Can answers we find on the web be trusted? Can we trust that our favorite e-commerce site is showing us all the options without bias and influence? It is a modern version of an old question. Can we trust a doctor who receives industry payments, a commissioned salesperson, a transaction-paid broker or a newspaper supported by advertisers? Human beings have always relied on people and institutions with interests of their own.
Trust has never required perfect neutrality. It requires enough understanding of a source to interpret its motives alongside its advice. Because we cannot independently verify every product, business, study or report, trust lowers both the cognitive and economic cost of decision-making.
A study of 344 supplier relationships in the United States, Japan and Korea found procurement transaction costs five times higher at the least-trusted automaker than at the most-trusted. Trust reduces verification and bargaining costs, which helps explain why trusted intermediaries acquire economic value.
The modern internet complicates that relationship because systems that reduce uncertainty often monetize the people they influence. Search sells visibility, marketplaces collect transaction and placement fees, social networks monetize attention, and publishers rely on subscriptions, sponsors and affiliates.
None of this proves misconduct. Commercial incentives can fund useful services and reduce search costs. The problem is informational asymmetry: the intermediary usually understands its revenue model and ranking logic far better than the user does.
Motives may also be political, institutional or reputational. The old challenge was deciding whether to trust an interested source. The modern challenge is harder because the source, its limits and the interests behind the answer may be difficult to see.
Can We Trust the Answer? Evidence on Trust and Transaction Costs
| Measure | Higher Trust | Lower Trust |
|---|---|---|
| Time spent contracting and haggling | About 21% | 47% |
| Face-to-face productivity | Higher | About 50% more contact time needed |
| Procurement transaction costs | Baseline | Up to 5× higher |
Sources: Dyer and Chu, INFORMS Organization Science
How the Internet Rebuilt Trust Around Systems
The early web largely preserved the traditional relationship between source and reader. Search organized access to newspapers, government documents, academic papers and retailers, but the source remained visible.
That relationship has changed. Digital systems now select, rank and summarize information; generative AI synthesizes it. A modern answer has an information supply chain in which content is created or licensed, filtered through access rules, retrieved, ranked and ultimately synthesized.
Content provenance asks who created the information. Availability provenance asks what the system could access. Retrieval determines what entered consideration, while synthesis turns those inputs into the final answer.
The user sees an answer. The system sees sources, rules, probabilities and commercial relationships; it may also see a profile of the person asking.
Those information universes can be commercially bounded. Publishers license archives, research databases restrict access, and Cloudflare’s Pay Per Crawl lets participating site owners charge AI crawlers. Contractual and technical boundaries can shape retrieval before ranking begins. Absence can have provenance too.
Profiling adds another layer. An FTC surveillance-pricing study found that intermediaries serving at least 250 business clients could use signals as granular as location, browsing history, shopping behavior and mouse movements to tailor prices or product presentation. The findings remain preliminary, but they show how behavioral data can become an economic input.
How the Internet Rebuilt Trust Around Systems — Evidence on Machine Access
| Access Condition | Current Cloudflare Mechanism |
|---|---|
| Minimum paid crawl price | $0.01 per successful crawl |
| Publisher choices | Allow, charge or block |
| Payment required response | HTTP 402 |
| Billable retrieval | Successful HTTP 200 response |
| Cloudflare network reach | More than 20% of the web |
Sources: Cloudflare
The Business and Power of Being Trusted
The commercial web has to pay for itself, and business models create incentives. Paid search monetizes visibility, sponsored placement monetizes position, and affiliate programs monetize transactions.
The trust problem begins when users cannot distinguish relevance from commercial preference. Advertising around information is different from commercial influence inside information selection.
That distinction grows more consequential as search evolves into conversation. A sponsored result occupies visible space, while a conversational system can potentially shape comparison and framing as the user forms a decision.
Advertising beside an AI answer is not evidence that the answer itself was commercially altered. The question is whether future business models preserve that separation as conversational systems gain influence.
Experimental evidence shows why the possibility matters. In two preregistered studies involving 2,012 participants and five frontier models, 61.2% of participants using conversational systems selected products randomly designated for promotion, compared with 22.4% under traditional search placement. Fewer than one in ten accurately detected concealed promotional steering.
These are experimental results, not measurements of deployed commercial AI. They nevertheless show that conversational persuasion can be examined as an economic mechanism rather than dismissed as a hypothetical risk.
Businesses are already trying to improve visibility inside generative systems. Political actors can pursue the same logic by shaping what a trusted intermediary sees. Trust has become infrastructure: it can be monetized, but it can also be depleted.
