Organizations
Five companies appear in the records: Coca-Cola, Starbucks, McDonald's, Netflix and TikTok. For each one this page separates what the sources document from what they do not show. A company appearing in a study does not mean the company used the study, and a plausible psychological reading is not evidence of intent.
Coca-Cola
Coca-Cola appears in four substantial records and in one myth check. They have different authors and different evidence: a study the company paid for, company accounts of its own campaigns, and an independent brain-scan study that used the drink as a stimulus.
What the sources document
In 1911 the company’s lawyers commissioned Harry Hollingworth to test caffeine for a court case, and his preface says Coca-Cola financed the experiments and their publication. The company’s own pages describe the New Coke change of 1985 and the Share a Coke campaign. An independent academic team scanned volunteers tasting Coca-Cola and Pepsi with and without brand cues (2004).
What they do not show
The 2004 brain-scan study was not commissioned or funded by either company, so it is not evidence of company practice. The 1957 claim that flashed messages sold Coca-Cola and popcorn rests on no published data, and no source shows the company took part. No company document read says New Coke or Share a Coke applied a named psychological theory; the sales figures are the company’s or its agency’s own.
Records
- 1911March
Hollingworth's caffeine study for Coca-Cola and the Chattanooga trial
The Coca-Cola Company's attorneys hired Harry Hollingworth to test caffeine on 16 volunteers for the 1911 Chattanooga trial. The company financed the placebo-controlled, blinded 40-day study, which reported a genuine stimulant effect. It did not settle the legal question.
- 1957
Vicary's subliminal popcorn claim
In 1957 James Vicary said he had boosted popcorn and Coca-Cola sales in a cinema with messages flashed too briefly to see. He published no data, and a 1992 column says he later admitted his data were too small to be meaningful. The claim is unsupported.
- 1985April
New Coke and the return of Coca-Cola Classic
Coca-Cola changed its formula in April 1985 after taste tests that the company says involved nearly 200,000 people, then restored the original in July. The company says its tests missed attachment to the brand. Other readings differ.
- 2004
The Coke and Pepsi brain-scan study and the Pepsi paradox
Academic researchers ran taste tests and brain scans with Coca-Cola and Pepsi, with and without brand cues. Brand knowledge changed stated preference and brain responses. The brands were stimuli. The funding statements read name a foundation and US health agencies, not either company.
Starbucks
Starbucks has two records where the company is the subject. A third, the goal-gradient coffee-card paper, is related research that is not about Starbucks.
What the sources document
The company’s own releases and filings describe its card-based rewards scheme, its merger into My Starbucks Rewards and the 2016 change to earning Stars by spend. Three economists used Starbucks transaction data to study New York City’s calorie-posting rule; they say the company supplied the data and that they have no consulting relationship with it.
What they do not show
The goal-gradient paper studied a campus cafe that is not part of any chain, so it is not attributed to Starbucks. No Starbucks document read says its rewards scheme was designed from that research or from any named theory. Membership and card-balance figures are the company’s own reporting and are not independently audited.
Records
- 2008
Starbucks Card Rewards and My Starbucks Rewards
Starbucks ran a card-based rewards scheme from 2008, merged it into My Starbucks Rewards in 2009 and in 2016 began awarding Stars by money spent. Membership and card-balance figures are the company's own reporting. No company document read cites a psychological theory.
- 2011
Calorie posting at Starbucks: the Bollinger, Leslie and Sorensen study
Three economists used Starbucks transaction data from New York City, Boston and Philadelphia to measure what happened after mandatory calorie posting began in New York. Calories per transaction fell about 6 percent, mostly from food. The authors say they persuaded Starbucks to provide the data.
McDonald’s
McDonald’s records cover its Happy Meal, personalized menus and app rewards, an independent packaging experiment, and two public-policy cases involving several chains.
What the sources document
The Happy Meal has competing origin accounts. Company announcements also describe context-sensitive menu displays in 2019 and a points program in 2021. In an independent Stanford experiment with 63 preschoolers, a news summary reports more children rating the McDonald’s-wrapped version of four identical foods tastier; the funders named were Stanford and a health foundation, not the company. New York’s calorie labeling and San Francisco’s children’s-meal ordinance are documented policy cases in which findings were mixed.
What they do not show
The branding study shows an effect of packaging on young children’s ratings in one experiment. It does not show what McDonald’s intended, and nothing read says the company commissioned or applied it. The labeling studies cover several chains, and their results differ from one another, so they do not settle whether menu labels change what people buy.
Records
- 1979
The McDonald's Happy Meal
McDonald's tested a boxed children's meal with a toy from 1977 and launched it across the United States in 1979. Press accounts disagree on who invented it. A 2012 FTC release covers food marketing to children in general and does not name McDonald's.
- 2007August
Preschoolers taste the same food in McDonald's wrappers
Stanford researchers gave 63 preschoolers identical foods in McDonald's or plain wrappers. The original abstract reports an overall preference for branded samples. The brand was a stimulus; the study does not establish company intent or sponsorship. A later reassessment also points to packaging appearance.
