The framing effect is the tendency to make different choices when the same facts are described in different ways, such as "200 of 600 people will be saved" versus "400 of 600 people will die."
In marketing it shows up whenever two descriptions of the same price, discount or outcome produce different conversion rates.
The effect is well replicated in the lab, but in A/B tests a new frame only wins when it changes what the shopper notices or understands.
This page covers what the research on framing bias actually shows, then what real tests in the LiftArchive database show when stores, SaaS companies and apps reframe prices and copy. It sits alongside our other psychology pages.
Framing bias: what the research shows
Framing bias is not a single quirk. It is a family of effects in which the description, not the substance, moves the decision.
Three findings matter most for anyone writing copy or setting prices.
The original experiment
Amos Tversky and Daniel Kahneman published the classic demonstration in Science in 1981.
Respondents imagined a disease expected to kill 600 people and picked between two programs with the same expected outcome.
| Frame | Sure option | Risky option |
|---|---|---|
| Gain frame (N = 152) | "200 people will be saved": 72% chose it | 1/3 chance all 600 saved: 28% |
| Loss frame (N = 155) | "400 people will die": 22% chose it | 1/3 chance nobody dies: 78% |
The two problems are identical. Only the wording changed, and the majority flipped from playing safe to taking the gamble.
Tversky and Kahneman tied this to loss aversion and the broader prospect theory account of how people value gains and losses.
The same paper has a finding that is closer to pricing.
68% of respondents would drive 20 minutes to save $5 on a $15 calculator, but only 29% would make the same trip to save $5 on a $125 calculator.
The saving was identical in dollars. People judged it relative to the price.
Three kinds of framing
Irwin Levin, Sandra Schneider and Gary Gaeth argued in a 1998 review that mixed results in the literature came from lumping different effects together.
They split framing into three types, and the split is useful for marketers.
| Type | What changes | Classic example | Marketing version |
|---|---|---|---|
| Risky choice framing | Willingness to take a risk | Lives saved vs lives lost | Guarantees, trials, "cancel anytime" |
| Attribute framing | How good one feature seems | "75% lean" vs "25% fat" beef | Percent off vs dollars off, per-day prices |
| Goal framing | How persuasive a message is | Benefits of acting vs costs of not acting | Benefit copy vs "what you lose without it" |
The beef study comes from Levin and Gaeth (1988).
Consumers rated beef labeled "75% lean" more favorably than beef labeled "25% fat," and the gap shrank once they actually tasted it.
That last detail matters for ecommerce: a frame shapes expectations, but product experience can override it.
For pricing, the closest research is John Gourville's "pennies-a-day" study (1998).
Restating a yearly cost as a small daily amount made transactions look better in lab studies across a range of products, even though the actual payment stayed the same.
How solid is the evidence?
Better than most biases. The disease problem was one of the classic effects retested in the Many Labs projects, and Many Labs 2 reproduced it, though only about half as strong as the original.
A critique that the effect was a wording artifact (people reading "200 will be saved" as "at least 200") was tested by Simmons and Nelson at Data Colada, who still found a strong effect after adding the word "exactly."
Goal framing is much weaker. A meta-analysis by O'Keefe and Jensen (2007) of 93 studies on disease-prevention messages found gain-framed messages only slightly more persuasive than loss-framed ones (r = .03).
So "fear sells" is not a law. It depends on the product and the audience.
The lab also differs from a store in one important way.
Lab participants see exactly one frame and must answer.
Shoppers skim, compare tabs and often do not read the line you changed.
That is why lab effect sizes do not carry over to conversion rates.
Framing effect marketing: what A/B tests show
In our database, framing tests cluster around four questions: how to state a discount, whether to show the math, how to state a recurring price, and whether copy should stress gains or losses.
Here is what real tests found, with the numbers as reported by whoever ran them.
Convertibles
@DaveDiederen
GrowthRock
Conversion FanaticsPercent off vs dollars off
The calculator problem predicts that the same saving feels bigger when it is large relative to the price.
Dave Diederen tested this directly on one brand.
Showing savings as a percentage instead of a dollar amount lifted conversion rate 7.55% and revenue per visitor 6.42%.
Dave Diederen then tested the common rule of thumb that dollar amounts work better for large savings.
On the same brand, with savings above $100, the dollar version came out net negative: conversion rate fell 1.08% and revenue per visitor did not change.
Dave Diederen's guess is that the large dollar figure looked too good to be true.
