The Psychology of Selling Online
15 Evidence-Based Principles

A feature is something a product has. Value is what that feature changes for the customer.

The distinction sounds obvious, yet many online shops and digital product pages are built almost entirely around specifications. A bicycle may have a particular carbon lay-up. Software may include automated reporting. A course may contain 40 video lessons. None of those facts has much persuasive force until the buyer understands why it matters.

Research into customer-perceived value describes a purchase as a balance between expected benefits and expected sacrifices. The sacrifices include more than the price. They can also include time, effort, risk, inconvenience and the opportunity cost of choosing one product instead of another. A comprehensive meta-analysis of customer-perceived value found that value is connected with outcomes including satisfaction, word of mouth and repurchase intentions.

That gives us a more useful way to construct an offer:

• Identify the customer for whom the product was designed.
• Define the problem, task or desire that brings that person to the page.
• Explain the practical outcome the product helps them achieve.
• Use features, specifications and evidence to support that outcome.
• Be honest about the time, money or effort the customer must contribute.

For a physical product, this might mean translating "double-wall stainless steel" into better temperature retention and durability. For a digital product, it might mean explaining how an automated report replaces a repetitive monthly task, not merely stating that "automation" is included.

Buying online requires a decision based on incomplete information. The customer cannot always touch the material, test the fit, inspect the construction or speak to the person responsible for the product. Digital products introduce a different version of the same problem: the buyer cannot be sure whether the content is sufficiently detailed, the software will fit their workflow or the promised outcome is realistic.

Studies of online product presentation suggest that demonstrations can improve perceived diagnosticity (the customer's sense that the information helps them judge the product) and make it easier to imagine using it. For example, experimental research on product presentation videos found that the type of video influenced mental imagery and purchase intention. Research into touch and perceived ownership also helps explain why tactile products are difficult to sell using generic catalogue imagery alone.

The objective is not to make every product page longer. It is to remove the particular uncertainty that prevents a sensible customer from deciding.

"Premium." "Powerful." "Seamless." "Next-generation."

These words are easy to write because they do not require the seller to commit to anything. That is also why they are weak. The customer has to translate them into a meaningful claim, and different readers will make different assumptions.

Concrete language gives the buyer something they can understand and evaluate:

This is not just a copywriting preference. Research by Grant Packard and Jonah Berger combined controlled studies with analysis of more than 1,000 real customer-service interactions. Customers responded more favourably when employees used concrete language, and the field data connected that language with satisfaction and subsequent purchasing.

Specificity can improve several parts of an online offer:

• It demonstrates that the seller understands the customer's question.
• It reduces the chance that buyer and seller are imagining different things.
• It makes a claim easier to verify.
• It sets clearer expectations, which can reduce disappointment after purchase.

Concrete does not mean drowning the reader in numbers. It means choosing details that help someone make the decision. The exact weight of a portable product may matter; the internal reference number of one of its components probably does not.

Customers do not experience a product page as a neutral container of information. They experience the effort required to make sense of it.

Processing fluency is the subjective ease with which something can be perceived or understood. Research has associated greater fluency with more favourable judgements in a range of settings. The important commercial lesson is not that everything must look simplistic. It is that avoidable difficulty can become part of the customer's judgement of the offer itself.

In a well-known experiment, instructions shown in a difficult-to-read typeface caused people to estimate that the described task would require more time and effort. The paper's title captures the result neatly: "If It's Hard to Read, It's Hard to Do". A broader review of processing fluency explains why contrast, clarity, familiarity and perceptual organisation can influence aesthetic responses.

For an online seller, fluency means that a visitor should be able to establish four things without detective work:

Descriptive headings, legible typography, familiar navigation and a clear visual hierarchy all help. So does writing in ordinary language. A product page should not require the customer to decode internal terminology before they can compare two models.

Trust is sometimes treated as a collection of visual accessories: a padlock icon, a row of logos and a badge announcing that the site is "100% secure". Those elements are meaningful only when they represent something real.

