← CHI 2026 map

CHI '26 · Honorable mention · full-paper review · confidence medium-high

Prompt Coaching for Inclusiveness: A Media Literacy Approach to Increase Users’ Awareness of Algorithmic Bias and Prompting Efficacy

Cheng Chen , Mengqi Liao , Aditya Anand Phadnis , Yao Li , Andrew High , Saeed Abdullah , S. Shyam Sundar

DOI PDF Program page Page JSON

This is a strong CHI intervention paper: the novelty is not a new model, but a reframing of prompting guidance as media-literacy-oriented coaching with deliberate friction. The evidence supports immediate attitudinal and perceived-efficacy effects, while also showing a real usability cost that matters for deployment.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

This paper’s main contribution is best read as a normative interaction/framework contribution rather than a technical algorithmic advance. The authors explicitly position inclusive prompt coaching as a media literacy intervention and as a strategy that leverages design friction, which is a meaningful departure from the more common-sense view that prompting tools should simply make users faster or more efficient. The novelty claim is grounded in the paper’s own language about “proposing and testing” inclusive prompt coaching for communicating and combating algorithmic bias. The validation is reasonably aligned with that claim: the reported evidence comes from a one-time online user study with N=344, comparing coaching versus no coaching and measuring awareness of algorithmic bias, perceived prompting efficacy, trust, and trust calibration, alongside qualitative responses. That is enough to support claims about immediate perceptions and self-reported efficacy, but not enough to establish durable behavior change, real-world adoption, or actual reductions in biased model outputs. The paper is also candid about a key tradeoff: inclusive prompt coaching made the experience less satisfying, and the authors note that not defining “inclusive” and “bias” explicitly may lead to variability in the prompts the system suggests. That limitation matters because it constrains deployment and suggests the intervention may work differently across contexts or user groups. Overall, the paper looks like a solid honorable-mention-level CHI contribution: conceptually sharp, empirically supported for its stated scope, and important for the field’s ongoing discussion of ethical prompting and AI literacy, but not a broad technical solution to algorithmic bias.

What Changed

Canon before

Prior CHI work on prompting and bias reduction has typically emphasized prompt engineering, safety filters, or user guidance that aims to improve outputs directly; this paper instead frames inclusive prompting as a media-literacy intervention with intentional friction to prompt reflection.

Departure from common sense

The paper treats inclusive prompt coaching as a form of design friction and media literacy intervention that intentionally slows users down so they reflect on bias, rather than assuming users will naturally notice bias or that prompt help should only optimize output quality and speed.

Actual novelty

It proposes and empirically tests inclusive prompt coaching as a user-centered media literacy intervention for generative-AI prompting, combining bias-aware prompt rewriting with deliberate friction to raise awareness of algorithmic bias and improve perceived prompting efficacy and trust calibration.

Evidence

The paper combines a designed prompting intervention with a one-time online user study (N=344) and reports direct effects on awareness of algorithmic bias and perceived prompting efficacy, plus indirect effects on trust and trust calibration through cognitive elaboration. It also reports a usability cost in lower satisfaction and higher frustration, and it explicitly discusses definitional ambiguity around “inclusive” and “bias” as well as the limits of one-shot exposure.

Limits

Method limits

Validation is based on a single online study with self-reported outcomes and open-ended responses; the evidence supports immediate perceptions and reported efficacy, not long-term behavior change, durable adoption, or downstream improvements in actual model outputs.

Deployment limits

The intervention adds friction and can reduce satisfaction and increase frustration, so adoption may be harder in real prompting workflows. The authors also note that not defining “inclusive” and “bias” explicitly may produce variability in AI-suggested prompts and limit consistency across contexts.

Boundary conditions

Effects are demonstrated in an online experiment on image-generation prompting with a one-time interaction. The paper itself suggests outcomes may depend on users’ goals, prompt content, and whether the intervention is optional or contextually tailored.

Manual check: No manual check required for the core review summary; however, the exact evidence spans available in the packet are limited, so downstream QA may want to verify offset-to-quote alignment.

Position in field

This sits at the intersection of HCI, media literacy, and generative-AI prompting: a normative intervention paper that reframes prompt guidance as bias-awareness coaching rather than only output optimization, and that foregrounds the tradeoff between machine agency and user agency.

Abstract

Large language models often produce biased or stereotypical outputs. One way to reduce this possibility is to be more inclusive in our prompts, but doing so may not come naturally to most users. Therefore, we designed a tool that coaches users to write more inclusive prompts—a strategy that leverages design friction to provide a media literacy intervention. Data from a user study (N=344) show that compared to no coaching, inclusive prompt coaching directly increased users’ awareness of algorithmic bias and their perceived prompting efficacy. It also indirectly enhanced their trust in the system and perceived trust calibration through cognitive elaboration. However, inclusive prompt coaching resulted in a less satisfying user experience. These findings have implications for ethical interventions in prompting for better communicating and combating algorithmic bias. We discuss the benefits and limitations of inclusive prompt coaching, as well as ways to balance usability for long-term adoption of generative AI systems.