INP observes the visual response after an action
A website can display content promptly yet feel difficult to use. You open a menu, select an option or type in a search box and wait for a visible change. INP is a measure that helps assess this response to user actions.
The initials stand for Interaction to Next Paint. It observes how long the browser takes to show the next visual change after an interaction. It does not necessarily measure the time needed to complete a search, purchase or other full operation.
The official INP definition on web.dev covers clicks, taps and key presses. Scrolling and hovering are not interactions that INP measures. Describe other frustrations directly rather than assuming every delay is an INP problem.
Separate the first response from results arriving later
Consider a product filter. Selecting a brand may immediately display a loading indicator while the products take longer to appear. The initial visual response and the wait for results are separate parts of the journey.
In another case, nothing changes after a click and the visitor tries again because it seems not to have worked. That initial lack of response deserves investigation. Without measurement, its relationship to browser work, presentation or another condition remains unknown.
A good INP does not certify that every operation finishes quickly. It also does not explain the cause of a slow interaction. Describe both waits so that improving one part is not mistaken for resolving the whole problem.
A high value guides investigation but does not identify a particular button
The published field assessment thresholds are: up to 200 milliseconds for good responsiveness, over 200 and up to 500 for responsiveness needing improvement, and over 500 for poor responsiveness. The linked definition explains how visits are aggregated for that assessment; it is not the average of every click.
Check which page or group of pages the data describes and whether it concerns phones or desktop computers. Insufficient data is not evidence of a good experience. A test that only loads a page without using it may also miss the actions you care about.
The web.dev INP investigation guide distinguishes real-visit data from controlled reproduction. A general report may indicate an issue without identifying its interaction; more detailed measurement, where available, provides context. For the general report, see how to interpret PageSpeed.
Record the action, first visible response and what happens afterwards
| Action to try | First response to observe | What to record afterwards |
|---|---|---|
| Open the mobile menu. | Whether the panel begins opening or the button changes state. | Whether options work and any further delays. |
| Filter products by brand. | Whether the selection changes or a loading indicator appears. | When the list arrives and whether it matches the selection. |
| Type in the search box. | Whether letters appear normally. | Any separate delay in suggestions or results. |
| Submit a test form. | Whether the button or message acknowledges the action. | Whether the expected confirmation appears, separately from email delivery. |
Add the page address, device, browser, connection, test time and whether the page was still loading. Record occasions when it responds normally too. Agree tests beforehand where actions create orders, bookings or messages, avoiding unintended real transactions.
A recording can help reproduce the situation, with customer details and credentials kept out of view. There is no need to estimate an exact time by eye: a clear description gives your team something to measure.
What your technical team needs to investigate
Once the action is identified, the team can examine what the browser was doing before handling it, what work the action triggers and how the next update is displayed. These are separate stages of a response.
Tests may connect a delay with the website’s code, work performed on entry, a complex screen or third-party tools. That connection should come from evidence rather than counting plugins or assuming slow hosting.
If an integration is implicated, use the external tools inventory to assess its purpose before changing it. A business owner can explain which task matters and how it should behave without diagnosing the code.
An example: a filter that seems unresponsive
Fictional example separating stages. It contains no actual measurements or confirmed technical cause.
A shop receives feedback that “the filter does not work”. Reproducing the journey should establish whether the selection appears late, whether it changes promptly but products arrive later, or whether the result is incorrect. Each situation raises a different question.
A prompt visual response can help customers understand that their action was received. If the product wait continues, that needs investigation too. Improving INP does not replace validating the result or checking the full task duration.
Verify improvement in the same interaction
Before changes, agree the action to improve and how to repeat its test. Afterwards compare equivalent conditions, keep results and verify that the function remains correct. Aggregated real-visit data may concern a different period from the new test.
If opening pages is also slow, bring both observations together using the slow journey guide. Avoid judging the entire experience from one metric or a homepage screenshot.
We can help assess website performance and responsiveness, connect data with real tasks and prioritise further work. The assessment defines the checks; corrections have their own agreed scope.