Conversion and UX psychology: the science of why users click or bounce

Every click is a small decision

Every click on your website is a decision. Every form field is a small piece of work you ask a stranger to do for you. Most of the rules that describe how people handle those decisions were worked out decades ago in psychology labs, long before anyone built a checkout: Hick on choice in 1952, Fitts on pointing in 1954, Miller on memory in 1956.

They are useful. They are also routinely oversold. Hick's law and Fitts's law describe reaction times in controlled experiments, measured in fractions of a second. They do not, on their own, predict your conversion rate. What they give you is a way of seeing your pages: where you make people choose too much, reach too far, remember too much or work too hard. The conversion effect you then find out by testing.

So for each idea below I give the research, what it does and does not say, and a worked example on an invented Irish business, every step shown. I have been applying these frameworks to digital products across eCommerce, SaaS and B2B for over a decade. Read them as a set of questions for a page review, not as laws that predict a lift. Every number in the examples is invented so I can show the arithmetic.

Hick's law: more options, slower decisions

In 1952 the British psychologist W. E. Hick published "On the rate of gain of information" in the Quarterly Journal of Experimental Psychology. It established one of the few widely accepted laws in psychology: choice reaction time grows with the number of alternatives, or more precisely with the amount of uncertainty in the task.¹ The common form is:

$RT = a + b \log_2 n$

Here $RT$ is the reaction time, the time to make the choice. $a$ is the part that has nothing to do with choosing (seeing the screen, moving a finger). $b$ is the extra time per bit of information, a constant you measure for a given task. $n$ is the number of equally likely options, and $\log_2 n$ is the number of bits needed to pick one of them. The logarithm is explained in Foundations: logarithms.

The logarithm is the important part. Going from two options to four adds the same time as going from four to eight, or eight to 16. Each doubling costs one more bit. So the first few options you add hurt less than you might fear, and the 30th option costs almost nothing on its own. What hurts is the habit of doubling.

Worked example: a menu at Annacotty Coffee Supply

Annacotty Coffee Supply is an invented coffee shop in Limerick selling beans online. Its main menu has 12 items. For the example, assume $a = 0.2$ seconds and $b = 0.15$ seconds per bit. These are the example's assumptions, chosen to be in a plausible range, not measured values.

With 12 items, $\log_2 12 = 3.585$:

$RT_{12} = 0.2 + 0.15 \times 3.585 = 0.738 \text{ seconds}$

With four items, $\log_2 4 = 2$:

$RT_{4} = 0.2 + 0.15 \times 2 = 0.5 \text{ seconds}$

Cutting the menu from 12 to four saves 0.238 seconds, which is $0.238 / 0.738 = 32\%$ of the decision time. The old version of this article said 60%. The formula does not support that.

The horizontal axis is the number of equally likely menu options, the vertical axis the predicted decision time in seconds, using the example's two constants. The takeaway is the shape: steep at the start, flat later. The absolute numbers are small. A quarter of a second is not why someone leaves Annacotty's site. Hick's law tells you where the friction sits, not how many orders it costs.

Where it matters, and where it does not

Hick's law bites where people must genuinely choose among options they do not know: a navigation menu on the first visit, a pricing page, a set of filters, a page with four competing calls to action. Choice overload, the bigger and messier effect of too many options on whether people choose at all, is covered in the behavioural economics article. The classic three tier pricing page is a related case.

The research also lists its own limits. Practice, a natural fit between what people see and what they do (stimulus response compatibility), very large sets of options and the order of earlier choices all change the result.¹ In plain terms:

• Familiar interfaces. Jakob Nielsen's observation, known as Jakob's law, is that users spend most of their time on other sites, so they prefer yours to work the way the sites they already know do.² A familiar layout with more items can beat an unfamiliar one with fewer.
• Search. Someone who knows what they want types it. A long country list with a search box is faster than a short list without one.
• Browsing. When the goal is discovery, not a decision, more to look at can be the point.

