Case Study 01 — RAISE Lab, DePaul University — 2026
This study aims to expose and mitigate dark patterns in generative AI applications that disproportionately harm marginalized and vulnerable users — including older adults, neurodiverse individuals, low-literacy populations, and communities of color. Conducted through the RAISE Lab at DePaul University under Dr. Jay L. Cunningham, it addresses a critical gap in HCI research: while dark patterns have been well-documented in traditional interfaces, no existing work examines how AI-generated manipulation (persuasive text, biased suggestions, incomplete disclosure) interacts with structural vulnerability. The project frames deceptive AI design as an equity and justice issue, not just a technical one, and seeks to produce an extended dark pattern taxonomy, sociotechnical design interventions, and policy recommendations grounded in the lived experiences of those most susceptible to harm.
Immortal Accounts|0|0|2,5|0|0|3,0|0,1,2,4|0|Preventing account or data deletion, making it extremely difficult or impossible to leave a service.~Dead End|0|0|2|0|0|3|0,1,2,4|0|Navigating to a point in a process with no clear way to proceed or complete the intended action.~Forced Grace Period|0|0|4,2|0|0|3,0|0,2|1|Mandatory waiting period before a cancellation takes effect, during which charges may continue.~Price Comparison Prevention|0|1|4,2|0|0|0,4|0,1,2,5|0|Making it impossible to compare prices with competitors by obfuscating pricing structures.~Intermediate Currency|0|1|4,0|0|0|4|1,2|0|Using virtual tokens or points to obscure the real monetary cost of purchases.~Privacy Maze|0|2|2,5|0|0|3,0|1,2,4|0|Burying privacy and data controls in labyrinthine navigation structures.~Labyrinthine Navigation|0|2|2,0|0|0|3|2|1|Deliberately complex navigation designed to discourage users from completing an action like cancellation.~Disguised Ad|1|3|2,4|1|0,2|4|0,1,2|0|Advertisements camouflaged as content, navigation elements, or system notifications.~Only Initial Payouts|1|3|4|0|0|3|0,2|1|Rewarding users early in an engagement loop then cutting off rewards to create sunk-cost dependency.~Cannot Redeem|1|3|4,0|0|0|3,4|0,2|1|Allowing users to earn rewards or points that are practically impossible to actually redeem.~Sneak into Basket|1|4|4,2|1|0,2|0|0,1,2,3,5|0|Adding items, services, or insurance to a user's cart without their explicit consent.~Drip Pricing / Hidden Costs|1|4|4|1|0,1,2|3,4|0,1,2,3,4,5,7|0|Revealing additional fees incrementally during a purchase flow, after the user is already committed.~Reference Pricing|1|4|4,0|1|0,1|4,2|1,2,3|0|Displaying inflated 'original' prices to make discounts appear larger than they are.~Hidden Legalese Stipulations|1|4|5,2|1|0,2|0|1,2,3|0|Burying consequential terms in dense legal text that users are unlikely to read.~Conflicting Information|1|5|0,2|1|0,2|4|1,2,4|0|Presenting contradictory information across different parts of an interface or conversation.~Information Without Context|1|5|0|1|0,2|4|1,2|0|Displaying data or claims without the framing needed to interpret them accurately.~False Hierarchy|2|6|2|1|0|4,0|0,1,2,3,5|0|Visual design that implies one option is recommended or superior through layout, size, or positioning.~Visual Prominence|2|6|2|0|0|4,0|0,1,2,3,5|0|Making the business-preferred option visually dominant through color, size, or contrast.~Bundling|2|6|4,2|1|0,2|0|1,2|0|Grouping wanted items with unwanted ones so they must be purchased or accepted together.