Beyond the Bargain: How Brands can Retain Burned Out Consumers

By: Lena Kleinwechter, Customer Engagement & Loyalty Strategist at Talon.One

The biggest shopping season of the year is among us, and with it comes the bombardment of promotions through email, social media, text and in-person mail. Black Friday and holiday shopping were once defined by consumer excitement with limited-time discounts that encouraged shoppers to form lines hours before stores opened.

While discounts continue to grow and start earlier each year, a quieter shift is happening: consumers are tuning out. Customers are experiencing “discount burnout” as retailers continue to rely on overused discounting to engage and retain customers. A report by AlixPartners found that the importance consumers place on price has dropped by 13% compared to 2024. Customers are becoming more selective about which brands they interact with, focusing far more on brands that provide value year-round and relying less on large markdowns. This shift is a clear sign that consumers are searching for authentic value and personalized experiences that go beyond holiday sales.

The Price of Year-Round Markdowns

Overdiscounting hurts brands in more ways than one. The overuse of discounts has made discount-driven campaigns significantly less effective. By engaging customers only through promotions, brands are diluting their offerings and conditioning customers to only purchase when discounts or promotions are available. This “next sale” mentality keeps companies trapped in a reactive cycle where customer engagement depends on lowering prices instead of increasing value.

This approach weakens brand value for customers because they begin associating a company solely with markdowns. Eventually, customers question if the original price reflects the true value of the product, making it harder for them to compete on different aspects like quality, durability or experience. Over time, this erodes customer trust and creates the assumption that a better discount is always around the corner. This shifts the customer mindset from excitement to skepticism, undercutting true brand loyalty.

The financial consequences can be daunting for companies. Heavy discounting can result in a “discount death spiral” that cuts directly into profit margins and leaves little to no room for investment into innovation and growth. Companies stuck in discount spirals are at the mercy of customer behavior. In challenging conditions, when spending slows, an overreliance on offers can leave companies especially vulnerable.

Even consumers are tired of the “next sale” strategy. According to a report by Alixpartners, service and experience have become two of the most important drivers of customer spending . Customers are no longer looking at price tags for purchasing decisions. Now they’re focused on finding a brand that can anticipate demands and provide a variety of benefits. This may not have always been the case and discounts only blurred and distracted the customer’s focus. By moving away from the “next sale” thought process, brands can think about the customer experience as a whole, aiming to become the solution to customer’s problems and frustrations through efforts such as seamless returns or high quality service.

Curing Discount Fatigue with a Personal Touch

With discounts and promotions out, what’s the next engagement strategy for retailers? Personalization.

Early on, personalization focused on inserting a name in an email or sending a one-size-fits-all discount code. However, as technology evolves, so does the opportunity to create shopping experiences that reflect unique preferences and behaviors. Customers now expect brands to use their data responsibly and proactively to deliver value that feels relevant and timely. In fact, companies that moved away from mass discounting toward personalized offerings create more value for their customers. A Harvard Business Review report with Talon.One found that of the organizations that started personalizing offerings, 62% say they’ve seen increased sales and 47% say it has increased customer loyalty as a result.

This demonstrates a brand recognizes who their customers are and what they care about. For example, a beauty brand might send simple replenishment reminders based on purchase frequency or a retailer could tailor promotions to local weather conditions. These moments reinforce that the brand understands customer needs on a deeper level, building reliability and trust. Companies that invest in strategies and tools for this have the key to unlock long-term engagement.

Retention Over Reaction: Playing the Long Game

Once you’ve delivered a customer their best moment, maintaining that relationship becomes far more valuable than chasing a quick conversion. Brands that are thriving in this environment have also shifted their focus from conversion to retention, which is a helpful long-term strategy for retailers.