Fitts's law: bigger and closer is easier to hit

In 1954 Paul Fitts measured how long people took to move to a target and hit it, and found that the time depends on the distance and the size of the target.³ He defined a binary index of difficulty, $\log_2 \frac{2D}{W}$, and the movement time grows in a straight line with it:

$MT = a + b \log_2 \frac{2D}{W}$

$MT$ is the movement time. $a$ is the fixed time to start and stop, and $b$ the time per bit of difficulty, which depends on the person and the device. $D$ is the distance to the target and $W$ its width along the line of movement, so $\frac{2D}{W}$ grows as the target gets further away or smaller, and $\log_2 \frac{2D}{W}$ is the index of difficulty in bits. The logarithm is explained in Foundations: logarithms. Fitts chose $2D$ rather than $D$ so the index stays above zero in practical cases.³

Two consequences follow. Doubling the width of a button removes one bit of difficulty. Halving the distance removes the same bit. Big targets near where the eye and finger already are get hit faster and missed less.

Worked example: the basket button at Bennettsbridge Print Co

Bennettsbridge Print Co is an invented poster shop in Kilkenny. On desktop, its "Add to basket" button is 40 pixels wide and sits 400 pixels from where the cursor usually rests after choosing a size. For the example, assume $a = 0.1$ and $b = 0.1$ seconds per bit.

$MT_{\text{old}} = 0.1 + 0.1 \times \log_2 \frac{2 \times 400}{40} = 0.1 + 0.1 \times 4.32 = 0.532 \text{ seconds}$

The index of difficulty is $\log_2 20 = 4.32$ bits. The redesign makes the button 120 pixels wide and moves it to 150 pixels from the size selector:

$MT_{\text{new}} = 0.1 + 0.1 \times \log_2 \frac{2 \times 150}{120} = 0.1 + 0.1 \times 1.32 = 0.232 \text{ seconds}$

Now the index is $\log_2 2.5 = 1.32$ bits. Three bits less difficulty, 0.3 seconds faster. Again, the time itself is not the prize. The prize is fewer near misses on a phone, fewer taps on the wrong thing, and a primary action that is obviously the primary action.

Touch targets and the accessibility line

On touch screens the target is a fingertip, which is far less precise than a cursor. Apple's Human Interface Guidelines give a default minimum control size of 44 by 44 points on iPhone and iPad.⁴ The Web Content Accessibility Guidelines (WCAG) 2.1 set 44 by 44 CSS pixels for pointer targets at their strictest level, AAA, with exceptions such as links inside a sentence.⁵ I build to WCAG 2.1 AA as the baseline and treat 44 pixels as the sensible size for anything people tap often. In Ireland that baseline has a legal side as well: the European Accessibility Act, transposed by S.I. No. 636 of 2023, has applied to eCommerce services since 28 June 2025.⁶

Practical rules that follow from Fitts:

• The primary action is the biggest target. Secondary actions are smaller and further away.
• Destructive actions keep their distance. "Delete" should not sit next to "Save" at the same size.
• Full width buttons on mobile. On a narrow screen, full width is the largest $W$ you can give.
• Edges and corners. On a desktop, a cursor stops at the screen edge, so targets there are effectively very deep. That is why operating systems put menus at the top.

The peak end rule: people remember the peak and the ending

Donald Redelmeier and Daniel Kahneman recorded the pain patients felt in real time during colonoscopy (154 patients) and lithotripsy (133 patients) and then asked them afterwards how painful the whole procedure had been. The remembered total was strongly related to the worst moment and to the last three minutes. Lengthy procedures were not remembered as particularly bad.⁷ Barbara Fredrickson and Kahneman called the second finding duration neglect: in their film clip studies, people's overall judgements looked like a weighted average of snapshots of the experience, as if duration did not matter.⁸

The striking test came later. In a randomised trial with 682 colonoscopy patients, half had a short, gentler interval added to the end of the procedure. They rated the final moments as less painful (1.7 against 2.5 on a ten point scale) and the whole experience as less unpleasant (4.4 against 4.9), even though their procedure was longer.⁹ Adding a better ending made a longer experience remembered as a better one.