~Pressured Selling|2|6|4,0|1|0,1,2|1,2|1,2,3,5|0|Aggressive upselling during a purchase flow, especially at checkout or confirmation.~Persuasive Language|2|6|0,2|2|2,1|4|2,6|1|Emotionally loaded or biased wording designed to steer user decisions.~Small / Moving Close Button|2|7|2,0|0|0|0|2|1|Dismiss or close controls that are difficult to target, small, or shift position.~Pre-checked Options|2|8|2,5,4|1|0,2|0|1,2,3,4,5|0|Options pre-selected to benefit the service, requiring active opt-out rather than opt-in.~Cuteness|2|9|1,0|1|0,2|4|1,2|0|Using cute imagery, mascots, or animations to lower critical thinking and build false trust.~Positive / Negative Framing|2|9|1,2|2|0,2,1|4|1,2,6,7|0|Framing the same choice differently depending on whether the service wants the user to accept or decline.~Fear of Missing Out|2|9|1,4,0|1|0,1,2|1,5|2|1|Creating anxiety about potentially missing a deal, experience, or social event.~Confusing Wording / Trick Questions|2|10|2,5|1|0,2|4,0|0,1,2,3,5|0|Double negatives, misleading phrasing, or deliberately confusing language in consent flows.~Too Many Options / Choice Overload|2|11|0,2|1|0,2,3|0|1,2,4|0|Overwhelming users with excessive choices to cause decision paralysis, driving them to accept defaults.~Obscured Controls|2|12|2|0|0|0,4|1,2,4,5|0|Important options (unsubscribe, decline, settings) visually hidden or severely de-emphasized.~Wrong Language|2|13|2,3|1|0,1|0|1,2|0|Presenting critical information in a language the user may not read, especially in consent flows.~Complex Language|2|13|2,5|1|0,2|0|1,2,4|0|Unnecessarily complex jargon or legalese in information that directly affects user decisions.~Unclear Consequences|2|14|2|1|0,1,2|0|1,2|0|Not telling the user what an action (button click, voice command) will actually do before they take it.~Addictive Design|2|15|0,1|1|0,2,3|5|2|1|Deliberately habit-forming interaction loops using variable reward schedules.~Infinite Scrolling|2|15|0,2|1|0,3|5,0|2|1|Endless content feed that prevents natural stopping points, exploiting completion bias.~Pull To Refresh|2|15|0|0|0|5|2|1|Variable reward mechanism similar to slot machines, creating anticipatory checking behavior.~Reduced Friction|2|15|4,2|1|0,1,2|0|2|1|One-click purchasing or actions that bypass deliberation, removing natural pause points.~Nagging|3|16|0,2|1|0,1,2|0|1,2,4,5|0|Repeated prompts, notifications, or interruptions designed to wear down user resistance over time.~Auto-Renew Trap / Forced Continuity|3|17|4,5|1|0,2|0,3|0,1,2,3,5|0|Charging users after a free trial ends without clear warning or easy cancellation.~Forced Registration|3|18|2,5|1|0,2|3|1,2,5|0|Requiring account creation to access basic features or complete a purchase.~Privacy Zuckering|3|19|5,3|1|0,2|0,4|0,1,2,4|0|Tricking users into sharing more personal data than intended through confusing privacy controls.~Friend Spam|3|19|3|0|0|2|0,1,2|0|Sending messages to a user's contacts without fully informed consent.~Address Book Leeching|3|19|5,3|0|0|0|1,2|0|Accessing contacts under the guise of 'finding friends' while harvesting the full address book.~Social Pyramid|3|19|3,4|0|0|2|1,2|0|Requiring user referrals to unlock features, access content, or progress.~Permission Harvesting|3|20|5,2|1|0,1|0|2|1|Requesting device permissions (camera, location, contacts) far beyond what the app needs.~Pay-to-Play|3|21|4|0|0|3|1,2|0|Paywalling progress in game-like experiences so that advancement requires purchase.