To achieve customer retention, brands should think about customers as relational rather than transactional, and rely on gamifying the engagement to battle discount burnout. Companies should measure success through lifetime value and advocacy. Instead of giving customers instant but forgetful gratification, loyalty programs are built to encourage customers to continue doing an action in exchange for points, personalized offers or exclusive product drops. A number of fast-casual restaurants have revamped their loyalty programs to increase personalization offerings and deliver additional value to their customers. This personalization results in customers feeling recognized and rewarded for consistent engagement and encourages brands to think of long-term retention above short-term promotions. However, businesses should still be cautious of making strong personalization claims. These claims can make customers have unrealistic expectations and believe that offers will 100% match their wishes. When the first couple of offers don’t deliver an identical match, customers may be disappointed and see personalization attempts through their negative experience.

Redefining What Value Means

This blueprint should guide brands to avoid aggressive discounting as they head into the holiday season.

Brands should redefine what “value” means in the modern retail environment with new strategies and tactics. The goal is to understand customers enough to contextualize benefits and build programs that encourage patterns of behavior. By investing in personalization and long-term engagement, brands can build trusting relationships that keep customers coming back even without discounts. The brands that lean into transparent and authentic outreach will have a competitive advantage over others and build genuine relationships that retain customers after the sale banners come down.

Read next: As AI Reshapes Talent Demand, a New Study Shows These Four Professions Are Coming Out on Top

by Web Desk via Digital Information World

A Small Pilot Signals YouTube’s Renewed Interest in Built-In Messaging

YouTube has started a controlled test that brings direct messages back to the mobile app. The feature gives people a way to share videos with each other and talk about them in the same place. It supports long clips, Shorts, and live streams.

The test is limited to adults who are eighteen or older. Only users in Ireland and Poland are included in this first round. People in the experiment need to send an invite before they can start a chat. Anyone who receives an invite can turn it down. They can also block another user or report a conversation.

YouTube ran an earlier messaging system for a few years and removed it in 2019. This new trial follows a different setup but keeps the idea of simple sharing and conversations inside the app. The support page explains how messages can be reviewed to enforce Community Guidelines. The system may scan for content that violates policies or could cause real-world harm. These rules match the standards applied to videos and comments.

People can keep sharing links through other apps if they prefer. The test only adds an option inside YouTube. The company has not provided any timeline for a wider release.

Notes: This post was edited/created using GenAI tools with human oversight. Image: DIW-Aigen.

Read next:

• AI Conversations Weaken Conspiracy Beliefs, Regardless of Messenger

• As AI Reshapes Talent Demand, a New Study Shows These Four Professions Are Coming Out on Top
by Irfan Ahmad via Digital Information World

As AI Reshapes Talent Demand, a New Study Shows These Four Professions Are Coming Out on Top

The widespread adoption of AI is having far-reaching impacts on the global job market.

In addition to driving elevated layoffs in the tech sector and beyond, the expanding influence of AI is reshuffling which skills are most in demand, and contributing to friction between talent supply and that demand, according to a new study from Toptal .

The High-skilled Job Report for October 2025 outlines an array of both positive and negative month-over-month and year-over-year trends, some of which appear dramatic and even contradictory at first glance. But upon careful review, it’s clear that the spikes and troughs are due to the same thing: wide-reaching instability as organizations adjust to quickly evolving technology and volatile macroeconomic conditions.

In an effort to meet this moment, companies are rapidly adjusting business models, strategies, talent portfolios, and hiring to support their changing priorities. In the job market, this is playing out as dramatic increases in demand for certain skill sets, while other types of expertise are temporarily deprioritized.

What’s Different About This Job Data

It has been extremely difficult for economists, labor experts, and job seekers to understand the current white-collar job market. Job listings are plentiful, but many openings are overwhelmed with hundreds or even thousands of applicants, and many laid-off professionals—and recent graduates—are struggling to find work. One reason for this confusion is the continued persistence of ghost job postings that companies advertise but have no intention of filling immediately (or ever).

The Toptal Market Strength Scores in the report solve for this by equally weighting stated demand, as measured by unique open job listings and offered compensation, against actual hiring. The scores also differ from general labor data because they exclude entry-level roles, focusing on demand for technology and professional services professionals with at least five years of experience. Because Toptal is the world’s largest remote workforce, they filter their demand data for explicitly remote or hybrid roles.

Which Workers Are Most in Demand Now?

