Uber Eats — UX Heuristic Evaluation Report

Task 1 · 5 steps

Search for "Sandwich" and add an item to cart

Happy path · Uber Eats mobile (iOS)

Open the Uber Eats app

Screen: Home feed

PURE 1 — Low friction Cognitive load 1 / 5

Friction points & heuristic issues

Home screen launches with clear hierarchy — address, category chips, and featured restaurants are immediately visible. No heuristic violations. The fee disclaimer banner introduces a cautionary tone but does not block action.

Cognitive load

User simply opens the app. No decision-making required. Familiar food-delivery layout minimises the learning curve.

Error prevention

Minimal risk

Tap search bar and type "sandwich"

Screen: Search input + autocomplete

PURE 2 — Moderate friction Cognitive load 2 / 5

Friction points & heuristic issues

Search lives in the bottom navigation bar, not at the top where most food-delivery apps position it. A "Sandwich" category chip is already visible on the home feed — users may tap that instead of the search bar, diverging from the happy path. The keyboard displays Chinese labels ("搜索", "空格"), a localization inconsistency that creates minor visual noise.

H4 — Consistency & standards H8 — Aesthetic & minimalist design Localization conflict

Cognitive load

Dual affordances (category chip vs. search bar) create a minor decision point before the user can proceed.

Error prevention

Medium — wrong-path risk via category chip

Select a restaurant from results

Screen: Search results list

PURE 2 — Moderate friction Cognitive load 3 / 5

Friction points & heuristic issues

Inline item previews (Clubhouse CA$16.20, Druxy's Famous Reuben CA$20.40) appear before the user selects a restaurant, blurring the list/menu boundary. The prominent "Buy 1, get 1 free" badge nudges users toward a promotional decision before they've chosen a restaurant. McDonald's carries a "Most popular" badge despite a lower rating (4.2) than Williams Fresh Cafe (4.5), creating a trust inconsistency.

H1 — Visibility of system status H7 — Flexibility & efficiency Choice overload Promotional dark pattern

Cognitive load

User must scan ratings, delivery fees, ETAs, and promo badges simultaneously while deciding whether to engage with inline item cards or the restaurant row itself.

Error prevention

High — inline item tap bypasses restaurant menu selection

Browse menu and tap a sandwich item

Screen: Restaurant page (bottom sheet) + item detail

PURE 2 — Moderate friction Cognitive load 3 / 5

Friction points & heuristic issues

The restaurant opens as a bottom sheet modal, placing fee info and the Delivery/Pickup toggle above the menu — the primary goal is buried. The BOGO section is the first content block. Item descriptions are truncated with no "read more" affordance — a concern for users with dietary restrictions. Customisation options use internal shorthand codes ("BGL Everything", "CHS Cheddar 2 slices") violating real-world language expectations.

H2 — Match: system & real world H8 — Aesthetic & minimalist design Truncated descriptions — allergen risk Abbreviated option labels

Cognitive load

Layered decisions: promo mechanics, service info, truncated descriptions, and cryptic customisation labels compound sequentially.

Error prevention

Medium — wrong bagel variant selected due to label shorthand

Tap "Add to cart"

Screen: Item detail CTA + cart view

PURE 2 — Moderate friction Cognitive load 2 / 5

Friction points & heuristic issues

The CTA reads "Add 2 · $9.00 $18.00" — quantity 2 is pre-set by the BOGO promotion without explicit user consent or inline explanation. The dual-price display in one compact button is visually dense. "Saving $9 with promotions" only appears after adding — not before — removing pre-commitment transparency. An upsell carousel in the cart introduces distraction at the point of commitment.

H1 — Visibility of system status H5 — Error prevention Pre-set quantity without consent Post-add saving disclosure

Cognitive load

Core action is simple, but surrounding quantity and pricing information requires extra parsing before the user can commit confidently.

Error prevention

High — quantity defaults to 2 without clear pre-action warning

Task 1 — Overall assessment

Overall cognitive load

2.2 / 5

Avg across 5 steps

Steps at PURE 2+

4 / 5

Moderate friction dominant

Heuristic violations

H1, H2, H4, H5, H7, H8

Friction trajectory

Step 1 — Open app

Step 2 — Search

Step 3 — Results

Step 4 — Menu

Step 5 — Add cart

Primary friction points (ranked)

Pre-set quantity of 2 on "Add to cart" CTA — users do not consent explicitly to the BOGO pair before committing.