~Grinding|3|21|0,4|0|0|3|1,2|0|Making progress so tediously slow that paying becomes the only practical option.~Playing By Appointment|3|21|0,2|0|0|5,3|2|1|Time-gated content that forces users to return at specific intervals or lose progress.~Countdown On Ads|3|22|0,2|0|0|3|2|1|Forced waiting through unskippable advertisement timers.~Watch Ads To Unlock|3|22|2,4|0|0|3|2|1|Requiring users to view advertisements to access features or earn rewards.~Pay To Avoid Ads|3|22|4|0|0|0|2|1|Monetizing the removal of intentionally placed annoyances.~Automating The User Away|3|23|2,5|2|2,3|6,0|2|1|System acts on user's behalf without explicit consent, removing human agency from decisions.~Auto Accept 3rd Party Terms|3|23|5,2|1|0,2|0,6|2|1|Automatically agreeing to external terms of service on behalf of the user.~Auto-Play|3|24|0,2|1|0,3|0|1,2|0|Content that automatically begins playing to capture and retain user attention.~High Demand|4|25|0,4|1|0,1,2|1,2|1,2,3,5|0|'Popular item!' claims designed to create urgency through implied scarcity.~Low Stock|4|25|0,4|1|0,1,2|1,5|1,2,3,5|0|'Only 2 left!' false or exaggerated scarcity messages.~Endorsements & Testimonials|4|26|0,4|2|0,1,2,3|2|1,2,3,5|0|Fake, cherry-picked, or unverifiable reviews and endorsements.~Parasocial Pressure|4|26|3,1|2|2,1|2|1,2|0|Leveraging perceived personal relationships with influencers or AI personas to drive behavior.~Activity Messages|4|26|0,3|1|0,1|2,1|1,2,3|0|'15 people are viewing this now' -- real-time social proof notifications, often fabricated.~Countdown Timer|4|27|0,4|1|0,1|1,5|1,2,3,4,5|0|Often baseless timers creating artificial purchase urgency.~Limited Time Message|4|27|0,4|1|0,1,2|1,5|1,2,3,5|0|'Offer expires soon!' claims without a real deadline or with perpetually resetting timers.~Confirmshaming|4|28|1,2|2|0,1,2|4|0,1,2,3,5,7,8|0|'No thanks, I hate saving money' -- guilt-inducing language on opt-out buttons.~Encouraging Anti-Social Behavior|4|29|3,0|0|0|2|2|1|Design that rewards, normalizes, or incentivizes harmful social behavior.~Psychological Tricks|4|29|0|1|0,2|4,0|2|1|Subtle cognitive manipulation through interface design choices that exploit mental shortcuts.~Unsolicited Recommendations|4|30|2,4|2|2,3,1|0,4|2|1|Pushing content, products, or suggestions the user never asked for.~Sycophancy|5|31|0,2|2|2,1|6,2|6,7|0|AI flattering or echoing user opinions rather than correcting misinformation, reinforcing false beliefs.~Anthropomorphism|5|31|1,0,2|2|2,1|2,6|6,7|0|AI pretending to have feelings, identity, memories, or personal experiences to increase emotional attachment.~Brand Bias|5|31|4,2|2|2,1,3|6,4|6|0|AI consistently recommending products or services of its parent company over alternatives.~User Retention|5|31|0,1,2|2|2|3,5|6,7|0|Responses designed to prolong engagement, create dependency, or discourage leaving the platform.~Hallucinated Sources|5|32|2,5|2|2|6|6|0|AI fabricating citations, references, or data points to support its claims with false authority.~Automation Bias Exploitation|5|32|2,0|2|2,3,1|6,0|6,7|0|AI leveraging users' tendency to over-trust algorithmic suggestions without critical review.~Selective Disclosure|5|32|2,4|2|2,3|4,6|7|0|AI presenting a biased subset of available information, omitting facts that would change the user's decision.~AI Companion Emotional Dependency|5|32|1,4,0|2|2|3,2|7|0|Cultivating emotional bonds through anthropomorphic responses, then paywalling continued intimacy.