Demand for Data Science Experts Grew 23% YoY

Given the fundamental importance of data in artificial intelligence, it's perhaps unsurprising that data science expertise is experiencing an increase in demand. While demand for these experts dipped 7% month-over-month, according to Toptal’s October 2025 report, year-over-year demand grew by double-digits (23%), for a Toptal Market Strength Score of Strong.

A negative change in demand of more than -15% equals a Toptal Market Strength Score of Poor, a change between -15% and +15% equals a score of Moderate, and a positive change greater than +15% equals a score of Strong. Image Courtesy of Toptal

This strong demand continues a trend of increasing demand for data science experts that Toptal has noted for several quarters in a row. For example, year-over-year demand was up by 40% as of the end of Q3 2025, by 28% as of Q2, and by 32% as of Q1. Toptal data science leaders also say that a premium is being placed on data professionals who can translate their technical skills into business insights. There is an ongoing strategy gap in AI, with organizations struggling to determine the most valuable AI use cases and ROI measurement throughout production. Given that, demand favors experts “who can connect the dots between technical fluency, commercial savvy, and product thinking,” says Brad DeFrank, Director of Delivery and AI Strategy at Toptal. “Analytics in this landscape isn’t about dashboards, it’s about decisions.”

Demand for Developers Increased 24% YoY

Toptal data suggests that some experienced software developers are also benefiting from the rapid adoption of AI. Demand grew 5% month-over-month in October 2025 for a Toptal Market Strength Score of Moderate, while year-over-year growth was Strong, with a surge of 24%.

Vrinda Dabke, Toptal’s VP of Global Technology Services, said she is seeing a sustained—but selective—demand specifically for senior-level development specialists, driven by organizations’ need for immediate, high-impact ROI. “The market continues to favor strategy-based hiring, prioritizing candidates who can demonstrate immediate, transformative ROI through skills like generative AI integration, MLOps, and advanced cloud security,” she notes in the report.

Demand for Product Managers Increased 21% YoY

Demand for experienced product managers increased by 13% month over month and by 21% year over year, according to Toptal’s October 2025 report. These increases were also likely due in part to rapid AI adoption. Érico Sabino, Enterprise Matching Team Lead at Toptal, is seeing the highest demand for product managers with expertise in LLMs and artificial intelligence, particularly roles focused on AI integration, such as AI product managers for internal workflow automation and AI-driven personal assistant applications. Furthermore, a “focus on minimum viable products and product roadmaps for AI solutions underscores a strategic trend toward rapid innovation and market responsiveness,” he says.

Demand for Marketing Experts Increased 18% YoY

The need for marketers may initially seem unrelated to AI adoption. But product marketing expertise is extremely important for differentiated messaging and positioning—particularly in the age of AI, when companies must be able to clearly explain their AI integrations and offerings, says Chris Krohn, GM of Toptal’s Marketing Agency. Demand for experienced marketing experts increased by 4% month over month, and by 18% year over year, for a Toptal Market Strength Score of Strong.

	Month-Over-Month Toptal Market Strength Score*	Year-Over-Year Toptal Market Strength Score*
Data Science Experts	-7%, Moderate	+23%, Strong
Developers	+5%, Moderate	+24%, Strong
Product Managers	+13%, Moderate	+21%, Strong
Marketing Experts	+4%, Moderate	+18%, Strong

 * A negative change in demand of more than -15% equals a score of Poor, a change between -15% and +15% equals a score of Moderate, and a positive change greater than +15% equals a score of Strong.

Which Professional Sectors Are Suffering Most in the Face of AI?

Toptal includes 10 technology and professional services areas of expertise in its report: data science, design, development, finance, information security, management consulting, marketing, product management, project management, and sales. The October report notes that year-over-year demand for experienced designers, information security experts, and management consultants all fell by double digits.

As noted earlier, rapid adoption of AI is temporarily shifting talent priorities at many organizations while they focus on filling the most urgent data- and AI-related roles. But there is a possibility that automation of certain tasks may shift the market for some areas of expertise in a more lasting way.