Inline item cards on the results list blur the list/menu boundary and create an accidental direct-add shortcut that bypasses restaurant review.

Abbreviated customisation labels ("BGL Everything", "CHS Cheddar") violate real-world language expectations and elevate error risk at ordering.

Search bar placement in bottom nav deviates from platform standards — the home screen Sandwich chip creates a competing, off-path affordance.

Promotional mechanics (BOGO, upsell carousel) surface at every decision point, consistently prioritising conversion over task clarity.

Task 2 · 7 steps

Change payment method, set a custom tip, and verify the final total

Happy path · Uber Eats mobile (iOS)

Tap "View cart" to navigate to cart page

Screen: Cart summary (Williams Fresh Cafe)

PURE 1 — Low friction Cognitive load 1 / 5

Friction points & heuristic issues

The cart screen clearly displays the selected item, quantity stepper, promotion badge, and "Go to checkout" CTA. The "Offers for you" upsell carousel is visible but does not block the primary path. "Saving $9 with promotions" banner provides a positive transparency signal.

No violations Upsell carousel adds visual noise

Cognitive load

The user is returning to a state they already created. Recognising their item and tapping the clear CTA requires minimal effort.

Error prevention

Minimal risk

Tap "Go to checkout"

Screen: Cart → Checkout transition

PURE 1 — Low friction Cognitive load 1 / 5

Friction points & heuristic issues

"Go to checkout" is a prominent full-width black CTA pinned to the bottom of the screen. Label is unambiguous and action-oriented. No heuristic issues at this step.

Clear, standard CTA pattern

Cognitive load

Single, obvious action. No competing affordances at the bottom of screen.

Error prevention

Minimal risk

Tap "No thanks" to dismiss upsell

Screen: "Complete your order" interstitial

PURE 2 — Moderate friction Cognitive load 2 / 5

Friction points & heuristic issues

An interstitial upsell screen intercepts the checkout flow with 6+ add-on items. The dismiss action "No thanks" is a socially loaded phrase that introduces mild guilt-inducing friction (confirmshaming). The screen does not display the user's current order total, removing contextual grounding right before payment. No progress indicator shows how many checkout steps remain.

H1 — Visibility of system status H3 — User control & freedom Confirmshaming — "No thanks" Mandatory interruption pre-payment No order total shown

Cognitive load

User must scan 6 upsell items, resist social pressure, and re-orient to their goal of changing payment method.

Error prevention

Medium — user may accidentally add an upsell item

Tap currently selected payment method

Screen: Checkout summary → Pay with modal

PURE 2 — Moderate friction Cognitive load 2 / 5

Friction points & heuristic issues

The checkout screen shows "Apple Pay" as a summary row with a right chevron, suggesting it is tappable. However, this interaction — tapping a list row to open a full payment modal — is not clearly signalled. Users accustomed to a dedicated "Change" or "Edit" button may overlook it. A partially cut-off courier upsell banner at the top competes with the order summary.

H6 — Recognition over recall H4 — Consistency & standards No "Edit payment" label — low discoverability Courier upsell occupies top zone

Cognitive load

User must infer the payment row is tappable to edit rather than being given a clear affordance.

Error prevention

Medium — users may tap "Next" without realising payment can be changed

Select a different payment method

Screen: "Pay with" modal

PURE 2 — Moderate friction Cognitive load 3 / 5

Friction points & heuristic issues

The "Pay with" modal lists two identical entries labelled "Apple Pay" — no distinguishing detail (last 4 digits, card type, default badge). The Personal/Business toggle adds unexpected complexity. "Uber Cash: $0.00" draws attention to an empty wallet. No confirmation feedback is provided when a method is selected.

H6 — Recognition over recall H1 — Visibility of system status Duplicate "Apple Pay" entries Personal / Business toggle adds unexpected complexity No selection confirmation feedback

Cognitive load

User must interpret duplicate labels, decide between Personal/Business, assess Uber Cash balance, and confirm selection — all in one unguided modal.

Error prevention

High — duplicate labels make correct selection ambiguous

Tap "Next" to proceed to tip page

Screen: Checkout summary → Add a tip

PURE 1 — Low friction Cognitive load 1 / 5

Friction points & heuristic issues

"Next" is a clear full-width CTA. The checkout summary provides a well-structured cost breakdown. Minor: the CTA label "Next" is generic — "Proceed to tip" would better set expectations.