The Business and Power of Being Trusted — Evidence on Commercial Persuasion
| Experimental Measure | Finding |
|---|---|
| Products randomly designated sponsored | 20% of catalog |
| Sponsored selection with conversational AI | 61.2% |
| Sponsored selection with traditional search | 22.4% |
| Relative selection difference | Nearly 3× |
| Detection under concealed promotion | Below 10% |
| Effect of a Sponsored label | No significant reduction |
Sources: Salvi, Cuevas and Ribeiro
Steering Without Lying
Steering does not require a lie.
Platforms can influence behavior through omission, ranking, defaults, framing and social proof while every displayed statement remains technically accurate. The effects can be substantial.
A CMA-cited meta-analysis found default-based interventions produced an average effect of d=0.62, compared with d=0.36 for information-based interventions. Another found a preselected option was 27% more likely to be chosen.
A 2024 UK study found defaults on 49% of sampled websites and apps. Yet 64% of the observed defaults directed users toward cheaper or similarly priced options, an important reminder that choice architecture is not inherently harmful. In the controlled experiment, however, a preselected default made consumers 60% to 70% more likely to choose the more expensive option.
Ranking creates similar effects. CMA evidence shows consumers disproportionately inspect and select options near the top of lists. In one cited case, placement on Spotify’s Today’s Top Hits playlist was associated with roughly 20 million additional streams of potential exposure. At scale, ranking distributes economic opportunity.
Humans use shortcuts because the information environment is too large to evaluate exhaustively. Rankings suggest relevance, popularity creates social proof, familiarity lowers perceived risk, and conversational fluency can feel like expertise.
The evidence does not show uniform algorithmic trust. Some settings produce appreciation and others aversion. The deeper problem is calibration: users decide when verification is worth the effort using incomplete signals about expertise, neutrality and source quality.
Those trust signals have economic value. The FTC now prohibits fake reviews, AI-generated fake reviews and compensation conditioned on review sentiment, evidence that apparent popularity itself can become a market.
These effects also vary geographically. Communities poorly represented in machine-readable or retrievable information can become harder to discover or recommend. Information inequality can become discovery inequality.
Accuracy therefore answers only part of the authenticity question. An answer may be factually defensible while the surrounding choice environment remains selective. The harder question is why this source, product or interpretation became the answer this person received.
Steering Without Lying — Measured Effects of Choice Architecture
| Choice Intervention | Measured Effect |
|---|---|
| Defaults | d = 0.62 |
| Providing information | d = 0.36 |
| Self-commitment | d = 0.30 |
| Preselected option | 27% more likely to be chosen |
| Observed defaults favoring cheaper or similar options | 64% |
Sources: UK Competition and Markets Authority, UK Department for Business and Trade
Who Governs the Answer
Platforms govern ranking, recommendation, data access and AI behavior, while governments govern platforms through consumer protection, privacy and competition rules. Ranking itself is already treated as economic power: in July 2026, the European Commission fined Google €890 million in two Digital Markets Act decisions, including €460 million relating to self-preferencing in Search.
Governance creates its own trust problem. Governments have political interests, regulators operate within institutions, universities have funding structures, and companies lobby. The lesson is not that nothing is trustworthy.
The better objective is calibrated trust.
Labeling content AI-generated addresses only part of the problem. Answer provenance also asks what the system could access, what was excluded, which relationships mattered and what rules transformed those inputs into the answer.
Complete technical transparency is neither practical nor useful. Decision-useful transparency means enough information to judge reliability, incentives and uncertainty without requiring users to audit an entire system.
Regulation matters, but it should not transfer unquestioned trust from platforms to governments. Consumers do not need thousands of pages of algorithmic documentation, yet complete opacity makes informed trust impossible. The governance challenge is not perfect neutrality; it is meaningful visibility into consequential conflicts.
Who Governs the Answer — Scale of Digital Gatekeeper Enforcement
| DMA Measure | Scale |
|---|---|
| Initial designated gatekeepers | 6 companies |
| Initially designated core platform services | 22 services |
| Maximum standard non-compliance fine | 10% of worldwide turnover |
| Maximum repeated-infringement fine | 20% of worldwide turnover |
| Initial compliance period after designation | 6 months |
Sources: European Commission
From Choosing Sources to Choosing Systems
AI does not create the trust problem. It may complete a progression already underway. Traditional search presented sources and left users to evaluate them; generative AI increasingly presents synthesis: not “here are places to look,” but “here is the answer.”
Different AI systems need not construct that answer from the same world. Training histories, retrieval indexes, licensing, databases and jurisdiction can produce materially different information environments without explicit ideological design.
The Reuters Institute’s 2026 Digital News Report, based on 97,520 respondents across 48 markets, found social media and video networks were the leading route to online news at 54%, ahead of news organizations’ websites and apps at 51%. Including AI chatbots raised third-party intermediation to 56%.