For example, access to AI-driven analytics and strategy tools may decrease

1981organizations’ needs for certain types of management consultants. Indeed, Toptal’s analysis noted a moderate 13% decrease in overall demand for these experts year over year.

But at the same time, Michael Valocchi, Toptal’s Senior Client Solutions Advisor, notes that demand specific to business consultants who can help create use cases for AI was actually up. “We also saw an increased demand for people and change management experts as new challenges arose from AI and transformation efforts. We anticipate even greater demand for leadership development as companies continue to reinvent themselves,” he notes.

Similarly, demand is down for more basic information security expertise that can be replaced or augmented by automation. Meanwhile, “we saw an urgent, emerging need for professionals who could help define and implement an AI security posture management layer to mitigate new risks, such as model poisoning and LLM jailbreaking, positioning AI risk assessment and AI/ML security architecture as critical governance, risk, and compliance (GRC) components,” notes Zohra Ibrahimi, Practice Director, Cyber and Information Security Services, at Toptal.

	Month-Over-Month Toptal Market Strength Score	Year-Over-Year Toptal Market Strength Score
Designers	-14%, Moderate	-24%, Poor
Information Security Experts	-7%, Moderate	-23%, Poor
Management Consultants	-6%, Moderate	-13%, Moderate

AI Will Continue to Reshuffle Demand and Complicate the Job Market

The Toptal data focuses on roles that require at least five years of experience, so it can’t speak to whether entry-level roles in data science, development, product management, or marketing are experiencing the same lift as more senior positions. One thing the data did highlight, however, was the contrast between how highly experienced professionals are faring in the face of AI versus those who are just starting out. Trends in the broader job market that include all levels of experience were generally more negative, according to the Toptal report, which cited a 6.7% year-over-year drop in new job postings on Hacker News, as well as a drop in new job postings in the US, UK, Germany, France, and most other major economies. (Canada was the one exception.)

The job market is likely to continue to remain volatile and uneven until new business models and talent strategies stabilize. Workers of all experience levels will need to stay flexible and tuned into trends that suggest which skills are rising in value and which are being automated away.

By Erik Stettler, Chief Economist, Toptal

Erik is the chief economist at Toptal, the world’s largest fully remote workforce of more than 20,000 highly vetted professionals in technology, design, finance, marketing, and strategic consulting. A data scientist and management consultant, Erik is a former senior analyst at economic consulting firm NERA and co-founder of the global venture capital fund Firstrock Capital. Erik holds an MBA with distinction from Harvard.

by Web Desk via Digital Information World

A Global Account Mapping Event Reveals What WhatsApp Metadata Can Expose

Researchers at the University of Vienna uncovered a weakness in WhatsApp’s contact discovery process that let them confirm more than 3.5 billion active accounts across 245 countries. The team relied on the same basic mechanism that helps users find contacts through phone numbers. WhatsApp checks each number against its registry. The researchers found that the system allowed an unusually high volume of lookups from a single source, which opened the door to automated enumeration at a massive scale.

Their testing reached a pace of more than one hundred million number checks per hour. The data made available through these lookups matched what any person could access when already aware of a phone number. That limited set included numbers, public keys, timestamps and public profile details. Even so, the researchers linked these pieces to patterns that revealed operating systems, account ages and companion device counts. They also spotted rare cases where cryptographic keys appeared to be reused across devices or numbers. Those findings pointed toward unofficial clients or improper implementations.

The dataset captured a broader snapshot of global behavior. Millions of active accounts appeared in regions where WhatsApp is officially blocked, including China, Iran and Myanmar. Platform distribution leaned heavily toward Android with a global share near eighty percent. The remaining group used iOS. Privacy habits varied by country. Some regions showed heavier use of public profile photos or public status text, while others leaned toward a more locked down setup.

The study highlighted long term risks tied to older exposures. Nearly half of the numbers seen in the major Facebook scraping incident from 2018 remained active on WhatsApp in 2021. That persistence raised concerns about continued targeting through scams and other unwanted contact.

No message content was ever accessed, and the researchers deleted the collected data before publishing their work. End to end encryption protects chats, but the team stressed that metadata can still reveal patterns that matter. They noted that even limited signals can be combined to build a picture of a user’s activity window or device environment.