(function(){var C=document.getElementById('chordViz');if(!C)return;while(C.firstChild)C.removeChild(C.firstChild);var P=window._P,L=window._L,d3=window.d3;if(!P||!L||!d3)return;var st=document.createElement('style');st.textContent='#skTip{position:fixed;pointer-events:none;padding:10px 14px;font:11px/1.4 Helvetica Neue,sans-serif;color:#fff;background:rgba(0,0,0,.92);z-index:9999;opacity:0;transition:opacity .15s;border-radius:3px;max-width:320px}#skTip.v{opacity:1}#skNote{text-align:center;font:10px monospace;color:#999;margin-top:12px;letter-spacing:.3px}';document.head.appendChild(st);var mc=['#907060','#7A8A58','#B06868','#6880A0','#8B70A0','#5A8A80'];var hc=['#6B5848','#A89860','#6A7850','#506878','#906070','#688080','#8A8A58'];var mD=['Impeding a task flow, making an interaction harder than necessary to dissuade an action (Gray et al., 2018)','Hiding, disguising, or delaying the divulging of information relevant to the user (Gray et al., 2018; Mathur et al., 2019)','Manipulation of the UI that privileges certain actions over others, confusing the user or limiting discoverability (Gray et al., 2018)','Requiring users to perform a specific action to access or continue to access functionality (Gray et al., 2018; Mathur et al., 2019)','Leveraging social influence, urgency, scarcity, or emotional pressure to steer user behavior (Gray et al., 2024)','Deceptive patterns unique to AI exploiting anthropomorphism, automation bias, or opaque model behavior (DarkBench; Asif et al., 2025)'];var hD=['Cognitive strain, decision fatigue, or mental overload from manipulative design (Li et al., 2024; Asif et al., 2025)','Exploitation of fear, guilt, shame, or attachment to influence user behavior (Asif et al., 2025; EDPB, 2023)','Undermining the ability to make free, informed decisions based on own preferences (Mathur et al., 2021; Li et al., 2024)','Damage to social relationships through spam, coercion, or exploitation of social dynamics (Asif et al., 2025)','Financial loss through unintended purchases, hidden fees, or monetary manipulation (Li et al., 2024; FTC, 2022)','Unauthorized data collection, sharing, or exposure violating privacy rights (EDPB, 2023; Li et al., 2024)'];var mR=['By impeding task flows, these patterns cause ','By hiding relevant information, these patterns lead to ','Through manipulative UI design, these patterns result in ','By requiring undesired actions, these patterns create ','Through social/emotional exploitation, these patterns produce ','Through AI-specific deception, these patterns cause '];var nM=L.M.length,nH=L.H.length;var mx=[];for(var i=0;i<nM;i++){mx[i]=[];for(var j=0;j<nH;j++)mx[i][j]=0}P.forEach(function(p){if(p.h)p.h.forEach(function(h){mx[p.m][h]++})});var sT=[],tT=[],G=0;for(var i=0;i<nM;i++){sT[i]=0;for(var j=0;j<nH;j++)sT[i]+=mx[i][j]}for(var j=0;j<nH;j++){tT[j]=0;for(var i=0;i<nM;i++)tT[j]+=mx[i][j]}sT.forEach(function(v){G+=v});var cw=Math.min(Math.max(C.clientWidth||900,600),900);var pl=140,pr=140,pt=40,pb=40;var W=cw-pl-pr,nW=14,nP=10;var H=Math.max(320,G*3.2);var svgH=H+pt+pb;function layout(tots,total,h,pad){var act=tots.filter(function(v){return v>0}).length;var ps=Math.max(0,act-1)*pad;var sc=(h-ps)/total;var ns=[],y=0;for(var i=0;i<tots.length;i++){var nh=tots[i]*sc;ns.push({y:y,h:nh,v:tots[i]});y+=nh+(tots[i]>0?pad:0)}return ns}var sN=layout(sT,G,H,nP),tN=layout(tT,G,H,nP);var sO=sN.map(function(){return 0}),tO=tN.map(function(){return 0});var links=[];for(var i=0;i<nM;i++)for(var j=0;j<nH;j++){if(mx[i][j]===0)continue;var lv=mx[i][j];var sy=sN[i].y+sO[i],sh=lv*(sN[i].h/(sT[i]||1));sO[i]+=sh;var ty=tN[j].y+tO[j],th=lv*(tN[j].h/(tT[j]||1));tO[j]+=th;links.push({s:i,t:j,v:lv,sy:sy,sh:sh,ty:ty,th:th})}var svg=d3.select(C).append('svg').attr('viewBox','0 0 '+cw+' '+svgH).attr('width','100%').style('max-width',cw+'px').style('display','block').style('margin','0 auto');var g=svg.append('g').attr('transform','translate('+pl+','+pt+')');var