The page is only useful if the task works
A visitor submits the form and sees 'Thank you', but sales receives nothing. Trace one labelled test request to the receiving system before changing the design. A confirmation on screen is not yet a delivered enquiry.
Follow the whole promised task: open the link, enter data, correct an error and locate the final record. That gives you a specific defect to fix instead of the vague conclusion that the page does not convert.
Readable design and a convincing offer still matter. But a broken action blocks people who already want to proceed, so fix it before sending more traffic into the same failure.
Every field should have a reason
Ask the receiving team what it will do with each answer at this stage. Keep what is needed to fulfil the request, defer later requirements and explain unusual questions.
Show which fields are required before submission. Keep labels visible while typing; a disappearing hint should not be the only reminder of what the field means.
In a hypothetical consultation form, the first response needs a name, contact route and short description, while detailed qualification can wait. Check enquiry quality after simplifying; fewer fields are not automatically better.
Write labels around the user's information
Use the customer's words: 'Your website', not internal database terminology. Explain an unusual format before the person has to guess it.
Test the formats relevant to the market: international phone numbers, names with spaces and local address conventions. Your developer's own details are not a complete test set.
Carry forward information already entered where appropriate and allow correction. Also name the commitment accurately: requesting a quote and placing an order are different actions.
An error should tell the user how to recover
Validation checks whether an entry can be accepted. An error should name the field and correction rather than simply say 'Invalid input'.
W3C's guidance calls for understandable success and error feedback. Put the explanation near the affected control and make the overall outcome discoverable. A red border alone may not explain what happened.
Retain valid entries after an error and test several errors together. Choose feedback timing that helps completion rather than treating an unfinished email address mid-typing as a final failure.
Unhelpful
- A red border with no explanation.
- A general error that clears the form.
Recoverable
- The field and correction are named.
- Other valid entries remain available.
Keep correct entries intact while the visitor fixes the affected field.
Check the occupied screen
Test an empty form, a focused field with the keyboard open and a form showing errors. Reach the next control and submit in every state. Working only before typing is not enough.
Inspect fixed footers, chat and persistent buttons together. They can cover the active field or its conditions. Check limited screen height as well as width.
Test touch and keyboard navigation separately. A hover-only hint may be unavailable on a phone; a control unreachable by keyboard blocks another way through the task.
Links need a predictable destination
Open the links, not just their labels. Verify the policy destination, phone number and messaging account. Reused templates may retain old addresses behind new text.
Explain an external handoff and test the return. Does the person still have their entries and the correct next action? 'Request a quote' should not silently become a purchase commitment.
Make links recognisable and keyboard focus visible. A person should be able to predict the destination before choosing the link.
Submission has two sides
Compare the browser message with the receiving record. Success without a record points to delivery; a record without clear confirmation can prompt repeated submissions.
Use a unique test marker and agree how to exclude it from operational reporting. Verify any required downstream step, such as a notification, not only initial storage.
Interrupt or delay the test connection safely. The page should not claim success before it knows the result, or leave the visitor unsure whether retrying will create a duplicate.
Inspect the page and receiving system.
- Message only
Find why delivery did not complete.
- Record only
Repair the visitor's confirmation.
- Both agree
Check that the request appears only once.
A successful submission should produce one clear confirmation and one corresponding business record.
Watch a person use the form
Ask someone unfamiliar with the page to complete the task using test data. Give the goal, not a guided tour. Watch where they hesitate before explaining anything.
Record the observation separately from the theory. Opening the privacy link before entering a phone number is observable; 'nobody trusts us' is a much broader claim. Investigate the specific wording first.
Automated checks find some technical and accessibility defects. They do not establish that the offer or explanation makes sense. Combine them with a manual walkthrough and observation of a real person.
Repair in an order that protects the task
- Fix missing submissions, wrong destinations and blocked controls.
- Make required information and material conditions clear.
- Add specific error recovery and reliable confirmation.
- Resolve mobile overlap and keyboard navigation failures.
- Then evaluate visual refinements against the same working journey.
Keep the event definition unchanged when comparing before and after. Moving a measurement trigger can raise a reported rate without improving the form. Record the release and exact behaviour changed.
Finish with a new person completing the task without help, receiving an accurate result and appearing once in the receiving system. Visual improvements can then support a working action.
Questions answered
Should a form always have fewer fields?
No. Request what is necessary at this stage and defer the rest. Check whether the revised form still gives the receiving team enough information to act.
Is a success message enough to verify a form?
No. Find the matching record and verify any required onward delivery using clearly labelled test data.
