Claims that consumers “prefer phone” or “have moved to digital” often combine evidence designed to answer different questions. Satisfaction indexes describe evaluated experiences; administrative systems describe submitted problems; survey-method research concerns respondent contact; and household internet tables measure access. None directly measures which service channel consumers would choose under otherwise identical conditions.
This review separates those evidence classes. The main geography is the United States. The periods differ: the American Customer Satisfaction Index (ACSI) publishes industry results for specified study periods; the Consumer Financial Protection Bureau (CFPB) and Federal Communications Commission (FCC) offer continually updated administrative data; the Federal Trade Commission (FTC) source summarizes calendar year 2024; and the Census table covers the 2023 American Community Survey (ACS) one-year estimate. The units range from survey responses and modeled index scores to complaints, fraud reports, households, and people. The cohorts are therefore non-equivalent and must remain labeled as such.
Satisfaction is not channel preference
ACSI’s retail and consumer shipping results report customer-satisfaction evidence for industries and companies covered by its research. The relevant unit is an index-based satisfaction result derived from sampled customer evaluations, not a count of support calls and not an election among telephone, email, chat, social media, or self-service. Industry pages can support comparisons that preserve ACSI’s study period and modeled construct. They cannot establish that all U.S. consumers prefer the channel used in a highly rated experience.
ACSI’s science of customer satisfaction describes survey inputs and an econometric model connecting expectations, perceived quality, perceived value, satisfaction, complaints, and loyalty-related outcomes. This is original survey methodology for a defined construct, but that construct is satisfaction—not universal channel preference.
Respondents qualify through experiences with covered companies or industries during a designated period. ACSI publishes benchmarks and offers related commercial products, a sponsor context that should be considered alongside its disclosed methodology. Because scores can combine many aspects of an experience, they cannot isolate phone or digital service without a channel-specific design.
CFPB complaints measure submitted financial-product problems
The CFPB’s Consumer Complaint Database contains consumer-submitted complaints about financial products and services that pass the Bureau’s publication process. Its units are complaint records and associated fields, such as product, issue, company, dates, geography where supplied, response status, and optional narrative. Its cohort is people who knew of, accessed, and used the CFPB process—not all customers of financial institutions and not a probability sample of U.S. adults.
The database’s public API enables reproducible filtering. Analysis should record the query date, period, filters, geography, narrative requirement, and unit counted. Because records are processed over time, extracts made on different dates can differ.
Complaint counts depend on exposure and process. Larger companies have more opportunities for complaints, while awareness, willingness to complain, access, language, severity, and referrals affect submission. Counts without customer or transaction denominators are not rates. Channel-related text may describe several failed contacts, but filing does not reveal the consumer’s general channel preference.
FCC records cover communications issues, not a population poll
The FCC’s Consumer Help Center Data provides data associated with consumer complaints and inquiries in communications domains under FCC jurisdiction. These administrative records can identify subjects generating submissions over selected periods and places. Their unit is a submitted ticket or complaint as defined by the FCC data, and their cohort is consumers who reached that system. Duplicate contacts, categorization, filing assistance, and data-field completeness can affect analysis.
The regulated subject may itself be telephone, broadband, television, accessibility, or unwanted communications. A phone-service complaint does not express a preference for phone support, and a broadband complaint does not demonstrate preference for chat. The records remain conditioned on filing behavior and FCC scope.
FTC Sentinel reports reveal harms reported to a network
The FTC’s Consumer Sentinel Network Data Book 2024 summarizes reports received during calendar year 2024, including fraud, identity theft, and other consumer problems. The geography includes the United States with state and other breakdowns where data permit. The units are reports and reported monetary losses under Sentinel definitions. Reports reach the network through the FTC and participating data contributors, so source composition is part of the evidence.
This harmed-or-concerned reporter cohort is not a sample of routine service users. Volume can reflect underlying harm, awareness, reporting access, contributor mix, or categorization. Reported loss is not necessarily verified loss. Contact methods used by alleged fraudsters do not measure preferred channels for contacting legitimate organizations.
USA.gov’s consumer-complaint guide shows why complaint destination varies: consumers are directed toward different steps and agencies according to the product, problem, or company. That routing function is useful to a complainant. Analytically, however, it means no single complaint repository contains a census of all U.S. consumer dissatisfaction.
Survey collection mode is not respondent service preference
Pew Research Center’s survey data collection overview explains tradeoffs among telephone, online, mail, and mixed-mode approaches. This is methodological evidence about how researchers recruit and interview samples. It highlights coverage, response, measurement, cost, and operational considerations that can vary by mode.
Completing a survey online does not mean preferring online customer service. Survey mode may be assigned or constrained by the sample frame, and it can affect participation and answers. A channel-preference estimate needs a defined service scenario, transparent sample, field dates, weighting, response information, and uncertainty. A digital-only sample may select for digital access.
Pew’s role here is methodological rather than the source of a specific nationwide channel-preference percentage. Its discussion helps test a preference survey’s design; it should not be cited as if all surveys share Pew’s sampling or weighting practices.
ACS internet access describes capability and constraint
The Census Bureau’s ACS overview describes a continuing national survey producing social, economic, housing, and demographic estimates. The 2023 ACS one-year S2801 table reports types of computers and internet subscriptions. For this source, the period is 2023, the principal unit is the household for many table measures, the cohort is the ACS-covered population or occupied-household universe specified by the row, and the geography can range from the United States to eligible subnational areas for one-year estimates.
Internet subscription indicates capability, not preference for chat, email, apps, or self-service. Household access does not guarantee an appropriate device, accessibility, privacy, skills, or reliable connectivity when service is needed. People without a home subscription may obtain access elsewhere.
The Bureau’s ACS methodology documents sampling, data collection, weighting, nonresponse treatment, and estimation. Published estimates carry margins of error, and differences should be assessed with that uncertainty rather than by comparing point estimates alone. One-year estimates are not interchangeable with five-year estimates; different vintages and geographies should not be pooled casually.
What can responsibly be concluded
Together, these sources support a bounded conclusion: service access occurs within unequal technical conditions; satisfaction varies across evaluated industries and periods; and public complaint systems reveal important submitted problems. They do not yield one harmonized national ranking of phone, email, chat, messaging, social media, and self-service.
The strongest channel-preference evidence would directly ask a defined population about a defined task and allow channel choices that reflect realistic constraints. Even then, stated preference may differ from observed behavior, and behavior may reflect which channels an organization offers rather than what a consumer wants. Complaint data can identify failure modes worth investigating, but complaint data do not equal channel preference. Satisfaction scores can describe evaluated relationships, but they do not isolate channel causality. Internet access estimates establish neither desire nor service quality.
Accordingly, every channel claim should name its period, unit, cohort, geography, sampling or submission mechanism, denominator, and uncertainty. Keeping those labels visible is not a technicality; it is what prevents several valuable but non-equivalent evidence systems from being turned into a claim none of them measured.