Online trust is better understood as the customer's willingness to accept vulnerability. They may be paying an unfamiliar business, sharing personal information or relying on the seller to deliver something they cannot inspect in advance. The greater the possible loss, the more evidence they will usually require.

A meta-analysis of 150 studies of trust in e-commerce found significant relationships between online trust and factors such as perceived privacy and service quality, as well as outcomes including loyalty and repeat-purchase intentions. The relationships were not identical in every setting, which is precisely why a generic "trust badge" cannot solve every trust problem.

Useful trust evidence includes:

• A clearly identified business with a real address and accessible contact details.
• Payment methods customers recognise and know how to use.
• Plain-language privacy, delivery, return and cancellation policies.
• Accurate claims that can be checked against a credible source.
• Visible customer service and a realistic explanation of response times.
• Relevant professional accreditation where it genuinely applies.
• Consistency between the advertisement, landing page, checkout and post-purchase experience.

Trust is also cumulative. A confusing price, a copied product description or a policy written to conceal its meaning can undo the reassurance created elsewhere on the site.

Social proof is often reduced to volume: thousands of customers, millions of downloads or a row of five-star ratings. But the most useful proof does more than imply popularity. It helps the buyer answer, "Did this work for someone like me, in circumstances like mine?"

The evidence supporting online reviews is substantial. A 2024 meta-analysis of how reviews affect purchase intention found significant effects across the review characteristics it examined, with the strength varying between them. Earlier field research by Judith Chevalier and Dina Mayzlin linked online book reviews with sales rather than relying only on what participants said they intended to buy.

Credibility matters. A systematic review of online-review credibility found that readers use cues relating to the source, message and platform when deciding what to believe. This supports several sensible practices:

• Link reviews to verified transactions where possible.
• Show when the review was written and which product variant was purchased.
• Invite customers to describe the use case, not merely select a star rating.
• Retain legitimate criticism rather than manufacturing an implausibly perfect record.
• Disclose incentives and commercial relationships.
• Use case studies with a clear starting point, intervention and outcome.

Celebrity endorsement and high follower counts may attract attention, but attention is not the same as credible product evidence. Relevance usually matters more than fame.

Every purchase carries the possibility of regret. The item may not fit, the software may not integrate properly or the course may be pitched at the wrong level. Customers evaluate that possible downside alongside the potential benefit.

A clear return, refund or cancellation policy changes the calculation. It does not make the purchase risk-free, but it tells the customer what will happen if the product is unsuitable.

A meta-analysis of 21 papers on return-policy leniency found that more lenient policies increased purchases more than returns overall. The detail matters, however. The researchers distinguished between several dimensions of leniency, including time, money, effort, scope and exchange, and found that they did not influence purchases and returns in the same way.

That makes "offer free returns" too crude as a universal recommendation. Sellers must consider product margins, fraud, shipping costs and the condition in which returned goods can be resold. What should be universal is clarity.

A good policy is not only generous enough to reduce reasonable fear. It is also written so that a normal person can understand it before paying.

The famous lesson from "choice overload" is usually expressed as "fewer options produce more sales". The research is more nuanced.

A conceptual review and meta-analysis by Alexander Chernev and colleagues found that choice overload depends on identifiable conditions. It is more likely when the decision is difficult, the available options are complex, customers are unsure about their own preferences or the assortment does not provide an obvious way to compare alternatives.

This means that a large range is not automatically a problem. A specialist buyer may value it. Conversely, three poorly differentiated plans can be more frustrating than 30 well-organised products.

The better objective is choice assistance:

• Group products according to a distinction customers understand.
• Provide filters that reflect real buying criteria.
• Make differences between variants explicit.
• Offer a comparison table when several attributes matter.
• Explain who each option suits and who it does not.
• Recommend a product only when there is a defensible reason.

When customers abandon a category page, the answer may not be to sell fewer products. It may be to provide a better map.

Every extra demand placed on a customer asks them to decide whether the purchase is still worth the trouble.