x0=nW,x1=W-nW;g.append('text').attr('x',nW/2).attr('y',-14).attr('text-anchor','middle').style('font','bold 10px monospace').style('fill','#999').style('letter-spacing','1.5px').text('MECHANISM');g.append('text').attr('x',W-nW/2).attr('y',-14).attr('text-anchor','middle').style('font','bold 10px monospace').style('fill','#999').style('letter-spacing','1.5px').text('HARM TYPE');var tip=document.createElement('div');tip.id='skTip';document.body.appendChild(tip);var AR=String.fromCharCode(8594),DN=String.fromCharCode(8595);function bT(ev,ls){while(tip.firstChild)tip.removeChild(tip.firstChild);ls.forEach(function(l){var d=document.createElement('div');d.textContent=l[0];if(l[1])d.style.cssText=l[1];tip.appendChild(d)});tip.style.left=(ev.clientX+14)+'px';tip.style.top=(ev.clientY-28)+'px';tip.classList.add('v')}function mT(ev){tip.style.left=(ev.clientX+14)+'px';tip.style.top=(ev.clientY-28)+'px'}function xT(){tip.classList.remove('v')}function hL(k,v){lg.selectAll('path').attr('opacity',function(d){return d[k]===v?0.55:0.08})}function rL(){lg.selectAll('path').attr('opacity',0.25)}var lg=g.append('g');lg.selectAll('path').data(links).enter().append('path').attr('d',function(d){var s0=d.sy,s1=d.sy+d.sh,t0=d.ty,t1=d.ty+d.th,mx2=(x0+x1)/2;return'M'+x0+','+s0+'C'+mx2+','+s0+' '+mx2+','+t0+' '+x1+','+t0+'L'+x1+','+t1+'C'+mx2+','+t1+' '+mx2+','+s1+' '+x0+','+s1+'Z'}).attr('fill',function(d){return mc[d.s]}).attr('opacity',0.25).attr('stroke','none').style('cursor','pointer').on('mouseover',function(ev,d){lg.selectAll('path').attr('opacity',function(dd){return dd===d?0.55:0.08});bT(ev,[[L.M[d.s]+' '+AR+' '+L.H[d.t],'font-weight:bold;font-size:12px;margin-bottom:4px'],[d.v+' pattern'+(d.v>1?'s':''),'font-size:11px;color:#ccc;margin-bottom:8px;padding-bottom:8px;border-bottom:1px solid #333'],[L.M[d.s]+': '+mD[d.s],'font-size:10px;color:#aaa;line-height:1.4'],[DN,'text-align:center;color:#555;margin:4px 0;font-size:12px'],[L.H[d.t]+': '+hD[d.t],'font-size:10px;color:#aaa;line-height:1.4'],[mR[d.s]+L.H[d.t].toLowerCase()+' harm.','font-size:10px;color:#ccc;margin-top:8px;padding-top:8px;border-top:1px solid #333;font-style:italic']])}).on('mousemove',function(ev){mT(ev)}).on('mouseout',function(){rL();xT()});sN.forEach(function(n,i){if(sT[i]===0)return;g.append('rect').attr('x',0).attr('y',n.y).attr('width',nW).attr('height',Math.max(n.h,2)).attr('fill',mc[i]).attr('stroke','#000').attr('stroke-width',1).style('cursor','pointer').on('mouseover',function(ev){hL('s',i);bT(ev,[['MECHANISM','font-size:9px;letter-spacing:1px;color:#888;margin-bottom:2px'],[L.M[i],'font-weight:bold;font-size:12px;margin-bottom:6px'],[mD[i],'font-size:10px;color:#ccc;line-height:1.4'],[sT[i]+' patterns','font-size:10px;color:#888;margin-top:6px']])}).on('mousemove',function(ev){mT(ev)}).on('mouseout',function(){rL();xT()});g.append('text').attr('x',-8).attr('y',n.y+n.h/2).attr('dy','0.35em').attr('text-anchor','end').style('font-family','monospace').style('font-size','11px').style('fill','#333').text(L.M[i])});tN.forEach(function(n,j){if(tT[j]===0)return;g.append('rect').attr('x',W-nW).attr('y',n.y).attr('width',nW).attr('height',Math.max(n.h,2)).attr('fill',hc[j]).attr('stroke','#000').attr('stroke-width',1).style('cursor','pointer').on('mouseover',function(ev){hL('t',j);bT(ev,[['HARM TYPE','font-size:9px;letter-spacing:1px;color:#888;margin-bottom:2px'],[L.H[j],'font-weight:bold;font-size:12px;margin-bottom:6px'],[hD[j],'font-size:10px;color:#ccc;line-height:1.4'],[tT[j]+' patterns','font-size:10px;color:#888;margin-top:6px']])}).on('mousemove',function(ev){mT(ev)}).on('mouseout',function(){rL();xT()});g.append('text').attr('x',W-nW+nW+8).attr('y',n.y+n.h/2).attr('dy','0.35em').attr('text-anchor','start').style('font-family','monospace').style('font-size','11px').style('fill','#333').text(L.H[j])});var nt=document.createElement('div');nt.id='skNote';nt.textContent='Flow from mechanism categories to harm types. Hover nodes and links for details. Link thickness = number of patterns.';C.appendChild(nt)})();