A simple way to write the idea down is the peak end average:

$R \approx \frac{x_{\text{peak}} + x_{\text{end}}}{2}$

$R$ is how the experience is remembered, $x_{\text{peak}}$ the most intense moment and $x_{\text{end}}$ the final moment, both on the same rating scale. It is an approximation of the research, not a law with fitted constants, and an average of two values, as in Foundations: the mean.

Worked example: two checkouts at Faithlegg Tea

Faithlegg Tea is an invented tea shop in Waterford. It has frustration ratings from 0 (none) to 10 (furious) for five steps of two checkout designs, from a small usability test.

Checkout A: browse 2, basket 3, delivery details 6, payment 2, confirmation 1. The mean is $14 / 5 = 2.8$. The peak is 6, the end is 1, so $R \approx (6 + 1) / 2 = 3.5$.

Checkout B: browse 1, basket 1, delivery details 2, payment 2, confirmation 5 (the confirmation page does not say when the tea will arrive). The mean is $11 / 5 = 2.2$. The peak and the end are both 5, so $R \approx (5 + 5) / 2 = 5$.

Checkout B is less frustrating on average and remembered as worse, because its worst moment is its last one. The fix is not on the payment page. It is on the confirmation page: what was ordered, when it arrives, what happens next, and a real person to contact.

What it means for a shop

The confirmation page, the confirmation email and the delivery are your ending. They are where a purchase turns into a memory, and that memory decides the second purchase. Low cost ways to get the ending right: a clear delivery date, order tracking that tells the truth, a short personal note in the parcel, and a confirmation page that answers the questions people actually have. Peaks can be positive too, a free sample or a quick, human reply to a question. Fix the worst moment first, though. A bad peak outweighs a nice touch.

The serial position effect: first and last are remembered best

People recall the first and last items in a list best and the middle items worst. The term goes back to Hermann Ebbinghaus and his memory experiments. The advantage at the start is called the primacy effect, the one at the end the recency effect.¹⁰ Plotted, recall makes a U shape across the list.

The commercial reading is modest and useful: the start and the end of any sequence are where attention and memory are highest. Put your strongest point first in a feature list and your second strongest last. Open an email with the main message and end it with the one action. In a long navigation, the first and last items carry the most weight.

There is one exception worth knowing. With three pricing tiers, a recommended middle option often benefits from the compromise effect, people's tendency to choose the middle of a range, which works against serial position. The pricing psychology article covers it.

Worked example: reordering a feature list at Shanakiel Rugs

Shanakiel Rugs is an invented rug shop in Cork. Its product page lists seven points in the order they were written: "Hand knotted", "Wool from Donegal", "Free returns for 30 days", "Underlay included", "Stain resistant finish", "Made in Ireland", "Free delivery". Customer emails show the two questions people ask most are about cleaning and about delivery.

The reorder puts "Stain resistant finish" first and "Free delivery" last, the two strongest answers in the two positions people remember best. The middle five stay, in any order. The test is simple: same page, two orders, and the metric is add to basket rate. If the order makes no measurable difference, keep whichever reads better.

Miller's law: working memory is small

George Miller's 1956 paper "The magical number seven, plus or minus two" summarised experiments on how many things people can hold and judge at once, and found a limit around seven for simple, one dimensional judgements.¹¹ Later work by Nelson Cowan argued that Miller's seven was more a rough estimate than a real capacity, and that when people cannot group items together the limit is closer to four chunks.¹²

The practical lesson is not "never more than seven menu items". A menu stays visible, so nobody has to remember it. The lesson is about what people must hold in their heads while doing something else: a code from an email, a set of plan features to compare, the fields still to fill in. Keep that load small, and group items into chunks people can treat as one.

