comes the part that corrects the easy version of our argument.
Disclosure lowered brand attitudes, credibility and evaluations of the source, and separately, recognition of sponsored content rose without carrying evaluation with it. Those two statements are not in conflict, because they sit at different places in the same model. The lowered attitudes and credibility are the direct effect of disclosing. The mediated path is the separate question, and there recognition of sponsored content rose and did not carry evaluation with it, while understanding what the sponsorship meant did. Noticing a label and grasping it are two separate events, and the one the industry can count is the one that changed nothing.
| Evidence | Who was asked | What it measured | What it found |
|---|---|---|---|
| Wojdynski and Evans, 2016 | 242 viewers, plus an eye-tracking sample | Recognition, by label wording and position | 7 per cent recognised it, per the university's account |
| Amazeen and Wojdynski, 2019 and 2020 | 738 and 800 United States adults, representative | Recognition, and opinion of the publisher | Fewer than one in ten recognised it |
| Eisend and colleagues, 2020 | 61 papers, 57 data sets | Pooled effects of disclosure | Recognition rose, evaluation moved only with understanding |
| Aribarg and Schwartz, 2020, field | 138,436 newsletter subscribers | Clicks, by disclosure prominence | A quieter disclosure earned clicks, and native against display was nothing the study could detect |
| Aribarg and Schwartz, 2020, laboratory | 1,299 panel participants, 156 eye-tracked students | Attention, brand recognition, site trust | Native won click-through, display won the rest |
The Facebook and Instagram work points the same way and should be read for the platform it used. Sophie Boerman at Amsterdam traced a chain from recognising an advert to distrusting the post to declining to pass it on, in posts attributed to a celebrity. Van Reijmersdal and colleagues report, citing that paper, that 59 per cent of participants who saw a sponsored post did not recognise the disclosure when shown a picture of it later.
Who pays for a prominent label?
The advertiser pays, in cash, and the number comes from the field. Anocha Aribarg and Eric Schwartz at the University of Michigan ran the largest test of native advertising in this literature with the business newsletter Morning Brew, randomly assigning four disclosure conditions across 138,436 subscribers who received the email on a single morning in April 2017.
Making the legal disclosure more prominent cost clicks. Making it less prominent, while staying minimally compliant with the Federal Trade Commission's guidance, earned them.
In the field, at a scale of 138,436 subscribers, the label that works is the label that costs money.
Their laboratory studies, 1,299 panel participants browsing a constructed news site and 156 eye-tracked students, found the trade-off that the trade press rarely quotes. With position held constant, a native advert won click-through because it resembled the editorial around it. A display advert won on visual attention, on brand recognition, and on how trustworthy readers found the site.
The paper carries a declared interest, printed in it and recorded here as what it is: the second author is a minor investor in Morning Brew, which supplied the field data, and states the research was conducted without bias.
Set against that, the strongest case for labelling is also empirical, and it is not weak. In the representative sample of 738 adults, readers were most receptive to sponsored news when publishers were open about its commercial nature. Van Reijmersdal's experiment found two mechanisms running in opposite directions at once: a disclosure lowered trust in the brand, and it also made the post look more transparent, which lifted attitudes to the brand by a second route.
Both arms are real. Our reading is that they do not cancel, because they fall on different people, and the publisher's arm depends on a reader seeing the label in the first place.
Native advertising is cheap to label and expensive to label well.
How far does this evidence actually reach?
Not as far as the arguments built on it. Nothing in reach measures recognition of native advertising twice in the same way, so the fact that three studies land under one in ten is suggestive and is not a trend over time. We looked for a replication or a tracking study since 2022 and found none.
The field experiment is not a clean win either, and its authors say so. The damage to site trust showed up in the first study only among people who clicked the native advert. The brand-recognition effect from that study did not appear under repeated exposure in the second. And in the newsletter experiment itself, the native-versus-display differences were nothing the study could detect.
The platform work is about platforms. A sponsored Facebook post from a celebrity and a sponsored Instagram post are not a news page, and the outcomes measured there are sharing and brand attitude rather than money.
Williams cannot be cited for any of it, by his own instruction.
What survives all of that is narrow and worth having. Fewer than one reader in ten recognises the piece as advertising, which is not the same as failing to see a label. Among the minority who do, the brand loses and the publisher's position is genuinely mixed. And the one setting measured in the field rewards the advertiser for making the label quieter.
Which gives you the test, and it is not about wording. Before you ask whether a page is labelled, ask who would have to lose money for the label to be read at a glance. Then look at the page again.
Ask who paid, before deciding what it says.
Common questions
Why do readers miss native advertising labels?
Mostly because the label does not land, and on the eye-tracking evidence that is not the same as never looking at it. In two nationally representative United States samples, of 738 and 800 adults, fewer than one in ten readers recognised sponsored news as paid advertising. Position and wording both matter: on the university's account of the eye-tracking work, a label in the middle of a page was looked at by 90 per cent of viewers against 40 per cent at the top, and blunt words like sponsored worked better than brand voice.
