From Labels to Contracts: How Consumers Use AI Today

• ICERTIAS surveyed 6,000 adults across four continents to understand how consumers use AI in everyday life

• The main objective of the study was to understand not only how many consumers use AI for shopping advice, but also how many ask AI to explain small print on product packaging and complex contracts

• Only 14% of adults had ever sent AI a photo of a product or product label, including 4.5% who had done so in the past month

• Around one in ten adults (10.3%) had asked AI to explain a contract before signing or accepting it

• Food and beverages were the most common products photographed for AI analysis (35.2%), followed by beauty and personal care products (23%)

• After using AI to check a product photo, 22.1% of consumers bought a different product, while 12.1% decided not to buy

• Banking, telecommunications and insurance accounted for 65% of contracts submitted to AI, making them the three sectors where consumers most often sought AI explanations

The most consequential line on a product label may be the one printed in the smallest type. It could qualify a claim, identify an ingredient, or explain an exception. On a bank contract, its equivalent might describe a fee that appears months later. Both can be difficult to examine when someone is ready to decide.

Today, a growing number of consumers have another way to approach those words. A photograph of a package or a copy of an agreement can be submitted to an AI assistant with a plain question: What should I know before I accept this? The text stays where it was. The cost of questioning it changes.

That is the intriguing finding beneath the shopping statistics in ICERTIAS's October 2026 study, "AI at the Point of Choice". The organization surveyed 6,000 adults across the United States, United Kingdom, Germany, France, Japan, and Brazil, asking about AI advice before purchases, photographs of products, and explanations of contracts before signing.

The research does not establish that every photograph concerned fine print, nor does it demonstrate that AI explanations were accurate. It shows something narrower and commercially important: some consumers already seek outside help at moments when a company's own words may influence a decision. Understanding that behavior matters more than forecasting automated shopping carts.

The Cost of Understanding

For years, businesses have been able to distribute more information than most customers have time to examine. Packaging carries specifications and qualifications. Financial agreements contain definitions, conditions, and exceptions. Much of that information serves a legitimate purpose. But the effort needed to locate and interpret a relevant passage usually falls on the person making the decision.

A consumer might understand a headline promise yet miss the limitation that follows it. Reading every line is possible; knowing which line matters is harder. AI offers a practical shortcut because a person can ask a specific question about an unfamiliar term instead of first working out how the document is organized.

This does not make AI a lawyer, nutritionist, or product tester. An assistant can omit a condition, misinterpret a figure, or supply an answer that sounds certain without being supported. Its appeal lies in lowering the initial effort of asking questions. Its weakness is that a convincing explanation and a correct one are not necessarily the same.

A KPMG Consumer Snapshot published on October 8, 2026, illustrates the appeal of independent assistance. In its German research, 73% of respondents said they could envisage using an AI agent to compare products, prices, or offers. When objectivity was the issue, 64% preferred general-purpose AI over retailer or manufacturer systems, which only 7% favored.

The KPMG questions are different from those asked by ICERTIAS, and its findings should not be treated as confirmation of ICERTIAS's percentages. They do, however, explain why the shift deserves attention. Customers may increasingly prefer an explanation from a service they choose rather than from the company asking for their money.

When Packaging Becomes Evidence

In the ICERTIAS survey, 23% said they had sought AI advice before buying a product or service in the previous 30 days. Forty-three percent had done so within the previous 12 months, and 49% at least once. Those figures describe advice-seeking, not purchases completed by an AI agent.

Photographing a product is a more specific behavior. Fourteen percent said they had at least once sent an AI assistant a photograph of a product or its label for advice before purchasing. Eleven percent reported doing so within the past year, including 4.5% in the previous 30 days. Most respondents had never tried it.

Among those who had submitted a photograph, food and beverages accounted for 35.2% of the latest products checked. Beauty and personal care represented 23%, electronics and appliances 17.9%, cleaning and household goods 10%, and other products 10.9%. Another 3% could not recall the category.

Food, beverages, beauty, and personal care therefore accounted for 58.2% of the most recently photographed products within this subgroup. That is a distribution among photo-check users, not an adoption rate for shoppers in those categories. The distinction matters because the survey did not record the contents of the images or the questions accompanying them.

Still, the setting is consequential. A shopper may be holding a food item while deciding whether its ingredients suit a household's needs. Someone comparing skincare products might want an unfamiliar term explained. Those are illustrations of possible uses, not behaviors individually established by the survey. The common feature is that the product itself becomes evidence in a conversation.

The Interactive Advertising Bureau's September 9, 2026, study offers a related perspective on whose explanations people value. Among 2,200 recent AI shopping researchers across five markets, 56% preferred AI recommendations that included creator perspectives. That survey did not study label photographs. It nevertheless highlights demand for information beyond the seller's own account.