Worked example: chunking an order reference

The invented Bennettsbridge Print Co sends order references like KH748291156. Customers type it into a returns form and get it wrong often enough for the support inbox to notice. The same reference shown as KH7 4829 1156 is three chunks instead of 11 separate characters. The characters are identical. The load on memory is not. The same works for forms: Bennettsbridge splits a 12 field returns form into three short groups (your order, the problem, how you would like it sorted), each fitting comfortably within four or five items.

Cognitive load: every extra piece of work costs attention

John Sweller's work in the late 1980s, which became cognitive load theory, argued that working memory is limited and that tasks which make heavy demands on it leave little room for anything else. His starting point was learning: problem solving that demands a lot of working memory gets in the way of learning from it.¹³ Translated to a website: some effort comes with the task itself (buying a sofa is harder than buying a book), and some is added by the design (confusing labels, a field nobody needs, an error message that does not say what is wrong). You cannot remove the first. The second is yours.

What Baymard's checkout research shows

The Baymard Institute has tested checkout usability for years. Its numbers are the most useful benchmarks I know for this:

• The average documented cart abandonment rate is 70.22%, the unweighted average of 50 studies.¹⁴
• Among shoppers who abandoned an order, leaving aside those who were only browsing, 18% left because the site wanted them to create an account and 17% because the checkout was too long or complicated. The top reason was extra costs, at 40%, and 19% did not trust the site with their card details.¹⁴
• In its 2024 benchmark the average checkout had 11.3 form fields, while most sites need only eight.¹⁵
• Baymard estimates that the average large eCommerce site could gain a 35.26% increase in conversion rate through better checkout design.¹⁴

Two corrections to the old version of this article. It said 28% of abandoners cited a long checkout, and elsewhere 18%. Baymard's figure is 17%. It also claimed that guest checkout lifts conversion by 35% for first time buyers. The 35.26% figure is Baymard's estimate for better checkout design as a whole, not for guest checkout. A claim that cutting fields from 15 to eight lifts completion by 10% to 15% had no source I could find, so it is gone. The trust side of the checkout, and that 19%, is covered in the social proof and trust article.

Worked example: trimming the checkout at Moycullen Candles

Moycullen Candles is an invented candle shop in Galway. Its checkout has 14 fields. Going through them one by one:

1. Email, first name, last name, phone, address line 1, address line 2, town, county, Eircode, country, card number, name on card, expiry, security code.
2. Combine first and last name into one "Full name" field. 13 fields.
3. Drop county (the Eircode does the job) and hide address line 2 behind an "Add another line" link. 11 visible.
4. Default country to Ireland, changeable. Still a field, no longer work.
5. Make phone optional, with a note on why it helps the courier. Still visible, no longer required.
6. Use address lookup from the Eircode, so most customers type an Eircode and pick their address.
7. Let the payment provider's field handle card number, expiry and security code as one compact group.

The result is eight to nine fields most customers actually touch, in three groups: you, delivery, payment. None of this needed a theory of conversion. It needed someone to ask, for each field, who uses it and what happens if it goes.

Load time is the first friction

Before anyone chooses, reaches or remembers anything, the page has to appear and respond. Google's Core Web Vitals define three measures of that experience with thresholds for a good result: Largest Contentful Paint (how fast the main content loads) within 2.5 seconds, Interaction to Next Paint (how quickly the page responds to input) of 200 milliseconds or less, and Cumulative Layout Shift (how much the layout jumps around) of 0.1 or less, each measured at the 75th percentile of visits.¹⁶

You will often see figures claiming that each extra second of load time raises bounce rate by a precise percentage. The widely quoted Google page I would have cited no longer exists, so I leave the number out. The thresholds above are the standard that matters, and your own analytics will tell you how speed relates to bounce on your own pages. I go deeper in the Core Web Vitals article.

An example: an invented Limerick furniture shop finds its product pages have a Largest Contentful Paint of 4.1 seconds at the 75th percentile, caused by a 3MB hero photo. Serving a properly sized, compressed image brings it to 2.2 seconds, inside the threshold. The useful comparison afterwards is product page bounce rate in the four weeks before and after, with the same campaigns running.

