The Warm-Cup Study: Does a Hot Drink Make People Like You? (Probably Not)

We sell a $35 logo mug gift box, printed, packed and shipped to each person's address, so a study saying a warm cup makes people like you would suit us. It didn't replicate, and we don't claim it. This page reports the original, three published replication reports and the original authors' reply, with sample sizes.

Key facts

• 41 and 53 people: the original two experiments' samples. Their effects were large (r = .28 and .31) and "barely statistically significant" (Chabris et al., 2019, describing Williams & Bargh, 2008).
• 861 people, three labs, no effect: a 2014 replication of the heat-pad experiment did not find it "in any individual laboratory or when considering the results of the three replications together" (Lynott et al., 2014).
• More than triple the samples, near-zero effects: a 2019 replication of both experiments, with 128 and 177 people and double-blind procedures, found r = −.03 and .02 (Chabris et al., 2019).
• 127 people in 2024: "We found no main effect of physical warmth" (Koppang et al., 2024, Royal Society Open Science).
• The original authors' reply argues the replication's coffee was too hot, and calls its own re-analysis "a post hoc story that requires further testing" (Bargh & Melnikoff, 2019).

What the 2008 Warm-Cup Study Claimed

Lawrence Williams and John Bargh tested whether physical warmth spills over into social warmth. Their abstract in Science (October 24, 2008) describes two experiments: "In study 1, participants who briefly held a cup of hot (versus iced) coffee judged a target person as having a 'warmer' personality (generous, caring); in study 2, participants holding a hot (versus cold) therapeutic pad were more likely to choose a gift for a friend instead of for themselves" (Williams & Bargh, 2008).

They hypothesized that it would happen "without the person's awareness of this influence." The samples were small. Study 1 had 41 participants and Study 2 had 53 (Chabris et al., 2019). The 2024 replication team computed Study 1's effect size from its reported statistics (F(1,39) = 4.08, p = 0.05): Cohen's d = 0.65, which they call a medium effect size (Koppang et al., 2024). (Chabris and colleagues describe the original effects as large; the two teams use different measures.) A p-value of 0.05 sits right at the usual cut-off for calling a result significant, which is one reason a larger replication matters.

The Replications, Side by Side

Every replication here had a larger sample than the original experiment it re-tested.

Replication 1: Three Labs, 861 People (2014)

Dermot Lynott and six colleagues ran "three high-powered, independent replications of Study 2". Participants evaluated hot or cold instant therapeutic packs, then chose a reward framed as "a treat for a friend" or "a treat for the self". The prediction was that the hot pack would make the friend's treat more likely.

It didn't, anywhere: "We did not replicate the effect in any individual laboratory or when considering the results of the three replications together (total N = 861). We conclude that there is no evidence that brief exposure to warm therapeutic packs induces greater prosocial responding than exposure to cold therapeutic packs" (Social Psychology, 2014).

Note what this experiment was about: choosing a gift. If warmth made people more generous with gifts, this is where it should have shown up, with about 16 times the original sample.

Replication 2: Both Experiments, Triple the Sample (2019)

Christopher Chabris, Daniel Simons and three colleagues went back to both experiments, coffee included. Their abstract sums up the original and their result in three sentences:

"These experiments featured large effects (r = .28 and .31), small sample sizes (41 and 53 participants), and barely statistically significant results. We attempted to replicate both experiments in field settings with more than triple the sample sizes (128 and 177) and double-blind procedures, but found near-zero effects (r = −.03 and .02). In both cases, Bayesian analyses suggest there is substantially more evidence for the null hypothesis of no effect than for the original physical warmth priming hypothesis." (Social Psychology, 2019)

In plain terms: an effect that looked large in a small sample shrank to about zero in a sample three times bigger. In a double-blind study, neither the participants nor the people running the sessions know who is in which condition, which guards against experimenters nudging the result without meaning to.

Replication 3: Hot Tea, Ice Water and 127 Students (2024)

A team in California and Norway re-ran the warm-cup experiment (Study 1) in 2024 and added three questions: does it matter whether participants know the original study, how warm their own personality is, or the gender of the person they rate? Participants held "either a cup of hot tea (about 50°C/122°F) or ice-cold water (about 10°C/50°F)" while being led to the lab, a trip of about 3 to 4 minutes by elevator. In the lab, about half read the original paper's abstract before rating the stranger.

The result: "We found no main effect of physical warmth, and very little evidence for any moderating effects. It is clear from this and other recent studies that the embodiment effect is not simple to replicate and, therefore, is difficult to exploit for practical purposes" (Koppang et al., 2024).

"Difficult to exploit for practical purposes" is the line any seller of mugs, coffee or cocoa should read twice.