Meta received the disclosure and added stronger rate limits along with tighter controls around profile visibility. The company said it had already been developing stronger anti scraping systems and used this study to validate those defenses. Meta also said it found no signs that malicious actors used the technique at similar scale.

This event landed during a year in which Meta paid more than four million dollars to security researchers for valid bug reports across WhatsApp, Facebook, Instagram and its other platforms. The company processed about thirteen thousand submissions and accepted around eight hundred. Meta highlighted two issues in particular. One stemmed from the Vienna enumeration work. The other came from an internal analyst using a specialized proxy tool to examine WhatsApp’s network protocol. That review uncovered an incomplete validation problem in older client versions that could have triggered content retrieval from arbitrary URLs on a recipient’s device. Meta patched it before any harmful use surfaced.

The company also released a patch to address a separate high severity vulnerability, tracked as CVE 2025 59489, that affected Quest devices through Unity based applications. That flaw came from a different researcher and involved operating system level behavior rather than messaging.

Meta has started distributing the WhatsApp Research Proxy to select long term contributors who focus on protocol level issues. The goal is to support deeper analysis and lower the barrier for academic teams that want to study the platform. Meta said it plans to expand access later.

The enumeration study follows earlier work from the same research group. They previously examined how delivery receipts can be triggered in ways that reveal activity patterns, device switches and session counts. Their combined findings show how small fragments of metadata can be stitched together into meaningful profiles.

The researchers argue that constant scrutiny remains necessary as messaging systems change over time. Meta echoed the reminder that its platforms draw attention from attackers and researchers alike. The size of WhatsApp’s user base gives every flaw wider reach, which makes independent testing and clear disclosure important parts of the security ecosystem.

Notes: This post was edited/created using GenAI tools and reviewed by human editor. Image: DIW-Aigen

Read next: 

• Meta Wins Key Ruling as Judge Rejects FTC Push to Break Up Instagram and WhatsApp
by Irfan Ahmad via Digital Information World

Meta Wins Key Ruling as Judge Rejects FTC Push to Break Up Instagram and WhatsApp

Meta secured a major legal victory after a federal judge ruled that the company did not violate antitrust laws when it bought Instagram and WhatsApp. The decision ends the current attempt by the Federal Trade Commission to force a breakup of the two platforms. The agency is still reviewing its options, so the broader matter may not be fully settled, but this ruling closes the latest chapter of a case that began in 2020.

The FTC had alleged that Meta created a monopoly in personal social networking by acquiring fast growing rivals. Regulators claimed the company used large acquisitions to limit competition and wanted Instagram and WhatsApp separated to restore market balance. This push came years after Meta completed both purchases.

The first version of the FTC’s lawsuit was dismissed in 2021 because the agency did not provide enough evidence of monopoly power. The agency returned with an amended case a year later. It described Snapchat and MeWe as Meta’s closest competitors for sharing updates with friends and family. Meta challenged that framing and said the FTC left out major platforms that shape user behavior across social media.

Judge James Boasberg agreed that the FTC’s market definition was too narrow. Evidence showed that users often moved between apps like YouTube and TikTok when Meta’s services experienced outages. The judge noted that TikTok had become such a strong competitor that Meta spent four billion dollars on Reels last year. This level of investment, driven by competitive pressure, did not match the picture of a market under one company’s control.

Meta argued that acquiring companies with strong products is a legitimate way to build new features. The court accepted that reasoning and found that the FTC had not demonstrated that these acquisitions harmed competition in the current environment. The judge pointed out that the landscape had changed since the early years of Facebook’s rise and that platforms now compete in more varied ways.

The ruling prevents the FTC from forcing divestment for now. The agency said it was disappointed and would examine its next steps. The case forms part of a much broader antitrust push against several major technology firms. Other actions are still underway, including those involving Alphabet’s Google and Apple.

For Meta, the decision removes an immediate threat to the structure of its platform group. It also brings some clarity to a question that has surrounded the company for years. The outcome does not resolve every regulatory challenge, but it does reshape how the courts view competition across today’s social apps.