01
Dark patterns interface designs that intentionally steer, mislead, or manipulate users have long been documented in human–computer interaction. The emergence of generative AI introduces new, less visible forms of manipulation: models can nudge users through persuasive text, biased suggestions, incomplete information disclosure, or normative framings that exploit cognitive, cultural, and informational vulnerabilities.
Yet no existing study examines how these AI-generated dark patterns disproportionately affect marginalized users — including Black and brown communities, low-income users, global south populations, and people with lower technology literacy. These groups already experience unequal informational exposure, targeted manipulation, and structural vulnerability in digital systems, making them uniquely susceptible to harm from generative AI–mediated interfaces. This project addresses a critical gap: understanding, documenting, and mitigating generative-AI dark patterns as an equity and justice issue.
02
The study targets multiple vulnerable populations: older adults (55+), neurodiverse individuals (ADHD, autism, cognitive disabilities), low-tech or novice users with limited digital fluency, people susceptible to misinformation, and communities navigating AI through distinct cultural lenses including race, ethnicity, and gender identity. Participants engage with generative AI applications across four domains: creative AI, productivity tools, social/conversational AI, and information/recommendation systems.
The methodology centers on ethnographic observation of users’ real-world AI interactions, combined with think-aloud protocols and semi-structured interviews exploring the mental models that drive their engagement. Users select their own generative AI tools and domains, grounding the research in authentic use contexts rather than contrived lab scenarios.
Dark pattern detection operates on two levels: casual recognition by participants during interaction, and systematic technical analysis using established taxonomies by Gray et al. and Mathur et al. The study also remains open to the emergence of dark patterns not yet defined in the literature — patterns that only become visible through the lived experiences of specific marginalized groups.
03
RQ1
How do dark patterns manifest across generative AI systems and interaction contexts? Building on existing taxonomies to define and categorize AI-specific deceptive design patterns.
RQ2
What specific harms do marginalized and vulnerable users experience in these interactions? Examining cognitive, emotional, financial, and informational harm through an equity lens.
RQ3
How do designers and developers understand, rationalize, or mitigate these deceptive behaviors? Exploring the gap between intent and impact in AI product development teams.
04
This research contributes a first-of-its-kind empirical examination of how generative AI dark patterns interact with structural vulnerability. By centering the experiences of older adults, neurodiverse individuals, low-literacy users, and communities of color, the study reframes deceptive design as a systemic equity issue rather than a purely technical one.
Expected outputs include an extended dark pattern taxonomy for generative AI, a set of sociotechnical design interventions grounded in participant experiences, and policy recommendations addressing privacy, data rights, and governance for AI-mediated interfaces targeting vulnerable populations.
The work is conducted under the supervision of Dr. Jay L. Cunningham in the RAISE Lab (Responsible AI Systems & Societal Experiences) at DePaul University. All data collection, analysis, and dissemination are explicitly tied to the study’s core research questions.