United Kingdom version. Set to the United Kingdom: prompts generate UK-format test data (postcodes, Ofcom drama numbers, DD/MM/YYYY, pounds), target WCAG 2.2 AA for public sector services and apply UK GDPR to logs. US version

AI Prompts for Software Testers (UK)

Prompt templates for UK software testers: test cases, bug reports, Playwright skeletons and UK-format synthetic data, with WCAG 2.2 AA for public sector work.

The interactive tool loads here when JavaScript runs.

About this tool

What is AI Prompts for QA Engineers?

With the region set to the United Kingdom, every prompt in this kit tells the assistant to write user-facing text in British English, to format dates as DD/MM/YYYY and money in pounds with VAT as the product specifies, and to build test data from UK formats. It also forbids real identifiers: test phone numbers come from Ofcom's ranges reserved for drama (07700 900000 to 07700 900999), emails use example.com, and no real NHS or National Insurance number goes into a fixture.

Accessibility is where UK testing differs most. Public sector websites and apps fall under the Public Sector Bodies (Websites and Mobile Applications) Accessibility Regulations 2018, and GOV.UK guidance sets WCAG 2.2 level AA as the target, with an accessibility statement. The UK rules tell the assistant to use that level for public sector products and the team's agreed level otherwise, so a generated test charter or release summary names the right bar.

The synthetic data template gets a UK note: postcodes in the right outward and inward shape, Welsh place names with diacritics, surnames with apostrophes and BFPO addresses as edge cases. Those are the inputs that break postcode lookups and name fields in UK forms.

  • Ten QA templates from test design and bug reports to flaky-test triage and release readiness
  • Framework choice for Playwright, Cypress or Selenium skeletons
  • Boundary value and equivalence class instructions built into test-case prompts
  • Flags emails, phones, card-like numbers and IP addresses in pasted logs
  • Favourites, keyword search and a Markdown library export for your QA wiki
Read the guide: AI for Civil and QA Engineers: Prompts That Respect the Engineer of Record

Updated

How to use AI Prompts for QA Engineers

  1. Name your product and team context, for example B2B billing web app, and pick a region if formats like dates or postcodes matter.
  2. Choose a template: Test cases from a user story, Bug report, Automation skeleton, Exploratory charter, Regression scope and more.
  3. Paste the story, acceptance criteria, notes or diff summary and set options such as the framework or priority scheme.
  4. Run the privacy check on logs or data you pasted, then copy the prompt into your assistant or IDE chat.
  5. Review the output: run the code, prune duplicate cases and confirm expected results with the product owner.

Worked examples

Masking a checkout log before asking for triage help

Input

Checkout failed for jane@example.co.uk at 10.12.4.33 with postcode SW1A 1AA and NI number QQ 12 34 56 C

Output

Checkout failed for [EMAIL-1] at [IP-1] with postcode [POSTCODE-1] and NI number [NATIONAL-ID-1]

The email, IP address, postcode and labelled National Insurance number are all masked, so the log line can go into a prompt. Even fake-looking values are masked, because the check cannot tell a test value from a real one.

UK-shaped synthetic customers

Input

Synthetic test data set, region United Kingdom, 20 rows of customer records in CSV

Output

Prompt asking for valid-shape postcodes, 07700 900xxx numbers, DD/MM/YYYY dates and amounts in pounds, with edge-case rows labelled

Ofcom reserves the 07700 900xxx block for fiction, so the numbers cannot reach a real person if a test sends an SMS by mistake.

Background and insight

Boundary value analysis and equivalence partitioning are decades-old test design techniques that models apply well when asked explicitly. If a field accepts 1 to 100 items, the interesting inputs are 0, 1, 2, 99, 100 and 101, plus invalid types. Without that instruction, a model tends to produce happy-path cases. The templates ask for the partition table first, then cases derived from it, so you can see whether a class was missed.

Generated test code is a skeleton, not a suite. A model cannot see your DOM, API contract or fixtures, so it invents selectors and endpoints. The Playwright, Cypress and Selenium template asks for user-facing locators (role, label or text) or test ids rather than CSS chains, explicit TODO comments for anything unknown, and no hard-coded waits, which are a common source of flaky tests.

Test data and logs often contain production customer data. Do not paste real records into a public assistant; under GDPR, data that can still be linked to a person remains personal data even when pseudonymised (Recital 26). The synthetic data template generates fictional records from a schema instead, and the privacy check flags emails, phone numbers, card-like numbers and IP addresses in anything you paste.

Common use cases

  • Sprint test design

    Turn user stories and acceptance criteria into partitioned test cases with boundaries before the build lands.

  • Clear defect reports

    Convert rough notes into reproducible bug reports with severity, environment and evidence sections.

  • Automation starts

    Get a Playwright, Cypress or Selenium skeleton with page objects and TODOs to wire to your app.

  • Release decisions

    Summarise test results, open defects and risks into a release readiness note for the go or no-go meeting.

Tips and limitations

  • For a GOV.UK service, run the accessibility checks against WCAG 2.2 AA and keep the accessibility statement in step with what you find.
  • Do not paste production logs from a UK service into a consumer AI tool; mask them here first or use your company's approved tool.
  • Test postcode lookups with Northern Irish (BT) and Channel Islands postcodes as well as mainland ones; they are often handled differently.

Sources and further reading

Frequently asked questions

Which WCAG version do UK public sector services have to meet?

GOV.UK guidance under the Public Sector Bodies Accessibility Regulations 2018 points to WCAG 2.2 level AA, with an accessibility statement explaining any gaps. Private sector products are not covered by those regulations but still have duties under the Equality Act 2010.

What phone numbers are safe to use in UK test data?

Ofcom reserves number ranges for drama and fiction, including 07700 900000 to 07700 900999 for mobiles. They are not allocated to real subscribers, which makes them a good choice for fixtures and screenshots.

Can I paste a stack trace from a live UK system into an AI assistant?

Only after removing personal data and secrets, or into a tool your organisation has approved. Logs often contain emails, IP addresses and postcodes, which are personal data under the UK GDPR. The privacy check masks those patterns but not names written on their own.

How should test data handle National Insurance numbers?

Never use a real one. HMRC's National Insurance Manual gives the format (two prefix letters, six digits and a suffix of A, B, C or D) and says D, F, I, Q, U and V are never used in the prefix. HMRC's own example, QQ 12 34 56 A, has the right shape but can never be issued, which makes it a safe fixture; keep a few with valid prefixes, labelled fictional, for positive validation tests.

Will the prompts write Playwright tests that use British spelling in assertions?

User-facing text in the prompts is British English, but assertions should match your app's actual copy. If your app says "Colour", the generated test should too; check selectors against the app, not the draft.

Is there an in-depth guide to AI Prompts for QA Engineers?

Yes. “AI for Civil and QA Engineers: Prompts That Respect the Engineer of Record” covers the background, how it works, common pitfalls and worked examples. Read the full guide.