Does labelling sponsored content protect readers?
Only the minority who notice it. A meta-analysis of 61 papers found that disclosure lowered source evaluations directly, while on the mediated path recognition rose without carrying evaluation with it and understanding did. In a Dutch experiment with 206 recruits, the disclosure did not activate the sense that this was advertising, and among those who noticed it both lowered brand trust and raised perceived transparency.
Is native advertising better than display advertising?
Not on the measures a brand should care about. In the 138,436-subscriber newsletter experiment led by Anocha Aribarg, the native-versus-display differences were nothing the study could detect, though a quieter legal disclosure earned clicks. In their two laboratory studies, native won click-through when page position was controlled, while display won visual attention, brand recognition and how trustworthy readers rated the site.
What does James Williams actually claim about advertising?
That advertising moved from being an exception inside a medium to being its organising rule, because the analytics built for advertising turned clicks and impressions into the operating metrics of websites. He raises native advertising only to set it aside, and he twice asks not to be read as making a scientific claim. We think the blur he sets aside is where the measurable harm sits.
Sources
Wojdynski, B. W., & Evans, N. J. (2016). 'Going Native: Effects of Disclosure Position and Language on the Recognition and Evaluation of Online Native Advertising.' Journal of Advertising, 45(2), 157-168. doi:10.1080/00913367.2015.1115380. [ABSTRACT READ IN FULL at tandfonline.com in the built-in browser; FULL TEXT PAYWALLED AND NOT READ. Author order confirmed at the Crossref item record: Wojdynski first and corresponding, Evans second, both University of Georgia Grady College. DOI CORRECTION: the DOI is .2015.1115380, not .2016.1115380; Crossref gives published online 14 December 2015 and print issue 2 April 2016. The recognition figures are NOT in the abstract and are attributed in the article to the University of Georgia's own published account, UGA Today, 16 December 2015, by Sarah Freeman, read in full: 17 of 242 viewers, 7 per cent, in the first study; seven times likelier recognition with 'advertising' or 'sponsored content' wording; and 40, 90 and 60 per cent looking at top, middle and bottom disclosures. The second study's figure is not printed because the release says 17 per cent and a secondary summary says 18.3 per cent and the disagreement is unresolved]
Wojdynski, B. W. (2016). 'The Deceptiveness of Sponsored News Articles: How Readers Recognize and Perceive Native Advertising.' American Behavioral Scientist, 60(12), 1475-1491. doi:10.1177/0002764216660140. [Abstract read in full at OpenAlex. Sole author, corresponding, University of Georgia]. A between-subjects experiment with 343 participants testing proximity, visual prominence, wording clarity and logo presence; recognition lowered perceptions of article quality, attitude toward the sponsor and intent to share, while logo presence raised recognition and also raised misperception of the label as a stand-alone display advertisement.
Amazeen, M. A., & Wojdynski, B. W. (2020). 'The effects of disclosure format on native advertising recognition and audience perceptions of legacy and online news publishers.' Journalism, 21(12), 1965-1984. doi:10.1177/1464884918754829. [Abstract read in full at OpenAlex. Author order confirmed at OpenAlex: Amazeen first and corresponding, Boston University; Wojdynski second, University of Georgia. Online 2018, print issue December 2020; the 2020 issue is printed and the 2018 online date noted]. A representative sample of 800 United States adults; fewer than one in ten recognised native advertising, and those who did held lessened opinions of the publisher and of the institution of advertising. NOT VERIFIED: which design characteristics were manipulated, and the legacy against digital-first comparison, which the abstract poses without answering.
Amazeen, M. A., & Wojdynski, B. W. (2019). 'Reducing Native Advertising Deception: Revisiting the Antecedents and Consequences of Persuasion Knowledge in Digital News Contexts.' Mass Communication and Society, 22(2), 222-247. doi:10.1080/15205436.2018.1530792. [Abstract read in full at OpenAlex. Amazeen first and corresponding, Boston University; Wojdynski second, University of Georgia]. A nationally representative sample of 738 United States adults; fewer than one in ten readers recognised the sponsored article as advertising, recognition was likeliest among younger and more educated consumers reading for information, and news consumers were most receptive to native advertising when publishers were transparent about its commercial nature.
Amazeen, M. A. (2021). 'Native Advertising in a Mobile Era: Effects of Ability and Motivation on Recognition in Digital News Contexts.' Digital Journalism, 11(6), 1130-1153. doi:10.1080/21670811.2020.1860783. [Abstract read in full at OpenAlex. Sole author per the OpenAlex authorship record, Boston University, corresponding. THE YEAR IS UNRESOLVED: OpenAlex gives 2021 while volume 11 issue 6 is a 2023 print issue; published online 2021, in volume 11 issue 6]. Two experiments among United States adults: greater motivation to process content raised recognition of native advertising and was not enough to compensate for the greater cognitive effort a mobile device demands. NOT VERIFIED: the sample size of either experiment.