Clear price breakdown — good transparency Generic CTA label — no expectation setting

Cognitive load

The cost breakdown is clear and well-structured. The user simply reads the total and taps forward.

Error prevention

Minimal risk

Tap "Edit", enter custom tip, verify final price Key step

Screen: "Add a tip" → custom amount input → final total

PURE 3 — High friction Cognitive load 4 / 5

Friction points & heuristic issues

The tip screen uses emotionally loaded labels — "Cheers to you (15%)", "You're great (18%)", "You're my hero (25%)" — nudging users toward higher tips with positive identity language. The 18% option is pre-selected, exploiting anchoring bias.

System status visibility failure — special focus

When the user taps "Edit" and enters a custom amount (e.g. $4.00), the "Order total" on the input screen reads $17.51 — the food + fees subtotal only, excluding the tip. The all-in grand total (food + fees + tip) is never shown. After tapping "Save," only "Your tip: $4.00" appears — with no running grand total. The user must mentally add $17.51 + $4.00, with no app confirmation. This directly breaks the core task goal of verifying the total price — a serious violation of H1.

H1 — Visibility of system status H5 — Error prevention H2 — Match: system & real world Pre-selected 18% tip — anchoring bias Emotionally loaded preset labels Custom input shows subtotal, not grand total No all-in total after tip confirmation

Cognitive load

User must resist pre-selected nudges, interpret loaded labels, and manually perform mental arithmetic to determine true final charge — without any UI confirmation.

Error prevention

High — user cannot verify the final charge before tapping "Place order"

Task 2 — Overall assessment

Overall cognitive load

2.0 / 5

Avg across 7 steps

Steps at PURE 2+

5 / 7

1 step reaches PURE 3

Heuristic violations

H1, H2, H3, H4, H5, H6 affected

Friction trajectory

Step 1 — View cart

Step 2 — Checkout

Step 3 — Dismiss upsell

Step 4 — Tap payment

Step 5 — Select method

Step 6 — Tap Next

Step 7 — Custom tip

Primary friction points (ranked)

Custom tip input shows subtotal only ($17.51), not the all-in grand total — the core task goal is architecturally impossible to complete with confidence.

Pre-selected 18% tip with emotionally loaded label hierarchy constitutes social engineering at a financially consequential step.

Mandatory upsell interstitial with "No thanks" confirmshaming violates user control and adds cognitive cost before payment.

Duplicate "Apple Pay" entries in the payment modal with no differentiating metadata make correct selection ambiguous.

No payment edit affordance label on checkout screen — users must infer the row is tappable, reducing discoverability.

Task 3 · 5 steps

Find customer service and locate a support phone number

Happy path · Uber Eats mobile (iOS)

Locate and tap the Account icon on the Home screen

Screen: Home feed — bottom navigation bar

PURE 1 — Low friction Cognitive load 1 / 5

Friction points & heuristic issues

The Home feed is the starting point. The bottom navigation bar contains 5 icons: Home, Account (person silhouette), Search, Cart, and a second person-shaped icon. The Account icon's label is not persistently visible — icons rely on shape recognition alone with no text labels below them. For a distressed user in a hurry, the distinction between the two person-shaped icons may cause brief hesitation. Critically, there is no "Help" or "Support" shortcut anywhere on the Home screen — the user must first identify the Account icon as the indirect gateway to support.

Navigation entry point

A user seeking customer service must correctly identify the Account icon as the path to Help — an indirect route that is not signposted from the Home feed. No shortcut or contextual help entry point exists on the most-visited screen in the app.

H6 — Recognition over recall Bottom nav is a standard, familiar pattern No persistent icon labels — recognition relies on shape alone No direct "Support" shortcut from Home screen

Cognitive load

Tapping a bottom nav icon is a well-established mobile pattern requiring minimal effort. The absence of persistent text labels is a minor overhead, but the action is quick and familiar for most smartphone users.

