Cypress gives frontend engineers a superpower: the ability to write E2E tests that watch our app behave just like a real user would. But with great power comes… well, a lot of subtle flakiness if you’re not careful.
The cy.wait Illusion: What's Really Happening
The scenario is simple: you have a component that loads data, and after a user action, it loads new data using the same API endpoint. To ensure the new data has arrived, you intercept the request and then use cy.wait('@requestAlias') multiple times.
// A common, flawed approach:
cy.intercept('GET', '/api/items/*', { fixture: 'item-1' }).as('getItems');
cy.visit('/items');
// 1. Wait for the initial load
cy.wait('@getItems');
// ... User performs an action that triggers the SAME request ...
// 2. Wait for the second load
cy.wait('@getItems'); // <-- THIS IS THE PROBLEM
The Flaw
Cypress's cy.intercept logic is designed to capture a single match for an alias. When you call cy.wait('@getItems') for the first time, it finds the initial request, waits for its resolution, and then the alias is fulfilled.
When you call cy.wait('@getItems') a second time, Cypress does not reset the listener. Instead, it checks if a request has already been resolved with that alias. Because the first request has resolved, the second cy.wait command resolves immediately, without waiting for the new network call to finish. Your test is now racing against the network, not waiting for it.
Fix #1: Re-intercept before each expected request
(Works, explicit, but verbose)
cy.intercept('GET', '/api/items').as('getItems_1')
cy.get('[data-testid=refresh]').click()
cy.wait('@getItems_1')
cy.intercept('GET', '/api/items').as('getItems_2')
cy.get('[data-testid=load-more]').click()
cy.wait('@getItems_2')
Clear, deterministic, but repetitive.
Fix #2: Use times: 1 to force Cypress to “consume” intercepts
(Cleaner: Cypress forgets the intercept after one match)
This is the missing tool many engineers don’t realize exists.
cy.intercept({
method: 'GET',
pathname: '/api/items',
times: 1
}).as('getItems')
// trigger request 1
cy.get('[data-testid=refresh]').click()
cy.wait('@getItems')
cy.intercept({
method: 'GET',
pathname: '/api/items',
times: 1
}).as('getItems')
// trigger request 2
cy.get('[data-testid=load-more]').click()
cy.wait('@getItems')
Why this works:
times: 1means Cypress removes the intercept after a single matching request- Re-declaring the intercept creates a fresh listener
- Each
cy.wait('@getItems')now truly waits for the next occurrence
This technique gives you explicit, occurrence-specific intercepts without alias clutter. For tests that must assert network behavior (payloads, headers, error flows), it’s a clean and robust pattern.
Fix #3: Stop waiting for requests altogether
(The best fix. UI > network.)
Here’s the golden rule:
Users don’t observe network resolution — they observe UI state.
That means the most stable tests assert what the user sees:
- A loading spinner appears → disappears
- A button becomes disabled → enabled
- A success message appears when an action is complete.
- The newly loaded element is now visible in the DOM.
Example with user-visible cues:
cy.get('[data-testid=refresh]').click()
cy.get('[data-testid=spinner]').should('exist')
cy.get('[data-testid=spinner]').should('not.exist')
cy.get('[data-testid=item-list]')
.children()
.should('have.length.at.least', 1)
No reliance on internal network timing. No alias lifecycle. Zero flake.
Accessibility makes this even more robust
Accessible UI patterns make great Cypress hooks:
aria-busy attribute
<ul data-testid="item-list" aria-busy="true">
Test:
cy.get('[data-testid=item-list]').should('have.attr', 'aria-busy', 'false')
role="status" with live regions
<div role="status" aria-live="polite" data-testid="status">
Loading…
</div>
Test:
cy.get('[data-testid=status]').should('contain', 'Loaded 10 items')
Disabled states for actions
cy.get('[data-testid=submit]').should('be.disabled')
cy.get('[data-testid=submit]').should('not.be.disabled')
These patterns aid screen reader users and produce stable, deterministic E2E tests.
When waiting for requests is appropriate
There ARE valid scenarios:
- Asserting payloads or query params
- Mocking backend responses
- Validating request ordering
- Verifying retry logic
- Testing error handling flows
For those cases: Combine times: 1 with explicit, fresh intercepts defined right before triggers.
For other cases: the test should rely on the UI state.
A combined real-world example
(Network + UI, the best of both worlds)
// UI-driven loading signal
cy.get('[data-testid=create]').click()
cy.get('[data-testid=spinner]').should('exist')
// Network contract check
cy.intercept({
method: 'POST',
pathname: '/api/items',
times: 1
}).as('postItem')
cy.get('[data-testid=create]').click()
cy.wait('@postItem')
.its('request.body')
.should('deep.include', { title: 'New item' })
// Final user-visible assertion
cy.get('[data-testid=status]').should('contain', 'Item created')
The network part is accurate. The UI part is resilient. The test is rock-solid.
Final checklist
For accessible, deterministic, non-flaky Cypress tests
- Prefer user-visible UI state, not network events
- Use
aria-busy,role="status",aria-live, and disabled states - When waiting for requests:
- Re-intercept before each occurrence, OR
- Use
times: 1to auto-expire the intercept
- Avoid global, long-lived intercepts
- Never assume
cy.wait('@alias')waits “for the next request” - Make loading and completion states accessible (good for tests, good for users)