Friction works in both directions

Friction is not always bad. The principle is asymmetry: remove friction from the paths you want people to take, and keep sensible friction on paths where a mistake is costly.

The best known example of removing it is Amazon's one click ordering. The patent, "Method and system for placing a purchase order via a communications network", was filed in September 1997, granted in September 1999 and expired in September 2017.¹⁷ For returning customers it removed almost every step between deciding and buying. The old version of this article said the patent was "worth billions". I could not find a source for a figure, so I will only say that a large retailer thought the step was worth patenting.

When to add friction, and the legal line

Friction belongs where an action is hard to undo: deleting an account, deleting data, a large order, changing payment details. A confirmation step, a summary or an undo window protects the customer and cuts support work.

It does not belong on the way out. Making cancellation deliberately hard is a dark pattern, and the law is moving against it. In Ireland, section 115A of the Consumer Rights Act 2022, inserted by S.I. No. 309 of 2026, requires a trader whose customer concluded a contract online to offer an online cancellation function: labelled "cancel contract here" or something as clear, prominently displayed, available throughout the cancellation period, and followed by a "confirm cancellation" step and an acknowledgement. It covers the legal cancellation period rather than every later exit, but the direction is plain.¹⁸ Account deletion carries its own duty: under Article 17 of the GDPR, people have the right to have their data erased without undue delay where one of its grounds applies, as the Data Protection Commission (DPC) explains,¹⁹ and Article 12(3) gives the business one month to respond to the request.²⁰ A "type DELETE to confirm" box is fine. A maze is not. The dark patterns article has more on where design becomes a legal problem.

Worked example: a fair cancellation flow at Rochestown Coffee Club

Rochestown Coffee Club is an invented coffee subscription in Cork. Its cancellation flow has three screens:

1. "Before you go: would a pause help?" with options to pause for one, two or three months, and a clear "No, cancel my subscription" button of the same size.
2. One optional question, "What is the main reason?", with five choices and a skip link.
3. "Your subscription is cancelled. Your last delivery is on 14 May." and a confirmation email.

The pause offer answers the most common reason people leave a coffee subscription, having too much coffee, and it is one tap away from cancelling. Nobody has to call, nobody has to find a hidden link. The question gives the shop the data it needs to improve. That is friction used honestly.

Attribute framing: the same fact, described differently

Logically identical information leads to different judgements depending on how it is described. In Irwin Levin and Gary Gaeth's 1988 study, consumers rated ground beef labelled "75% lean" more favourably than the same beef labelled "25% fat".²¹ The old version of this article said "95% lean" and gave preference shares for this and three other "studies". The study used 75%, and I could not find sources for the other examples or their numbers, so they are gone. The same study found the effect was smaller once people had actually tasted the meat, which is a useful warning: framing moves first impressions more than it moves experience.

Some framing choices that follow from the idea, each one a copy test rather than a law:

Instead of	Try	Why
Delivery in 3 to 5 working days	Arrives by Friday	A date is easier to picture
Save 15%	Save €180 a year	A large absolute number can feel bigger than a small percentage
Unlimited storage	You will not run out of space	The benefit, not the feature
30 day refund policy	Not right? Send it back within 30 days	Plain words, not policy words

Framing must stay true. "Only 3 left at this price" is fine when it is true and a misleading action when it is not. The scarcity and urgency article covers that line.

Worked example: testing a delivery line at Faithlegg Tea

Faithlegg Tea tests "Delivery in 2 to 3 working days" against "Order today, arrives by Thursday" on its product pages. Each version gets 8,000 visitors. The control converts at 4.0% (320 orders), the new line at 4.6% (368 orders).