Error prevention

Low — user may tap the wrong person icon, but recovery is immediate with one back-tap

Locate "Help" on the Account profile page and tap it

Screen: Account profile page → Help hub

PURE 2 — Moderate friction Cognitive load 2 / 5

Friction points & heuristic issues

The Account profile page loads a flat, heterogeneous list of 9+ items with no visual grouping: quick-action shortcuts (Favourites, Wallet, Orders), a "Try Uber One free" promotional banner, and a long feature list (Family and teens, Rides, Promotions, Send a gift, Help, Saved groups, Set up your business profile, Partner Rewards). "Help" is the 5th item in this list, visually equal to commercial and non-support entries. The Uber One promotional banner sits above Help, further pushing it down the hierarchy. There is no dedicated support section, no icon differentiation, and no priority signal to guide a distressed user directly to Help.

Discoverability concern

Most users in distress expect a dedicated "Support" or "Contact Us" path — not a generic profile page where Help competes equally with "Rides," "Promotions," and "Partner Rewards." The promotional banner occupying the space above Help actively deprioritises the support entry point.

H4 — Consistency & standards H7 — Flexibility & efficiency Help buried at item 5 of 9 Promotional banner above Help reduces visual priority No visual grouping or hierarchy separating support from commercial items

Cognitive load

The user must scan a flat list of 9+ mixed-purpose items to locate "Help." The presence of commercial items (Uber One, Partner Rewards) alongside support items increases visual noise for a user already stressed about an order problem.

Error prevention

Medium — user may tap "Orders" instead of "Help," expecting support to live within their order history

Navigate the Help hub and tap the last order

Screen: Help hub — Restaurants / Stores tabs + Last Order card

PURE 2 — Moderate friction Cognitive load 3 / 5

Friction points & heuristic issues

The Help hub opens with an "Ongoing chat with Uber Support" banner at the top — implying an active session may already exist, which is confusing for a user who has not recently chatted. The "Restaurants" and "Stores" tabs require the user to classify their issue type before proceeding — an unnecessary pre-screening step. The "Last Order" card (Domino's Pizza, Apr 30, CA$31.71) is surfaced directly and is a genuinely strong UX shortcut. However, there is no visible "Talk to a person," "Call us," or "Live chat" option anywhere on this screen — human escalation is architecturally invisible at the hub level.

Human agent visibility

A user scanning this screen for a phone number or live chat button will find nothing. The only forward paths are the AI chatbot or topic category lists. Human escalation is not offered as an option at any visible point on this screen.

H1 — Visibility of system status H6 — Recognition over recall Last Order card surfaced directly — strong UX shortcut No human agent entry point visible anywhere on screen Ongoing chat banner creates false session confusion Restaurants/Stores tab forces pre-classification of issue

Cognitive load

Three micro-decisions before any support interaction begins: classify context (Restaurants vs. Stores), assess the existing chat state, identify the correct order or topic to tap. The Last Order shortcut reduces one decision, but the banner and tab overhead remain.

Error prevention

High — user selecting wrong tab (e.g. Stores for a restaurant order) receives irrelevant AI routing with no clear way back

Select an issue category — e.g. "Another order issue"

Screen: Order issue category list → AI chatbot (topic-routed)

PURE 2 — Moderate friction Cognitive load 3 / 5

Friction points & heuristic issues

Selecting a category opens the AI chatbot presenting near-duplicate issue chips — creating a disorienting sense of circular navigation. The chatbot header reads "Powered by AI" in 12px subdued text: the only indicator this is not a human. No free-text input box, no "Skip to agent" option, and no phone number exists — all paths are gated behind chip selection.

Chatbot containment architecture

The system is designed to maximise automation and minimise human agent contact. Every available path routes through the AI decision tree. This is a deliberate support-suppression strategy, not an oversight.

H1 — Visibility of system status H3 — User control & freedom H10 — Help & documentation Circular category duplication — same chips as previous screen "Powered by AI" disclosure buried in 12px subtext No free-text input — human bypass impossible

Cognitive load

User must recognise the duplicate menu, understand they are inside an AI system from minimal signposting, and begin processing that direct human contact may not be possible — all while managing frustration over a real order issue.

Error prevention

High — user may believe they are in a human chat session due to inadequate AI disclosure

Attempt to reach a human agent or find a phone number Critical failure point

Screen: Chatbot — "More options" branch / free-text attempt

PURE 3 — High friction Cognitive load 5 / 5

Friction points & heuristic issues — deep analysis

This step represents a complete architectural dead end. The task goal — reaching a human agent or finding a customer service phone number — is structurally unachievable within the app.