After the Photograph

The outcome question produces the study's most commercially arresting figures. Among respondents who had checked a product photograph with AI, 50% said they bought the original product. Another 22.1% bought a different product, while 12.1% decided not to buy. Ten percent postponed the purchase, and 4.8% were still deciding.

The remaining 0.9% did not know what happened. Combining alternative purchases and decisions not to buy gives 34.2%. Including postponed purchases brings those three outcomes to 44.2%, with another 4.8% still deciding. These are proportions of photo-check users, not of the entire adult sample or all shoppers.

It would be tempting to describe that 34.2% as sales diverted by AI. The survey cannot support that conclusion. Respondents reported what they did after receiving advice, but there was no comparison group showing what they would have purchased without it. A buyer might have changed their mind for several reasons, including price or availability.

There is another caution. Buying the original product does not prove that AI endorsed it; buying an alternative does not prove the assistant criticized the first. The outcomes establish what respondents reported next. They do not reveal the advice itself, its accuracy, or the degree to which the consumer followed it.

For a manufacturer, though, even the possibility of an extra check alters the questions worth asking. Does the product communicate its strongest evidence clearly? Could a shopper distinguish a material limitation from a routine warning? If a competing package makes its case more intelligibly, the comparison may begin before either product reaches the checkout.

PYMNTS Intelligence reported in its October 2026 Consumer AI Benchmark that 23% of US consumers primarily began online retail research with AI, while Amazon accounted for 60% of respondents' most recent AI-assisted purchases. Its measures differ from ICERTIAS's. Together, the studies separate influence over research from ownership of the final sale.

Contracts Under New Scrutiny

The same question becomes more consequential when the document is a contract. ICERTIAS found that 10.3% of respondents had at least once submitted a photograph, document, or text of an agreement to an AI assistant for explanation before accepting or signing it. Some 5.7% reported doing so within the previous 12 months.

Those numbers are much smaller than the figures for general shopping advice. They also describe a more demanding task. A comparison between two products can end with a changed preference. A misunderstood contract may leave a person with obligations that persist long after the initial decision has been made.

Of respondents who had sought AI explanations of contracts, 25% said the most recent concerned banking or credit. Telecommunications accounted for 21%, insurance 19%, employment 14%, other agreements 11%, and property purchase or rental 9%. One percent could not remember. Banking, telecommunications, and insurance together represented 65% of these latest contracts.

That concentration does not establish that contracts in those industries are worse written. It indicates where these respondents had last used the technology. Nor does the survey show which clauses they questioned. Cancellation provisions, premiums, interest calculations, exclusions, and renewal conditions are plausible subjects, but they were not separately measured.

How consumers supplied the material is revealing. Fifty-four percent uploaded a PDF or other document, 26% pasted copied text, and 15% sent a photograph. Four percent used another method, while 1% could not recall. Document uploads and pasted text together accounted for 80% of the reported submissions.

For companies, that creates a new audience for contract language. An agreement needs to work when a customer reads it directly and when a third-party assistant summarizes it. A qualification that is legally present but difficult to locate may be omitted from a poor summary. Clear structure and consistent terminology become practical safeguards.

There is a countervailing risk. Uploading an agreement can expose names, financial information, or other sensitive details to an external service. A September 30, 2026, study by Synchrony and Oxford Economics found that US consumers ranked data security and transparency about data use above convenience when considering AI shopping services. It did not assess contract uploads specifically.

Regulators are noticing adjacent behavior. On September 25, 2026, the UK's Ofcom reported that about one in twelve online adults had used AI to help manage telecommunications services. This is not equivalent to sending a contract for review. It suggests, however, that independent digital assistance is becoming relevant beyond finding products.

The Business of Explanation

If asking questions about fine print becomes easier, companies may have to reconsider what constitutes a successful disclosure. A term can be technically available yet poorly understood. A consumer who asks an assistant to identify the costliest exception may encounter that exception first, rather than reaching it after several pages of reading.

That creates a different design problem from traditional advertising. A prominent headline may attract attention, but the accompanying evidence must survive close inspection. Companies should be able to explain why a claim is true, which product or contract it covers, how long it applies, and what qualifications materially change its meaning.

The obvious place to begin is with actual packaging and agreements. Photograph them using ordinary phones, submit the same questions through several assistants, and compare the explanations with the underlying documents. Record omissions, incorrect interpretations, and unsupported conclusions. Such tests would not establish how every assistant behaves, but they would expose avoidable ambiguities.