Eisend, M., van Reijmersdal, E. A., Boerman, S. C., & Tarrahi, F. (2020). 'A Meta-Analysis of the Effects of Disclosing Sponsored Content.' Journal of Advertising, 49(3), 344-366. doi:10.1080/00913367.2020.1765909. [Abstract read in full at OpenAlex. Author order confirmed at OpenAlex: Eisend first and corresponding, European University Viadrina; van Reijmersdal second and Boerman third, both University of Amsterdam; Tarrahi last, Viadrina]. 61 papers using 57 distinct data sets: disclosure reduced brand attitudes, credibility and source evaluation and increased recognition, persuasion knowledge and resistance; the path model showed recognition of sponsored content increasing without influencing evaluation, while understanding of sponsored content did influence evaluation. NOT VERIFIED: any pooled effect size, the abstract reporting directions only.
Aribarg, A., & Schwartz, E. M. (2020). 'Native Advertising in Online News: Trade-Offs Among Clicks, Brand Recognition, and Website Trustworthiness.' Journal of Marketing Research, 57(1), 20-34. doi:10.1177/0022243719879711. [FULL TEXT READ, author copy at gwern.net. Author order confirmed at OpenAlex: Aribarg first and corresponding, Schwartz second, both Ross School of Business, University of Michigan; OpenAlex dates it 2019 online and the print issue 57(1) is 2020]. Study 1: 1,299 participants from the Lucid panel browsing a constructed news site. Study 2: 156 eye-tracked undergraduates under repeated exposure. Study 3: a randomised field experiment successfully delivered to 138,436 Morning Brew subscribers on 19 April 2017. Native won click-through with position controlled; display won visual attention, brand recognition and website trustworthiness; and advertisers could increase clicks by making disclosures less prominent while staying minimally compliant with FTC guidelines. LIMITS read directly: the trustworthiness effect appeared in Study 1 only among those who clicked, the Study 1 brand-recognition effect did not appear under repeated exposure in Study 2, and neither Study 3 nor the Web Appendix D pretest detected the native-versus-display format effects. DECLARED INTEREST, printed verbatim: 'The second author is a minor investor in Morning Brew, provider of data for Study 3, and maintains that the research was conducted without bias.'
Van Reijmersdal, E. A., Brussee, E., Evans, N. J., & Wojdynski, B. W. (2023). 'Disclosure-Driven Recognition of Native Advertising: A Test of Two Competing Mechanisms.' Journal of Interactive Advertising, 23(2), 85-97. doi:10.1080/15252019.2022.2146991. [FULL TEXT READ, CC BY-ND final published version at the University of Amsterdam repository. Author order from the published version: Van Reijmersdal first, Brussee second, both Amsterdam; Evans third and Wojdynski fourth, University of Georgia]. 206 participants recruited in the Netherlands, 133 analysed, those being the people who correctly recalled whether a disclosure was present. In the total sample the disclosure did not activate conceptual persuasion knowledge. Among those who noticed it, the disclosure lowered brand attitudes and purchase intention through activated persuasion knowledge and simultaneously raised them through perceived transparency.
Boerman, S. C., Willemsen, L. M., & Van Der Aa, E. P. (2017). '"This Post Is Sponsored": Effects of Sponsorship Disclosure on Persuasion Knowledge and Electronic Word of Mouth in the Context of Facebook.' Journal of Interactive Marketing, 38(1), 82-92. doi:10.1016/j.intmar.2016.12.002. [Abstract read in full at OpenAlex. Author order confirmed at OpenAlex: Boerman first, University of Amsterdam; Willemsen second, HU University of Applied Sciences Utrecht; Van Der Aa last, Amsterdam]. A two-by-two experiment on Facebook posts: sponsorship disclosure influenced the use of persuasion knowledge only for a celebrity source, and a disclosure began a chain from recognising advertising to distrusting the post to lower intention to pass it on. NOT VERIFIED: the sample size. The 59 per cent non-recognition figure is printed in the article as reported by Van Reijmersdal and colleagues in 2023, citing this paper, and was not verified in the paper itself.
Yang, G. (2022). 'The effects of sponsorship disclosures, advertising knowledge, and message involvement in sponsored influencer posts.' International Journal of Advertising, 41(8), 1502-1520. doi:10.1080/02650487.2022.2029260. [Abstract read in full at OpenAlex. Sole author, corresponding, Oakland University, Michigan]. An online experiment with 296 participants on Instagram posts: enhanced advertising knowledge raised recognition regardless of disclosure prominence, prominent disclosures outperformed subtle ones where knowledge was not enhanced or involvement was low, and recognition elicited resistance and less favourable brand attitude. SCOPE LIMIT stated in the article: influencer posts on Instagram, not sponsored news articles.