AI can help you turn a vague instruction such as “check the website” into a useful test plan. The important distinction is between suggesting a check and actually passing it. A chatbot that has read your homepage text has not necessarily opened the mobile menu, followed the enquiry journey or confirmed that a message reached your inbox.
This workflow shows a small business how to use AI to organise website quality assurance before a launch or after a change. It produces a short list of reproducible tests, evidence and repair priorities. You remain responsible for checking the results in a browser, rather than treating confident AI wording as proof.
Choose one important journey: finding a service, checking whether it suits the customer, then contacting the business. Start with the relevant pages rather than asking AI to assess every aspect of the website at once.
Prepare a small, non-sensitive information pack:
Do not upload customer enquiries, passwords, private analytics exports or unredacted administration screenshots. If an AI tool supports images, use screenshots of public pages without personal information. Check the tool’s data-handling settings and your business’s rules before sharing anything non-public.
For a substantial change, agree a recoverable backup and test environment with the person maintaining the site. Our guide to WordPress staging sites explains when a separate test copy can be useful. Do not grant an AI tool live editing access simply to generate a checklist.
Use the following prompt with your own approved information:
Act as a website QA planning assistant. The customer needs to [task]. The journey is [page addresses and intended steps]. Use only the page text and observations I provide. Create up to ten tests. For each, give the page, exact steps, expected result, evidence required and business impact if it fails. Separate observed problems from things that still need testing. Do not say a feature works unless I provide a test result. Do not invent prices, service areas or promises. Do not recommend design changes or write repair code yet.
If the tool cannot open the supplied URLs, give it the relevant public text instead. Record that limitation. A model may propose tests for features your site does not have, so remove irrelevant suggestions before using the list. The outcome should be a manageable testing sheet, not an impressive-looking but unverified audit score.
Start with navigation, content and the intended action. On each selected page, follow the main links and record the actual destination. Check that headings describe the current service and that contact details match the approved business information. A reachable page can still be the wrong destination.
Open the journey on a real phone as well as a desktop browser. Check whether you can open and close the menu, read the text and reach the primary button. Chrome’s Device Mode documentation describes desktop simulation as an approximation, not a substitute for running the page on a mobile device. It is useful for initial checks at different widths, but label simulated and real-device results separately.
Include basic accessibility checks such as keyboard operation, visible focus and form labels. The W3C Easy Checks guide gives a starting point and explicitly cautions that passing these checks is not a comprehensive accessibility evaluation. Our keyboard-only enquiry review provides a focused companion exercise. Neither an AI screenshot review nor a short checklist establishes accessibility compliance.
For forms, distinguish reaching the form, completing validation, receiving the success message and receiving the actual email. Only send test enquiries with the business’s approval. Use clearly labelled synthetic details and tell the recipient what to expect. Do not create live bookings, payments or orders just to complete an audit. Our guide to missing enquiries after a form says sent explains why the inbox check matters.
Imagine a fictional service business refreshing its enquiry page. Its goal is simple: a phone visitor should find the service, understand what information to provide and send an enquiry. This is an illustrative exercise, not a TempWebz customer case study.
Suppose A and B pass, while C displays success but no message is found in the agreed inbox or spam folder. Record “delivery not confirmed”, not “email server broken”. The evidence narrows the investigation but does not establish the cause. AI can help format the report; the support provider still needs to investigate the delivery path.
Give the tool your redacted observations using this follow-up prompt:
Turn these test results into a repair list. Keep the exact observed facts and label unknowns. For each failed test, include the URL, device/browser, reproduction steps, expected versus actual result, customer impact and next diagnostic check. Do not infer a root cause from a screenshot alone. Group duplicate observations. Put blocked customer journeys ahead of cosmetic suggestions. Leave successful tests marked passed and untested items marked untested.
Review every generated item. An unsupported diagnosis such as “the plugin is incompatible” can send a developer in the wrong direction. Replace it with the actual symptom and the evidence. If the AI proposes deleting a plugin, changing DNS or replacing a form, do not act on that recommendation without understanding the impact and having an authorised recovery plan.
Assign each confirmed issue an owner and a specific next action. After a targeted repair, repeat the same test on the same page and device. Save the before-and-after evidence in a website change log. A saved setting, cleared cache or “fixed” message is not equivalent to a successful customer journey.
Finish with three short lists: verified passes, confirmed failures and checks not yet completed. Decide which failures prevent release and which can be scheduled based on actual impact. Keep aesthetic preferences separate from functional defects so the review does not turn into an unplanned redesign.
The useful role of AI here is preparation and organisation: better test questions, clearer evidence and a readable repair brief. It does not certify security, guarantee search rankings or replace specialist investigation. If a confirmed website problem needs technical help, send the affected URL, reproduction steps and redacted evidence when requesting TempWebz technical support.