Notes: This post was edited/created using GenAI tools. Image: DIW-Aigen. 

Read next:

• Shift Brower Report Finds AI Adoption Is Increasing Amid Concerning Skepticism

• Global Web Access Faces Its Hardest Year Yet as Rights Recede
by Irfan Ahmad via Digital Information World

Shift Brower Report Finds AI Adoption Is Increasing Amid Concerning Skepticism

A new study released by next-generation Internet browser Shift unveils a growing divide over how Internet users approach artificial intelligence capabilities online, with deep divisions between AI capabilities and user trust in them.

Among the top concerns shared over the integration of AI into modern browsers are privacy issues and environmental impact. However these concerns have not noticeably tempered the demand for AI features.

Topline Context

According to the report, AI usage is both high and growing. Of those surveyed, 82% report engaging with AI at least occasionally, with tech workers and younger users indexing higher on their usage.

However AI is not yet overtaking traditional search as a source of answers to users' questions. According to the study, 68% still rely on traditional search engines like Google, compared to only 21% citing AI tools like ChatGPT (about 10% say they use both equally).

The outlook for that to change is only slightly leaning AI’s way. Less than a third (32%) say they expect to use AI tools more in the future, compared to 44% who say they’ll stick with traditional search.

Instead, AI use is centered around three main areas:

• Research Assistance: 49%
• Task Automation: 37%
• Drafting Content: 34%

These results shift a bit between demographics, with 50% of Gen Z saying they primarily want AI to provide personalized recommendations, compared to 33% of the total audience surveyed.

AI Barriers: Privacy and Planet

For all of AI’s potential and expressed interest among browser users, the single largest concern reported is privacy, with nearly half (45%) of respondents citing privacy concerns as their main reason for hesitancy. Driving this is a lack of trust in where their data and information goes.

Another 35% say they don’t trust how AI-generated content might be used, while a nearly equal number (34%) expressed doubts about the accuracy of AI results. The result is a trust gap that requires more than a slick user interface to address.

Instead, for AI to grow further, providers must make transparency a default feature, with explicit details provided about how data is handled, models are trained, and privacy protected.

The environmental cost of AI is another major concern, with 57% of respondents saying they are either “very” or “somewhat” concerned about the energy and water consumption of AI infrastructure. Only 24% say they’re not concerned, and 19% were not aware of the energy usage issue.

Interestingly, those figures increase among those using AI the most. IT and Tech workers (the most active AI users at 62% daily) say they are either very concerned (35%) or somewhat concerned (44%) about the environmental impact of their AI use.

These results don’t exist in a digital vacuum. AI isn’t the only energy-consuming technology that has users concerned. Streaming video, cloud-stored photos, and internet traffic may all feel innocuous enough, but have an impactful carbon cost. As a result, environmental responsibility has evolved from a platitude to an essential strategy for businesses. Those who accept this reality today will be better positioned to thrive tomorrow.

The AI Paradox

Taken together, a certain paradox emerges. Those who use AI the most are the ones with the greatest concerns about its impact. Those seeking the most personalized experiences also cite the greatest concerns over privacy. The tension between confidence and control, and between usage and impact, stands to shape the next phase of AI’s growth.

Adoption alone can no longer be used to measure the future success of AI tools and technologies. The path forward depends on trust, transparency, and tangible value. While users are always eager to embrace tools to make everyday tasks easier and more convenient, a sizable portion of the population will hold back as privacy, data use, and environmental concerns become more pronounced.

This suggests the next phase of AI adoption must hinge on how well technology providers address these worries and how successful they will be in turning curiosity into confidence. For technology to earn its place in our daily lives, it must prove it can serve both human productivity and human values equally.

The State of Browsing Report, based on a survey of 1,000 adults in the U.S., was conducted this past September. View it for free here.