For contracts, the test should include the clauses people are least likely to find unaided. Can the assistant identify a renewal trigger? Does it distinguish a promotional rate from a continuing rate? Does it preserve exclusions when explaining insurance coverage? The point is to improve the source document, not simply optimize language for a chatbot.

Independent evidence may gain practical importance in this environment, but it must remain verifiable. A quality mark or consumer award can supply a useful reference only when its issuer, research basis, dates, scope, and meaning are clear. Otherwise, a fluent AI explanation risks amplifying a claim that customers cannot check.

The prospect also raises a question about customer service. If customers arrive with machine-generated interpretations of product claims or contracts, employees will need to distinguish genuine ambiguity from AI error. Clear escalation routes and accessible explanations of disputed clauses may matter as much as improvements to the original text.

Businesses should not assume that widespread adoption is inevitable within a fixed number of years. The ICERTIAS data provide a snapshot, not a forecast. But waiting for a majority to use these tools would miss the operational opportunity already visible: ensuring that the information customers are permitted to inspect can withstand inspection.

The Question After Reading

The most useful insight from this survey is not that AI will make every decision. Many people have never sought its shopping advice, and most have never uploaded a product photograph or contract. Consumers who do use it may still ignore its answers or seek confirmation elsewhere. The evidence supports a change in practice, not a completed revolution.

There is nevertheless a material difference between information being disclosed and information being readily questioned. For a long time, the main barrier to understanding some consumer documents was the work required to read them closely. AI offers another reader, imperfect but available on demand. That makes the smallest line harder to assume nobody will notice.

A package can still win attention with its front label.

A contract can still secure acceptance with its signature box. Increasingly, however, the decision may turn on a sentence neither company intended to put in the spotlight.

The commercial test is whether that sentence remains defensible when someone asks AI what it really means.

How the Research Was Conducted

ICERTIAS interviewed 6,000 adults aged 18 or older between October 12 and 21, 2026, using computer-assisted web interviewing (CAWI). It covered Brazil, France, Germany, Japan, the United Kingdom and the United States, with 1,000 respondents per country. Participants were recruited through online advertisements.

The study used nonprobability quota sampling by age, gender, region and education. Results were weighted to official demographic statistics. ICERTIAS applied checks for duplicate entries, automated participation, implausibly fast completion and inconsistent responses. Respondents were real people; no synthetic survey participants or answers were included.

The sample was not random, so ICERTIAS reports no conventional margin of sampling error. Combined figures weight the countries equally, regardless of population. Exact unweighted bases for follow-up questions were not supplied with these figures.

The Questions Behind the Findings

The following English translations reflect the seven questions supplied by ICERTIAS. The survey began: “When was the last time you sought advice from an AI assistant before purchasing a product or service?” It then asked: “When was the last time you photographed a product or its label and sent the photograph to an AI assistant for advice before purchasing?”

Next came two questions about the most recent instance: “Which product did you most recently photograph and send to an AI assistant for analysis?” and “Did the AI assistant’s advice lead you to change your decision to buy that product?” These linked a specific item to the reported purchase outcome.

The contract section asked: “When was the last time you sent an AI assistant a photograph, document or text of a contract for explanation before accepting or signing it?” Respondents were then asked: “What type of contract was it?” and “How did you submit the contract to the AI assistant?”

What the Numbers Add

Nine percent had sought AI purchasing advice within seven days and 14% between eight and 30 days earlier. Overall, 23% had done so in the past month, 43% in the past year and 49% at least once. Forty-eight percent said they had never done so.

Photo-based checking remained less common. Some 4.5% had submitted a product or label photograph within 30 days, compared with 11% in the past year and 14% ever. Of the last products photographed, 35.2% were food or beverages, 23% beauty or personal care, 17.9% electronics or appliances, 10% cleaning and household products and 10.9% other products; 3% were not recalled.

Fifty percent of photo-check users reported buying the original product. A further 22.1% bought something else, 12.1% did not buy, 10% postponed buying and 4.8% were still deciding. The question concerned reported behavior following the advice. The figures cannot isolate what would have happened without AI.

Contract use was narrower: 5.7% said they had submitted contract material for explanation within 12 months and 10.3% had ever done so. Banking and credit agreements represented 25% of the most recent contracts in this subgroup, telecommunications 21% and insurance 19%. Employment agreements accounted for 14%, property 9%, other contracts 11% and uncertain recollections 1%.

Most users sent the contractual wording in a digital format: 54% uploaded a PDF or another document and 26% pasted text. Fifteen percent used a photograph, 4% another method and 1% could not recall. The survey did not test the accuracy of explanations or whether users removed sensitive information.

The results capture a specific early behavior: consumers asking a separate service to read information supplied by a seller or contracting party.

They do not predict adoption rates or establish better consumer outcomes.