The pooled conversion rate is $p = (320 + 368) / 16{,}000 = 0.043$. The two proportion z test from Foundations: hypothesis tests is:

$z = \frac{p_1 - p_2}{\sqrt{p(1-p)\left(\frac{1}{n_1} + \frac{1}{n_2}\right)}}$

where $p_1$ and $p_2$ are the two conversion rates, $p$ the pooled rate and $n_1$ and $n_2$ the visitors in each version. With the numbers:

$z = \frac{0.046 - 0.040}{\sqrt{0.043 \times 0.957 \times \left(\frac{1}{8{,}000} + \frac{1}{8{,}000}\right)}} = \frac{0.006}{0.00321} = 1.87$

A $z$ of 1.87 gives a two sided p value of about 0.06. That is not significant at the usual 5% level. A 15% relative lift looked convincing and is not yet proven. Faithlegg should keep the test running to a sample size set in advance, as Foundations: sample size explains, rather than declare victory. The statistical significance article goes further.

A small toolkit in Python

The old version of this article had more than a dozen Python scripts that turned these laws into predicted conversion lifts. The coefficients in them were made up, so the predictions were too. Here is the honest version: the formulas above, as functions, with the example numbers. Use them to compare designs, never to forecast revenue.

import math

def hick_time(n_options, a=0.2, b=0.15):
    """Hick's law decision time in seconds (a and b are assumptions)."""
    return a + b * math.log2(n_options)

def fitts_time(distance, width, a=0.1, b=0.1):
    """Fitts's law movement time in seconds, index of difficulty log2(2D/W)."""
    return a + b * math.log2(2 * distance / width)

def peak_end(ratings):
    """Peak end average of a list of ratings (higher = more intense)."""
    return (max(ratings) + ratings[-1]) / 2

def two_proportion_z(x1, n1, x2, n2):
    """z statistic and two sided p value for two conversion rates."""
    p1, p2 = x1 / n1, x2 / n2
    p = (x1 + x2) / (n1 + n2)
    se = math.sqrt(p * (1 - p) * (1 / n1 + 1 / n2))
    z = (p2 - p1) / se
    p_value = math.erfc(abs(z) / math.sqrt(2))
    return z, p_value

print(round(hick_time(12), 3), round(hick_time(4), 3))   # 0.738 0.5
print(round(fitts_time(400, 40), 3), round(fitts_time(150, 120), 3))  # 0.532 0.232
print(peak_end([2, 3, 6, 2, 1]), peak_end([1, 1, 2, 2, 5]))  # 3.5 5.0
print(two_proportion_z(320, 8000, 368, 8000))  # about (1.87, 0.061)

A page review in seven questions

Put together, the ideas become a set of questions to ask of any page, in roughly the order they tend to matter:

Read the flow from the top. Each step is one question, and each points to the idea behind it in brackets. The last step is the one that turns a review into evidence: every change the questions suggest is tested against the current version before anyone calls it a win.

Psychology gives the questions, testing gives the answers

These ideas are real and well researched, and none of them is a conversion rate in disguise. Hick and Fitts describe fractions of a second. Miller's seven is closer to four. The peak end rule came from hospital procedures, and framing effects shrink once people try the product. What they all do well is point at the places where a page asks too much of the person using it.

So use them to find the problem, fix it in the plainest way you can, and measure whether it made a difference. Fewer choices where people must choose, bigger targets where people must tap, less to remember, fewer fields, a good ending and words that describe the truth kindly. That is most of good user experience (UX) design, and most of it costs very little to try.

