Is framing always manipulation?

No. Every explanation uses some frame. The concern is whether the presentation helps people understand the choice or obscures relevant aspects.

Can showing both frames help?

It can make the equivalence clearer, though it does not guarantee elimination of every presentation effect.

Is framing the same as anchoring?

They are related topics in judgment research, but anchoring concerns influence from an initial reference value, while framing concerns the presentation of the decision.

AI Glossary / Graphs and symbolic AI / Intermediate

Framing effect

Definition

The framing effect is a change in judgment or preference caused by how substantially equivalent information is presented. A choice described in terms of gains, losses, success or failure can lead people to respond differently even when the underlying outcomes remain the same.

What it means

A description directs attention. Saying that 90 out of 100 attempts succeed emphasizes a different aspect from saying that 10 out of 100 fail. Both can describe the same data, yet readers may interpret them differently.

In decision research, framing can affect preferences between options, including how people respond to risk. The effect is not a promise that every person will react in the same direction or that wording can override all other considerations.

For AI applications, framing matters because assistants select wording, comparisons and reference points. An accurate statement can still emphasize one side of a choice while leaving another equally relevant description less visible.

How it works

A frame can establish a reference point and make particular consequences more salient. People may evaluate outcomes as gains or losses relative to that point rather than only as final states. Different descriptions can therefore alter the perceived attractiveness of the same options.

To examine a framing effect, keep the substantive choice constant while varying its presentation. Check that probabilities, outcome totals, time horizons and available alternatives are equivalent. A difference in behavior is not evidence of framing alone if the descriptions also change the actual information or consequences.

The observed effect depends on context, task and population. A careful study compares groups or conditions and accounts for ordinary variation rather than interpreting one person's changed answer as a universal psychological rule.

Three practical examples

AI summaries: An assistant describes a service as meeting its target in 95 percent of cases or missing it in 5 percent, influencing what readers notice.
Product comparisons: A decision aid presents the same tradeoff using retained capacity and lost capacity, then checks whether users understand both.
Research evaluation: A team tests whether equivalent prompt descriptions change an AI system's recommendations, without assuming that model behavior reveals human psychological mechanisms.

Think of it this way

Think of looking at the same landscape through windows facing different directions. Each window makes some features easier to see without changing the landscape itself. The analogy explains selective emphasis; real framing effects concern measurable judgments and preferences, not merely a change in visual viewpoint.

Worked example

Consider a fictional archive with 240 files that will become inaccessible unless a recovery action succeeds. There are two options.

Option A guarantees that 80 files remain accessible. Option B has a one-third chance that all 240 remain accessible and a two-thirds chance that none remain accessible.

The equivalent loss description says: Option A guarantees that 160 files become inaccessible. Option B has a one-third chance that no files become inaccessible and a two-thirds chance that all 240 become inaccessible.

The arithmetic is unchanged. Option A's 80 accessible files imply 240 - 80 = 160 inaccessible files. Option B's possible outcomes are the same in both descriptions. Both options have expected accessible count 80, since B's expectation is one-third × 240 + two-thirds × 0.

Equal expected counts do not make the options identical: one is certain and one is risky. A person may consistently prefer either based on their preferences. A framing effect would be a systematic preference change associated with the equivalent gain-versus-loss description, not merely a difference between a safe and risky choice.

This is an original teaching scenario. It includes no claimed experimental percentages and does not establish how any particular reader would choose.

Illustrative case study

Fictional case: Willow Analytics prepares AI-generated summaries of a service pilot. One version highlights the share of requests completed on time; another emphasizes delays. Reviewers appear to reach different impressions from the two drafts.

The team first checks that both versions use the same denominator, period and definition of on-time completion. It then revises the decision summary to present both complementary counts alongside the same underlying table.

For a formal evaluation, the team would compare equivalent presentations systematically and measure comprehension as well as preference. It does not treat a few reactions to draft wording as proof of a quantified framing effect. The case illustrates how editorial choices can influence interpretation even when the displayed numbers are accurate.

Common misconceptions and mistakes

Framing requires false information. Equivalent true descriptions can emphasize different aspects of a choice.
Any change in preference is a framing effect. New facts, changed options or different incentives can also explain it.
Gain language always makes people risk-averse and loss language always makes them risk-seeking. Such patterns are context-dependent, not universal laws for individuals.
An AI response shift proves the model thinks like a human. Similar outputs can arise through different mechanisms.

Business use

Framing is relevant to dashboards, product explanations and AI decision support. A system should help users understand the underlying tradeoff rather than depend on a favorable verbal angle.

Keep denominators and time periods visible, present complementary outcomes when useful, and separate expected values from the range of possible results. For evaluations, compare substantively equivalent prompts and document changes in recommendations. Measure whether users understand the decision, not only whether one wording increases selection of a preferred option.

Check your understanding

A report changes from “9 of 10 checks passed” to “1 of 10 checks failed,” with all other information unchanged. Did the numerical outcome change?

Answer: No. The descriptions are complementary frames of the same result. Whether judgments change is an empirical question.

Related terms

Decision theory
Cognitive Bias
Reference point
prospect theory
risk preference
anchoring
choice architecture
information presentation

Learn next

Rewrite the archive scenario in a table showing final accessible and inaccessible counts for each outcome. Ask whether both verbal descriptions map to the same table. Then identify which additional information would genuinely change the decision rather than merely reframe it.

Sources

Tversky and Kahneman's original research examines preference shifts under different descriptions of equivalent decision problems: https://stanford.edu/class/psych205/papers/Tversky-Kahneman-1981.pdf

The publisher's record identifies the original Science article: https://doi.org/10.1126/science.7455683

The archive arithmetic and Willow Analytics case are original illustrations. No empirical effect size is claimed for these scenarios or for present-day AI systems.

Previous: Frame problem Next: Frank - Wolfe algorithm