That effort can be physical, typing card details on a small screen, or cognitive, such as working out whether an account is required, interpreting an error or trying to remember a password. The customer may still want the product. They simply no longer want the process.

Baymard Institute's large-scale checkout usability research repeatedly identifies problems such as forced account creation, excessive form complexity, late costs and unclear delivery information. Its current survey data should not be treated as a universal forecast for every shop, but the underlying usability problems are concrete and testable.

Useful improvements include:

• Offering guest checkout when an account is not essential.
• Supporting browser and payment autofill.
• Asking only for information needed at that stage.
• Placing error messages beside the field that needs correction.
• Retaining entered information after an error.
• Offering payment methods familiar to the target market.
• Keeping optional promotions and cross-sells secondary to checkout.
• Making the final action and its consequences unmistakable.

The number of checkout pages is not a reliable measure of effort. One page containing 30 poorly organised fields can feel harder than three short, logical stages. Reduce actual and perceived complexity rather than chasing a fashionable "one-step checkout".

Unexpected charges do more than make a product more expensive. They change the meaning of the price the customer saw earlier.

If a product is advertised at £50 and becomes £64 only after several checkout steps, the problem is not simply the additional £14. The customer has been encouraged to evaluate one offer and then presented with another. That can create frustration even when the final price remains competitive.

Baymard's checkout-abandonment research currently identifies late additional costs as a leading reported reason for leaving a checkout. The UK Competition and Markets Authority has also published an extensive evidence review of online choice architecture, including practices that affect how consumers receive and interpret choice information.

The practical standard should be early, intelligible disclosure:

• Show delivery costs or a reliable estimate before the final checkout stage.
• Explain thresholds for free delivery where the customer is likely to see them.
• Include compulsory fees in the prominent price wherever possible.
• State whether tax is included and which customers may pay a different amount.
• For subscriptions, show the billing interval, renewal price and minimum commitment.
• Distinguish a monthly equivalent from the total amount the customer is committing to pay.

This is not an argument against delivery charges, setup fees or paid upgrades. Businesses have real costs. It is an argument against relying on the customer's accumulated effort to make a late price increase feel difficult to refuse.

A price has little meaning without a comparison. Is £120 expensive? The answer depends on what is being bought, what alternatives cost, how long the product lasts and what the customer receives.

Consumer research has repeatedly demonstrated that the presentation of reference prices can affect perceived savings. Yet a meta-analysis of price presentation also found important differences between types of reference price. For example, a stated regular price could increase perceived savings, while a manufacturer's suggested price did not produce the same main effect. Customers are not equally persuaded by every number a seller places beside an offer.

Useful reference points include:

• A genuine previous selling price.
• The unit price where pack sizes differ.
• The total cost over the expected period of use.
• A clear comparison between packages with different inclusions.
• The separate price of bundle components when they are genuinely sold separately.
• The cost of the problem the product replaces, where that comparison is realistic.

This is also why charm pricing, such as £9.99 instead of £10, should not be elevated into a general strategy. The left-digit effect has experimental support under particular conditions, but price endings also communicate positioning. A tactic suited to a price-led retailer may be wrong for a premium or professional product.

The job of price presentation is to help the customer evaluate the exchange, not to make arithmetic harder.

Claims ask the customer to trust the seller. A sample, preview or demonstration allows them to form part of the judgement themselves.

For physical goods, direct experience might take the form of a sample, trial size, virtual try-on, detailed demonstration or try-before-you-buy arrangement. For digital products, it could be an interactive demo, a sample chapter, a limited software trial, a representative template or a preview of the actual learning environment.

The attraction of "free" is well documented. In a series of experiments, Kristina Shampanier, Nina Mazar and Dan Ariely found that a zero price could produce a disproportionately large increase in demand compared with an equivalent price reduction that did not reach zero.