No phone number exists anywhere in the app

Uber Eats does not publish a customer service phone number in-app. A user whose primary goal is direct phone contact will exhaust every screen in this flow, find nothing, and be forced to leave the app entirely. This is a deliberate cost-reduction strategy — the app funnels 100% of support contacts through the chatbot.

No free-text input — "human" bypass is impossible

The chatbot presents only pre-defined chip options. The happy path step of typing "human" cannot be executed because no text box exists at this stage. Every chip routes deeper into automation, never toward a human agent.

"More options" is a dead-end branch

Tapping "More options" surfaces additional issue categories — not a "Talk to a person," "Call us," or "Request a callback" option. The resolution tree is finite and loops back to the same automation with no guaranteed human escalation path.

H1 — Visibility of system status H3 — User control & freedom H5 — Error prevention H9 — Help users recognise & recover H10 — Help & documentation Phone number: architecturally absent Human agent: no escalation path Free-text input: unavailable Exit path: user must leave app

Emotional design failure

The support flow uses a cold, transactional chip-only interface with no empathy signalling — no acknowledgement of inconvenience, no "we're sorry this happened," no estimated resolution time. The emotional register is entirely mismatched to a user's distressed state.

Error prevention

Maximum — goal is architecturally impossible. No system feedback informs the user that this path does not exist

Task 3 — Overall assessment

Overall cognitive load

3.0 / 5

Highest avg across all tasks

Steps at PURE 2+

4 / 5

1 step reaches PURE 3

Heuristic violations

H1, H3, H4, H5, H6, H7, H9, H10

Frustration trajectory

User emotional state as path narrows toward a dead end

Mild annoyance Maximum frustration

Friction trajectory

Step 1 — Find Account icon

Step 2 — Account page / Help

Step 3 — Help hub / Last order

Step 4 — Issue category

Step 5 — Human / phone

Primary friction points (ranked)

No phone number exists anywhere in-app — the stated task goal is architecturally impossible, with no system feedback informing the user of this fact.

Chatbot-first, human-suppressed architecture — every support path routes through automation. Escalation to a human is only possible after navigating the full decision tree, and is not guaranteed.

Circular topic duplication — same issue categories appear in the chatbot and the preceding issue list screen, creating disorienting circular navigation.

Help buried in profile menu at item 5 of 9 — no contextual link from order history or active order screens.

Emotionally mismatched UI — no empathy language, no apology, no estimated resolution time in the support flow, despite the user entering in a distressed state.

Emotional design failures

No empathy signalling

Neutral, transactional chip UI with no acknowledgement that something went wrong.

AI label buried

"Powered by AI" shown in 12px subdued text — users may believe they are talking to a human.

Issue categorisation language

Plain language labels are the one emotionally considerate element, reducing friction of problem framing.

Dead end with no exit

No "We can't help further in-app — here's what to do" message. User is simply blocked.

Cross-task analysis

Uber Eats — Evaluation summary

3 tasks · 17 steps · PURE + Nielsen heuristics

Overall metrics

Total steps evaluated

Across 3 tasks

Steps at PURE 2+

13 / 17

76% moderate or higher

Steps at PURE 3

3 / 17

Critical failures

Task comparison

Task	Steps	Avg cognitive load	PURE 2+ steps	PURE 3 steps	Heuristic violations
Task 1 — Search & Add to cart	5	2.2 / 5	4 / 5	0	6 (H1, H2, H4, H5, H7, H8)
Task 2 — Payment & Tip	7	2.0 / 5	5 / 7	1 (Step 7)	7 (H1, H2, H3, H4, H5, H6)
Task 3 — Customer Support	5	3.0 / 5	4 / 5	1 (Step 5)	8 (H1, H3, H4, H5, H6, H7, H9, H10)
Combined	17	2.4 / 5	13 / 17	2 critical	9 unique heuristics violated

Severity-ranked findings across all tasks

Task 3, Step 5 — Customer support phone number is architecturally absent. The primary task goal cannot be completed. No system feedback informs the user of this. Maximum severity.

Task 2, Step 7 — Custom tip input displays subtotal only, not the all-in grand total. The core task goal of verifying the final price is structurally impossible before "Place order" is tapped.

Task 1, Step 5 — "Add to cart" pre-sets quantity to 2 via BOGO without explicit user consent, creating a downstream order surprise at checkout.

Task 2, Step 5 — Duplicate "Apple Pay" entries in the payment modal with no differentiating metadata make correct payment selection ambiguous.