Footnotes

1. Robert W. Proctor and Darryl W. Schneider, Hick's law for choice reaction time: a review, Quarterly Journal of Experimental Psychology 71(6), 1281 to 1299, 2018, https://doi.org/10.1080/17470218.2017.1322622

2. Jakob Nielsen, Jakob's Law of Internet User Experience, Nielsen Norman Group, 2017, https://www.nngroup.com/videos/jakobs-law-internet-ux/

3. Paul M. Fitts, The information capacity of the human motor system in controlling the amplitude of movement, Journal of Experimental Psychology 47(6), 381 to 391, 1954, reprinted in Journal of Experimental Psychology: General 121(3), 1992, https://www.cs.princeton.edu/courses/archive/fall08/cos436/FittsJEP1954.pdf

4. Apple, Human Interface Guidelines: Accessibility (default and minimum control sizes), https://developer.apple.com/design/human-interface-guidelines/accessibility

5. W3C, Understanding Success Criterion 2.5.5: Target Size (Level AAA), WCAG 2.1, https://www.w3.org/WAI/WCAG21/Understanding/target-size.html

6. Government of Ireland, S.I. No. 636 of 2023, European Union (Accessibility Requirements of Products and Services) Regulations 2023, regulations 1(2) and 3(2)(g), Irish Statute Book, 2023, https://www.irishstatutebook.ie/eli/2023/si/636/made/en/html

7. Donald A. Redelmeier and Daniel Kahneman, Patients' memories of painful medical treatments: real time and retrospective evaluations of two minimally invasive procedures, Pain 66(1), 3 to 8, 1996, https://doi.org/10.1016/0304-3959%2896%2902994-6

8. Barbara L. Fredrickson and Daniel Kahneman, Duration neglect in retrospective evaluations of affective episodes, Journal of Personality and Social Psychology 65(1), 45 to 55, 1993, https://doi.org/10.1037/0022-3514.65.1.45

9. Donald A. Redelmeier, Joel Katz and Daniel Kahneman, Memories of colonoscopy: a randomized trial, Pain 104(1 to 2), 187 to 194, 2003, https://doi.org/10.1016/S0304-3959%2803%2900003-4

10. Wikipedia, Serial position effect (definition, citing Ebbinghaus, On memory, 1913 translation), https://en.wikipedia.org/wiki/Serial-position_effect

11. George A. Miller, The magical number seven, plus or minus two: some limits on our capacity for processing information, Psychological Review 63(2), 81 to 97, 1956, Classics in the History of Psychology, York University, https://psychclassics.yorku.ca/Miller/

12. Nelson Cowan, The magical number 4 in short term memory: a reconsideration of mental storage capacity, Behavioral and Brain Sciences 24(1), 87 to 114, 2001, https://doi.org/10.1017/S0140525X01003922

13. John Sweller, Cognitive load during problem solving: effects on learning, Cognitive Science 12(2), 257 to 285, 1988, Crossref record with abstract, https://api.crossref.org/works/10.1207/s15516709cog1202_4

14. Baymard Institute, Cart abandonment rate: 50 statistics and why people abandon, Baymard Institute, 2026, https://baymard.com/lists/cart-abandonment-rate

15. Edward Scott, 5 ways to minimize form fields in checkout, Baymard Institute, 2024, https://baymard.com/research-articles/checkout-flow-average-form-fields

16. Google, Web Vitals, web.dev, https://web.dev/articles/vitals

17. Amazon.com Inc, US patent 5,960,411, Method and system for placing a purchase order via a communications network, Google Patents, https://patents.google.com/patent/US5960411A/en

18. Government of Ireland, S.I. No. 309 of 2026, European Union (Distance Contracts for Financial Services) Regulations 2026, regulation 6 (new section 115A of the Consumer Rights Act 2022, online cancellation function), Irish Statute Book, 2026, https://www.irishstatutebook.ie/eli/2026/si/309/made/en/html

19. Data Protection Commission, The right to erasure (Articles 17 and 19 of the GDPR), accessed November 2026, https://www.dataprotection.ie/en/individuals/know-your-rights/right-erasure-articles-17-19-gdpr

20. Regulation (EU) 2016/679 (General Data Protection Regulation), Article 12(3), EUR-Lex, https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32016R0679

21. Irwin P. Levin and Gary J. Gaeth, How consumers are affected by the framing of attribute information before and after consuming the product, Journal of Consumer Research 15(3), 374 to 378, 1988, https://academic.oup.com/jcr/article/15/3/374/1821914