Popups point the other way. Convertibles ran an A/B/n test for an adult wellness brand where $30 off a $150+ order beat 15% off: +43% email capture, +21% AOV and +$45,000/month in revenue.
But that variant also changed the images, the layout and added community-deals copy, and the minimum order did part of the work.
It is not a clean test of dollars versus percent.
The takeaway: there is no universal winner. Test the frame on your own price points, and change only the number format when you want to learn something about framing.
Do the math for the shopper
Several of the cleanest wins come from replacing a frame the shopper has to decode with a plain number.
During Black Friday, a luxury pajama brand showed "25% off with code CHEER" under the regular price.
Convertibles tested showing the crossed-out price next to the calculated sale price instead, worth +$102,350/month in revenue.
DRIP Agency saw the same direction with BLACKROLL, where an explicit euro savings line under the bundle price lifted revenue per user 1.8%.
Stone Creek Coffee, tested by Blend Commerce, added "Save X% per oz" labels on larger bag sizes and saw a 41% conversion rate lift.
Both cases turn a hidden fact into a visible one.
That is a different mechanism from the lab: the frame works because it supplies information, not because it tricks anyone.
More examples are on our pricing and discount tests hub.
When the frame adds nothing, nothing happens
GrowthRock ran the best natural experiment on this in our database.
On a large apparel site where nearly every product had the same 30% discount, adding "You Save #%" next to prices did nothing: all three versions landed within 0.3% of each other across 280,000 visitors per variation.
GrowthRock reran it later, when discounts varied by product.
This time percent-off labels won: +2.57% conversion rate and +2.54% revenue per visitor across 700,000 visitors.
The frame only helped once it told shoppers something they could not already infer.
Per-month, per-week and per-unit framing
Gourville's pennies-a-day idea is the most tested frame on app paywalls.
Mojo, a video editing app, showed the annual plan as a monthly price billed annually.
New revenue per paywall impression rose 45% in Brazil, 26% in Mexico and 8% in the USA.
Stardust, a cycle tracking app, added a discount badge and a weekly-equivalent price to its yearly plan and reported +50% conversion and +62.5% proceeds per user.
Since two things changed at once, the weekly price alone cannot take full credit.
More on this pattern is in our paywall tests.
Smaller-unit framing does not always help. For a product bought in large quantities, Conversion Fanatics found that hiding the savings on quantity options lifted conversions 25.1%, and a per-unit-price version lifted them 12.9%.
Too many numbers can make a simple choice feel like a math problem.
Our page on anchoring covers how the first number shown sets the reference point.
Same value, different packaging
Attribute framing also applies to offers and plans.
Response Capture kept a $500 sweepstakes budget and offered ten $50 gift cards instead of twenty $25 cards; sign-ups rose 31%.
A retail subscription brand working with Optimizely presented its free trial as part of the plan rather than an add-on and saw a 37% lift in enrollment.
Powtoon, testing with VWO, changed how storage was described on its pricing page.
Showing the Business plan at 10 GB instead of "unlimited", next to the Pro plan's 2 GB, lifted revenue 27.9%.
A concrete number made the gap between plans easier to judge.
Gain vs loss copy
Goal framing is where marketers most often overclaim.
One strong result: Dave Diederen tested four product description angles for a Shopify brand, and the fear angle, describing the health risks of not using the product, won with +32.75% conversion rate and +41.90% revenue per visitor.
Small copy changes can also work in the positive direction.
Rewording a cart message from "Free shipping will be applied at checkout" to "Congrats! You unlocked FREE shipping!" lifted revenue per visitor 13.93%.
Both are single tests on single stores. Given the weak goal-framing evidence from research, treat each as a hypothesis to retest, not a rule. Our microcopy guide has more examples.
How to test a frame properly
- Change one thing. If the new frame ships with new images or a new offer, you will not know which part worked.
- Keep the substance identical. $30 off $150 is not the same offer as 15% off with no minimum. A true framing test keeps price, terms and product the same.
- Measure revenue, not just clicks. A frame that makes a deal look bigger can raise conversion and lower AOV, as Dave Diederen's percentage test did (-1.04% AOV).
- Expect smaller effects than the lab. Write the expected size into your test hypothesis and size the test to detect it.
- Do not hide the truth. Framing that misleads, such as a per-day price that hides an annual commitment, crosses into dark patterns and tends to raise refunds and chargebacks.
Framing is one of many cognitive biases that shape how people read a page.
The research shows it is real. The tests show it pays off mainly when the new frame makes the value easier to see.