What the Original Authors Say

John Bargh and David Melnikoff replied to the 2019 replication in the same journal. Their points, in their words (Bargh & Melnikoff, 2019; PDF via Yale's ACME Lab):

• The coffee study is one result among many. "The Williams and Bargh (2008) coffee study is one data point among many that bear on this question." They cite brain-imaging and body-temperature research on physical and social warmth.
• The replication's coffee may have been too hot. By the reply's account, the replication team bought coffee and ran the study right outside the shop, so "those participants likely held a piping hot coffee, unlike in the original study." They argue "'hot' is not the same as 'more warm'."
• A suggestive split, by their own account unproven. Broken down by how many people had held a cup before, the 4th and 5th holders of the hot coffee rated the stranger warmer than cold-coffee holders did; "this effect is not statistically significant, but close: p = .08." They add: "Of course, this is a post hoc story that requires further testing."
• A narrower claim. In their conclusion, "the question of the reliability of the coffee study paradigm bears on – but is not equivalent to – the reliability of the physical-to-social warmth effect."

They wrote that they were discussing a pre-registered study with the Chabris team to test the temperature explanation. We found no published result of that test as of October 29, 2026.

So, Does a Hot Drink Make People Like You?

Probably not, at least not in the way the 2008 study described. Three replication reports, each larger than the experiment it re-tested, found no main effect; the original authors argue for a narrower effect at the right temperature, which has not been shown in a pre-registered test that we found.

How to Read a Study Like This (Our Checklist)

Our suggestion, for any study a gift seller quotes, us included:

1. How many people? 41 is small for a surprising claim about everyday behavior.
2. How close to the line? p = 0.05 is the usual cut-off for "significant", so a result right at it is the one a retest most needs to check.
3. Has anyone else repeated it? A larger, independent replication outweighs one original.
4. Was the test planned in advance? A split of the data found after the fact, like the p = .08 cup-order split, is a hypothesis, not a finding (its authors said so).
5. Does it measure what the seller claims? Rating a stranger in a lab isn't liking a business.

What a Mug Gift Can Honestly Claim

Not warmth priming. What a gift can claim rests on other evidence, labeled for strength:

• A small gift changes behavior in field experiments, through reciprocity, not temperature. A chocolate with the check raised tips from 15.1% to 17.8% across 92 dining parties (Strohmetz et al., 2002): the restaurant mint study, from the paper. More: 9 studies on tiny gifts with outsized results.
• A mug gets used. In ASI's 2026 Ad Impressions Study, a promotional-products industry survey, 87% of end-users said they would keep a branded drinkware item "because of its usefulness", and more than three quarters said they would use one at least weekly and keep it at least a year (ASI). Self-reported, and no sample size is given in the article.
• Recipients favor practical gifts. Givers weight how desirable a gift is; receivers weight how practical it is (Baskin et al., 2014). More: practical beats impressive.

The evidence on whether client gifts work at all, both sides: do client gifts actually work?

What to Know Before You Order

• No claim about warmth. We don't say our mug, cocoa or truffles make anyone feel warmly toward you. We state what the box holds and what it costs.
• The treats. LINDOR truffles contain milk, soy and barley malt (gluten); read the labels.
• Gift limits. Each box is $35, delivered. Some employers and industries cap the gifts their people may accept; the limits, quoted from their sources: client gift limits by industry.
• Where we ship. The lower 48 states and Washington, DC, except Texas for now. The order page flags any row we can't ship to, so you can fix it or take it out.

Where Personalized Gifts Delivered Fits: A Useful Mug and Your Words, No Warmth Claim

What we sell rests on the parts that hold up: a useful object, a small treat and a note in your words. Each Personalized Gifts Delivered box holds:

• The mug: an 11 oz white ceramic mug with your logo printed small (Small, up to 1½ in wide, is the default), on the front or on both sides at no extra cost.
• The treat: a 1.5 oz packet of Ghirardelli premium hot cocoa and 4 Lindt LINDOR milk chocolate truffles, packed inside the mug, in a matte-black gift box with a magnetic lid.
• Your note: your message, printed on a textured card that greets each person by first name, in an envelope closed with a gold wax-seal sticker, lying on the lid so it's read before the box opens.

Putting 25 of these together yourself takes, by our estimate, about 9 hours of hands-on work plus a week or more waiting for supplies. Per box it comes out about even with our $35 (about $38 with USPS counter postage, about $34 with commercial postage), with $1,192.84 paid up front for the mugs, the retail supply packs and 25 postage labels, as of Oct 29, 2026 (DIY vs ordering, itemized).

Add your logo (or type your website and we'll find it), or a photo or a few words for a personal gift; check the preview of your mug and card; add your address list; and pay $35 per box. We print, pack and ship each box to its own address. It's $35 per box, delivered, with no minimum: a logo mug gift box with Ghirardelli cocoa, Lindt truffles and a personal card. When the boxes ship, you get one email with a spreadsheet of recipient, ship date, carrier, tracking number and $35 per box.