- 2009October
New York City calorie labeling: Elbel and Dumanovsky studies
Two studies of New York City's 2008 menu calorie rule reached different results. Elbel's receipts study found no change in calories bought. Dumanovsky's surveys found no overall decline but falls at some chains, including McDonald's. Both studied customers, not McDonald's decisions.
- 2010November
San Francisco's Happy Meal toy ordinance
In 2010 San Francisco passed an ordinance restricting toys with children's meals that exceeded nutrition limits. McDonald's opposed it. A later study of two unnamed chains found the toy sold for 10 cents extra and mixed effects on calories.
Netflix
Netflix is the best documented recommender case here, but most of the documents are its own. Independent work covers viewing and sleep and the limits of the company’s claims.
What the sources document
The company ran the Netflix Prize from 2006 and published descriptions of its recommender, its artwork testing and its autoplay and viewing-prompt features. A survey study links binge viewing with poorer sleep.
What they do not show
The recommender and artwork descriptions are company self-report and have not been audited, and several of Netflix’s own posts could not be opened, so some claims rest on outlets quoting them. No Netflix document read names a psychological mechanism behind autoplay. The sleep study is correlational and did not test autoplay, so it does not show that Netflix design caused later bedtimes.
Records
- 2006October
The Netflix Prize
Netflix released over 100 million ratings and offered one million dollars to any team that beat its Cinematch recommender by a further 10 percent. The contest framed recommending as predicting star ratings, a framing Netflix later moved away from.
- 2015December
Netflix describes its recommender system and its business value
Two Netflix employees described the company's recommender, said it influences about 80 percent of streamed hours and estimated its value at over one billion dollars a year. These are company claims; the paper does not show how the estimate was made.
- 2017April
Netflix autoplay, "Are you still watching?" and sleep
Netflix can play the next episode automatically and asks "Are you still watching?" after a run of episodes. A survey links binge viewing to poorer sleep, but it is correlational and did not test Netflix features. No document opened shows design intent.
- 2017December
Netflix chooses the artwork each member sees
Netflix posts describe testing several images per title and then choosing the image shown to each member from their viewing history. The performance claims are Netflix's own. In 2018 users alleged racial targeting; Netflix denied it.
TikTok
TikTok’s records separate what the company says, what outside researchers measured, and what litigants allege.
What the sources document
TikTok has published its own account of how the For You feed ranks videos and announced a default 60-minute prompt for under-18s in 2023. Two researchers ran scripted accounts on the TikTok website in 2021 and reported which actions changed the feed, with following a creator the strongest. US state attorneys general filed suits in 2024 whose complaints cite internal documents.
What they do not show
The company’s explanation is a self-report and gives no weights; it is not an independent audit of the algorithm. The scripted-account study is limited to a few factors on the web version and is not a full audit. The litigation claims are allegations as reported by news outlets, the complaints could not be opened, and TikTok disputes them. A Wall Street Journal bot experiment was left out because its details could only be found second-hand.
Records
- 2020June
TikTok explains how the For You feed recommends videos
TikTok's own account of its For You feed lists three groups of signals and says strong signals count for more than weak ones. It is a company description, not an audit, and it gives no weights.
- 2022January
Boeker and Urman's sock-puppet audit of TikTok personalization
Researchers ran scripted accounts on the TikTok website to test what changes the For You feed. Following a creator had the strongest effect. The abstract and main text disagree on the order of liking and watch time. This is a limited web audit, not a full test of the algorithm.
- 2023March
TikTok's default 60-minute prompt for under-18s
TikTok announced a default 60-minute daily limit for under-18s that a passcode can override. A state complaint, as reported, alleges its own test cut use by about 1.5 minutes a day. TikTok disputes the complaint. No independent evaluation was opened.
- 2024October
State lawsuits against TikTok and the Kentucky filing
US state attorneys general sued TikTok in October 2024. Kentucky's complaint, as reported, alleges an internal finding that habit forms after 260 videos. These are allegations. TikTok calls the quotes misleading and says it has safeguards.
Other organizations
These organizations appear in records where the evidence permits, as context rather than as full cases. Each record states the exact relationship.
- Facebook (Meta)
- Facebook's emotional contagion experiment; Bakshy, Messing and Adamic on news exposure on Facebook; The Meta 2020 US election studies on feeds
- YouTube
- Deep Neural Networks for YouTube Recommendations; YouTube radicalisation audits and YouTube's borderline content post
- Deep Neural Networks for YouTube Recommendations
- Tesco
- Tesco launches Clubcard
- American Airlines
- American Airlines launches AAdvantage
- J. Walter Thompson
- Watson joins J. Walter Thompson
- American Tobacco Company
- The Easter Parade smokers of 1929 (torches of freedom)
- Pepsi
- New Coke and the return of Coca-Cola Classic; The Coke and Pepsi brain-scan study and the Pepsi paradox