That does not mean every business should introduce a free tier. The customer who starts a free trial is not necessarily the customer who will pay. Free access can attract people with little need for the product, reduce the urgency to evaluate it or give away enough value to replace the paid version. Research into software trials explicitly describes a trade-off between reducing uncertainty and cannibalising paid demand. A more recent large-scale randomised field experiment found that extending a trial increased adoption and delayed conversion but did not simply improve every stage of the customer journey.

A useful preview demonstrates the core benefit. A software trial that opens onto an empty dashboard may increase uncertainty. A course preview consisting only of the welcome video proves very little. Give customers enough substance to make a better decision, then measure what happens after the trial begins.

Scarcity can affect demand because limited availability changes the decision. If stock may genuinely run out or enrolment closes on a fixed date, waiting has a potential cost.

A meta-analysis of scarcity tactics in marketing found an overall positive effect on purchase intention, but the result varied according to the form of scarcity, product and context. That qualification matters. "Scarcity works" is not a licence to attach the same countdown timer to every offer.

Legitimate scarcity includes:

• A physical product with an accurate stock count.
• A limited production run that will not immediately be replenished.
• A service constrained by the provider's actual capacity.
• An event or course with a real enrolment deadline.
• A promotion with a published and enforced end date.

There is another risk: urgency shortens deliberation. That may increase immediate purchases while also increasing mistaken purchases, cancellations and returns. If a product requires careful assessment, giving customers less time to think can select for the wrong buyers.

When scarcity is real, explain what is limited, why it is limited and when the constraint ends. If there is no genuine constraint, the product should be able to stand without one.

Defaults are powerful because accepting a preselected option requires less effort than changing it. They can also imply that the seller recommends the option or that it represents the normal choice.

A meta-analysis of 58 default studies involving 73,675 participants found a considerable average effect. It also found substantial variation: several studies reported no significant effect and two found negative effects. Defaults influence decisions, but their effect is neither uniform nor automatically beneficial.

Online sellers can use this insight constructively. A recommended plan can reduce comparison effort when the business explains who it suits. A sensible default quantity can save repeat customers time. A preselected delivery method may be helpful when it is the least expensive reasonable option and alternatives remain visible.

The CMA's review of online choice architecture makes the broader point that the same design tools can help customers or harm them, depending on how they are used.

An effect observed across many studies can still fail on a particular website. The audience may be different, the product may carry more risk or the intervention may interact with other parts of the page in an unexpected way.

That does not make the research useless. It means research should generate better hypotheses rather than replace measurement.

Suppose a retailer adds a prominent scarcity message and conversion rises. That appears successful until returns and complaints also rise. A software company may increase free-trial registrations while reducing the proportion of users who ever reach the product's useful features. A subscription default may raise initial revenue while creating cancellations, chargebacks and distrust later.

Measure the outcome that represents genuine commercial value:

• Completed and paid orders rather than button clicks.
• Revenue and margin rather than conversion rate alone.
• Returns, refunds and chargebacks.
• Subscription cancellation and retention.
• Product activation and meaningful use.
• Customer-service contacts and complaint themes.
• Repeat purchases and customer lifetime value.

Where traffic and technical conditions allow, a properly designed randomised test is more reliable than comparing results before and after a change. A 2024 systematic review of A/B testing discusses the method across software engineering and the issues involved in conducting and interpreting experiments.

Small businesses may not always have enough conversions to reach a reliable result quickly. In that case, customer interviews, usability observation, search data, support questions and analytics can still identify serious problems. What they cannot do is prove that a fashionable tactic will increase sales by a predetermined percentage.

The principles above apply to both, but the emphasis changes.

This difference should shape the evidence. A photograph can prove the appearance of a chair, but not its comfort in every home. A case study can show that a course helped one student, but not guarantee the same result for everybody. Good marketing identifies what the evidence can establish, and where uncertainty honestly remains.

The most durable principles in online selling are not tricks.

These practices can improve conversion because they improve the decision itself. They also attract better-matched customers, set more accurate expectations and protect the trust on which future sales depend.

The alternative is to optimise for the moment of purchase while ignoring everything that follows it. That may produce an impressive test result. It is a poor way to build a business.