Read next: Anthropic CEO Dario Amodei Says AI Is Moving Faster Than Its Oversight

by Web Desk via Digital Information World

Study Maps the Conditions That Trigger AI Citation Hallucinations

A new look at GPT-4o shows how easily citation trouble grows when the topic drifts into areas with thin research trails. The warning signs appear clearly in a controlled test where the model produced six mental health literature reviews and filled them with 176 citations. A closer inspection showed that 35 of those references did not exist and many of the remaining ones carried mistakes hidden in the details. The numbers set the tone for a pattern that shifts sharply depending on how familiar the topic is in the research world.

Researchers at Deakin University built the experiment around three disorders. Depression sat at the top of the visibility ladder, followed by binge eating disorder, then body dysmorphic disorder at the bottom. This mix created a natural gradient in research volume. Depression carries decades of trials and thousands of papers. The other two conditions occupy smaller footprints and offer far fewer studies on digital interventions. That uneven landscape became the test bed for the model’s strengths and misses.

Each disorder received two review requests. One prompt asked for a broad overview that covered causes, impacts and treatments. The other request drilled into digital interventions. The team wanted to see how topic familiarity and prompt depth shaped the reliability of the citations. They pulled every reference into a manual check across major academic databases. This process placed each citation into one of three buckets. Either it existed in the real world, it existed but contained errors, or it was fabricated outright.

The headline numbers make the problem easy to see. Out of 176 total citations, 35 were fabricated. Among the 141 real ones, 64 carried errors. Only 77 came through fully accurate. That means around half of all citations were unusable in scholarly work. DOI errors were the most common type of error. Wrong links, wrong codes, or completely invalid strings made many citations look correct at first glance but fail when checked against the actual paper.

The pattern became sharper when the team compared the three disorders. Depression showed the lowest fabrication count with only four fake citations out of 68. Binge eating disorder jumped to seventeen fabricated citations out of sixty. Body dysmorphic disorder followed closely with fourteen fabricated citations out of forty eight. Accuracy among the real citations also depended on the topic. Depression reached sixty four percent accuracy. Binge eating disorder reached sixty. Body dysmorphic disorder fell to twenty nine. The drop shows how the model struggles once the evidence base gets thin enough.

Prompt specificity also shaped outcomes, though not in a simple way. Binge eating disorder showed the clearest effect. Its specialized review saw fabrication rise to almost half of the citations. The general overview stayed closer to one out of six. Other disorders showed different patterns. Depression’s general overview delivered better accuracy than its specialized review. Body dysmorphic disorder flipped that pattern and showed better accuracy when the prompt narrowed. These differences suggest the model reacts to the structure of the request and the strength of the underlying literature in different ways.

The study’s authors point out how much the model leans on patterns in public information. When the topic sits on a wide and stable base of research, the model has clearer pathways to follow. When the topic shifts to areas with fewer papers or narrower lines of inquiry, the model relies more on guesswork. The results from body dysmorphic disorder show how quickly accuracy collapses when the system tries to piece together references from scattered or limited material.

These findings matter because more researchers have started using large language models to speed up routine tasks. Survey data shows strong adoption among mental health scientists. Many researchers believe these systems help with drafting, coding, and early idea formation. Efficiency gains look promising until the citations fall apart under verification. That creates problems for anyone who trusts the output without checking every reference. A fabricated citation can mislead a research team, distort the evidence trail, and send other scientists searching for sources that were never written.

The study pushes institutions and journals toward simple safeguards. Every AI generated citation needs to be verified. Every claim tied to those citations needs human confirmation. Editors can screen suspicious references by checking whether they match known publications. When a citation sits outside any recognized record, it becomes a clear red flag. With these checks in place, journals can block fabricated references before they reach print.

The authors also point to the need for stronger guidance at universities and research centers. Training programs can help researchers learn how to identify hallucinations and validate AI generated content before placing it in a manuscript. As AI tools become part of normal workflows, these checks will keep the academic record from drifting into mistaken territory.

The results show that reliability is not static. It depends on the openness of the research terrain. Well studied disorders give the model a broader map. Narrower or less familiar topics cut away those supports. For now, the safest way to use these systems in research is to treat their output as a starting point that always needs careful checking. The experiment makes that reality clear.

Read next: Meta Adds New Content Protection Tools to Help Creators Spot Copycats
by Asim BN via Digital Information World