AI use and trust do not move together neatly. Weekly chatbot use for news rose from 7% to 10% globally and reached 16% among people under 35. Only 20% globally said they trusted chatbot news, compared with 37% for news overall, while trust reached 44% among actual chatbot news users.
Verification has not disappeared either. Forty-two percent of chatbot news users said they often or always clicked through to original sources, while 33% said they used chatbots to assess a news source’s reliability. The intermediary is therefore becoming both a source of information and, for some users, a tool for judging other sources.
The stakes rise when interpretation becomes action. In August 2026, the UK Financial Conduct Authority reported that 56% of surveyed younger investors trusted AI tools. Yet 44% mistakenly believed AI-generated financial information was regulated, 38% would accept an investment decision based solely on AI output, and 32% expected compensation or ombudsman protection if AI advice caused harm.
Can AI be trusted? Sometimes. So can doctors, salespeople, newspapers, governments, marketplaces, academics and search engines. None deserves unlimited trust simply because of the role it occupies.
Humans have always needed trust to live, trade, cooperate, learn and decide. The internet did not eliminate that need. It industrialized the ability to shape where we place it.
The defining question of web authenticity may therefore no longer be simply, “Is this true?” It may increasingly be: Why did this become the answer I received, whose interests shaped it, and what do I actually know about the system asking me to trust it?
From Choosing Sources to Choosing Systems — AI News Use by Age
| Age Group | Used AI Chatbots for News in the Last Week |
|---|---|
| 18–24 | 17% |
| 25–34 | 15% |
| 35–44 | 11% |
| 45–54 | 8% |
| 55+ | 5% |
| AI as main news source | 1% overall |
Sources: Reuters Institute for the Study of Journalism
TL;DR Summary
- Trust lowers the cognitive and economic cost of independently verifying decisions.
- Digital systems increasingly select and synthesize information rather than merely transmitting it.
- Answer provenance includes source origin, availability, retrieval, ranking and synthesis.
- Commercial incentives do not imply misconduct, but opacity makes trust harder to calibrate.
- Conversational systems create new opportunities for commercial influence inside interpretation.
- Choice architecture can steer behavior without requiring false statements.
- Behavioral evidence points to trust calibration rather than universal algorithmic trust or distrust.
- Ranking can allocate visibility and economic opportunity.
- Poor digital representation can weaken discovery for communities and businesses.
- Regulation can improve accountability without making governments neutral arbiters.
- Different AI systems can operate over materially different information universes.
- As AI moves toward interpretation and action, calibrated trust increasingly depends on understanding why an answer appeared.
Sources
• INFORMS Organization Science; The Role of Trustworthiness in Reducing Transaction Costs and Improving Performance; – Link
• OECD; OECD Survey on Drivers of Trust in Public Institutions 2024 Results; – Link
• World Bank; What Is Trust Why Does It Matter for Development and How Do We Measure It; – Link
How the Internet Rebuilt Trust Around Systems
• Federal Trade Commission; Surveillance Pricing Study Indicates Wide Range of Personal Data Used to Set Individualized Consumer Prices; – Link
• Cloudflare; What Is Pay Per Crawl; – Link
• Pew Research Center; What Web Browsing Data Tells Us About How AI Appears Online; – Link
The Business and Power of Being Trusted
• Salvi Cuevas and Ribeiro; Commercial Persuasion in AI Mediated Conversations; – Link
• Google; A New Generation of Ads for the AI Era of Search; – Link
• OpenAI; Testing Ads in ChatGPT; – Link
• McKinsey & Company; The Agentic Advertising Economy From Attention to Action; – Link
Steering Without Lying
• UK Competition and Markets Authority; Evidence Review of Online Choice Architecture and Consumer and Competition Harm; – Link
• UK Department for Business and Trade; The Prevalence and Potential Harm of Defaults in Online Shopping; – Link
• Federal Trade Commission; Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials; – Link
Who Governs the Answer
• European Commission; Commission Fines Google €890 Million for Breaches of the Digital Markets Act; – Link
• European Commission; DMA Designated Gatekeepers; – Link
• European Commission; Guidelines on Transparency Obligations for Providers and Deployers of AI Systems; – Link
From Choosing Sources to Choosing Systems
• Reuters Institute for the Study of Journalism; Digital News Report 2026 Overview and Key Findings; – Link
• Reuters Institute for the Study of Journalism; Emerging Uses of AI Chatbots for News and What It Means for Journalism; – Link
• Financial Conduct Authority; Young Investors Trust AI More Than TV or Celebrities; – Link
• Pew Research Center; Americans and AI 2026 Chatbots Smart Devices and Views on Impact; – Link