Task 3, Step 4 — AI chatbot containment architecture routes 100% of support contacts through automation with no guaranteed human escalation path and inadequate AI disclosure.

Task 2, Step 7 — Pre-selected 18% tip with emotionally loaded label hierarchy ("You're my hero" = 25%) constitutes social engineering at a financially consequential step.

Task 1, Step 4 — Abbreviated customisation labels ("BGL Everything", "CHS Cheddar") violate real-world language expectations, raising error risk for allergen-sensitive users.

Task 2, Step 3 — "No thanks" confirmshaming on the upsell interstitial and Task 3's emotionally mismatched support UI represent a consistent pattern of user autonomy deprioritisation.

Systemic design patterns identified

Conversion over clarity

Promotional mechanics (BOGO, upsells, pre-selected tips) appear at every major decision point across all 3 tasks, consistently prioritising revenue over task focus.

Support suppression

The entire customer support architecture is designed to eliminate live agent contact. This reduces costs but creates maximum frustration for users with unresolvable issues.

Status transparency failures

H1 (Visibility of system status) is violated in all 3 tasks — from ambiguous cart quantities to missing grand totals to absent phone numbers.

Emotional asymmetry

Warm, identity-positive language is used to drive spending (tip labels). Cold, transactional language is used in support flows where empathy would reduce abandonment.

AI-Simulated Think-Aloud · SEQ Rating · Gemini Thinking · March 19 2026

User Testing Session — Alex

3 tasks · SEQ scores recorded · Gemini (Thinking) used as participant proxy

Alex — Simulated Participant

A time-pressured professional with approximately 40 minutes remaining on a lunch break at the start of the session. Motivated by speed and efficiency — wants to complete tasks quickly without distractions. Familiar with food delivery apps but not a power user.

Time-pressured Efficiency-focused Moderate app familiarity Low tolerance for friction Frustrated by automation barriers

SEQ Scores at a glance

Task 1 — Search & Add

6 / 7

Very Easy

"Very straightforward — BOGO was clearly labeled, though the required customisation felt like a tiny speed bump."

Task 2 — Payment & Tip

4 / 7

Slightly Difficult

"Navigating through an upsell page and a separate tipping flow felt tedious when watching the clock."

Task 3 — Customer Support

2 / 7

Extremely Difficult

"The app hides direct contact behind layers of automated menus and AI chatbots — the last thing I want in a hurry."

Task 1 Search for "Sandwich", pick a combo, handle promo pop-ups, proceed to cart ▼

Think-aloud transcript

Searching I'm on the home screen and I see a 'Sandwich' category icon, but I'll just hit the search bar at the top to see all my options. I'm typing in 'sandwich'.

Filtering The results are up. I see 'Williams Fresh Cafe' right at the top — it has a 'Buy 1, get 1 free' offer and a 22-minute delivery time. That's perfect; I can have the second one for lunch tomorrow. I'm tapping on that.

Choosing I see the 'Egg, BACON, Spinach & Avocado Bagel' under the BOGO deal. I'm tapping the '+' to add it.

Customising Now it's asking for 'Choices - Bagel'. It says 'Required', so I can't skip this. I'll go with 'BGL Everything'. I don't need extra cheese right now, so I'll just move on.

Cart review I'm clicking 'Add 2' — since it's a BOGO, it shows two items for the price of one, $18.00 total but with $9.00 in savings. I see the 'Saving $9 with promotions' banner at the bottom.

Checkout Everything looks right in the cart. I'm hitting 'Go to checkout' so I can actually eat before my next meeting.

SEQ Score

The process was very straightforward because the BOGO offer was clearly labeled, though having to fill out the 'Required' customisation fields felt like a tiny speed bump when in a rush.

Researcher note — PURE vs SEQ alignment

Alex's SEQ of 6 aligns with the PURE evaluation's finding of consistent moderate (PURE 2) friction — no catastrophic failure, but multiple micro-frictions. Notably, Alex independently noticed the abbreviated label "BGL Everything" as a friction point, corroborating the PURE Step 4 finding. Alex also navigated the pre-set quantity of 2 without objection, suggesting the BOGO framing was legible enough for a motivated, time-pressured user — but may be more problematic for less engaged users.

Task 2 Change payment method, set a custom tip, and check the total price ▼

Think-aloud transcript

Upsell I'm hitting 'Go to checkout', and now it's showing me a 'Complete your order' page with cookies and coffee. I don't have time for extras; I just want my sandwich. Tapping 'No thanks' at the bottom immediately.

Summary Now I'm on the main checkout screen. I see the total is $17.51. The math seems right: $18.00 subtotal minus the $9.00 promotion, plus fees and taxes.

Payment I see it's set to Apple Pay. I want to double-check my payment options, so I tap that. It opens the 'Pay with' menu. It looks like I'm sticking with Apple Pay, but finding where to tap to change it took a second because the 'Add payment method' is further down.

Tip discovery Now for the tip. I tap to 'Add a tip'. Wait — it says tips are based on the order total before discounts — $26.51? That's a bit annoying when my actual total is much lower. 18% is selected by default, but I'll go with a custom amount.

Custom tip I'm tapping 'Custom' and typing in '$4'. I hit 'Save'.

Final review Back on the tip screen, it shows 'Your tip: $4.00'. The 'Place order' button is right there. This whole payment and tipping part took way longer than just picking the food. I'm ready to just hit place order and be done with it.

SEQ Score

While the screens are clear, having to navigate through an upsell page and then a separate tipping flow felt tedious when watching the clock.

Researcher note — PURE vs SEQ alignment

Alex's SEQ of 4 is consistent with the PURE evaluation's finding of a single PURE 3 step (Step 7) dragging down an otherwise moderate-friction task. Two key PURE findings were directly echoed in the think-aloud: (1) the tip base calculated on the pre-discount total ($26.51 vs $17.51) caused visible confusion — Alex called it "a bit annoying", validating the H1 violation; (2) the difficulty locating the payment change affordance confirms the H6 / discoverability finding from PURE Step 4. Critically, Alex never received a confirmed all-in total before placing the order, exactly as predicted by the PURE Step 7 analysis.

Task 3 Find customer service for the last order and locate a phone number ▼

Think-aloud transcript

Finding help I'm on the main account page and I see 'Help' right there, so I'm tapping that to get started.

Locating order The 'Last Order' section shows my Domino's Pizza order from April 30th. I'll tap on that to see if I can get a refund or some help.

Chatbot barrier It immediately brings up a chat 'Powered by AI'. The bot is giving me canned options like 'My order never arrived' and 'Wrong or missing items'. This isn't what I want; I need a person.

Searching I'm looking all over this 'Help' screen for a phone icon or a 'Call Us' button. I'm tapping through 'Help with an order' and 'More options', but it just keeps giving me more menus for 'Change payment method' or 'Another order issue'.

Frustration I'm clicking everything, but it feels like the app is hiding a phone number on purpose to force me to use the bot. My break is almost over, and I'm still just staring at a list of links and an AI that doesn't understand I just want to speak to a human.

SEQ Score

It was extremely difficult because the app hides direct contact information behind layers of automated menus and AI chatbots, which is the last thing I want when in a hurry.

Researcher note — PURE vs SEQ alignment

Alex's SEQ of 2 is the strongest validation of the PURE evaluation across all three tasks. The think-aloud directly surfaces the same architectural dead end identified in PURE Step 5: no phone number, no free-text bypass, no human escalation path. Alex's unprompted observation — "it feels like the app is hiding a phone number on purpose" — independently identifies the deliberate support-suppression strategy flagged in the PURE analysis. The emotional trajectory also matches exactly: mild frustration at step 2 (chatbot barrier), escalating to task abandonment by step 5 as the clock ran out. The PURE analysis predicted maximum frustration; the think-aloud confirmed it verbatim.

PURE evaluation vs user testing — alignment summary

Task 1 alignment

PURE predicted moderate friction (avg 2.2). SEQ returned 6/7. Motivated users navigate BOGO mechanics successfully — friction is real but surmountable. PURE correctly identified "BGL" labels and pre-set quantity as the key friction points, both surfaced in the think-aloud.

Task 2 alignment

PURE predicted a PURE 3 failure at Step 7 (no grand total shown). SEQ returned 4/7. Alex confirmed the tip base confusion ($26.51 vs $17.51) and the payment edit discoverability issue — both predicted by PURE Steps 4 and 7.

Task 3 alignment

PURE predicted an architecturally impossible goal (PURE 3, CL 5). SEQ returned 2/7 — the lowest score across all tasks. Alex independently concluded the app deliberately hides contact info, exactly matching the PURE "support-suppression" finding.