Over the past week other two, I’ve been using ChatGPT 5.6 extensively to help develop assessment templates for vocational education and training. This has not involved simply asking it to “create an assessment tool” and accepting whatever comes back. My process has been much more deliberate, iterative and, at times, frustrating.
Rather than building another collection of static Word documents, my goal has been to turn our existing assessment templates into interactive HTML forms that can be hosted on a website, completed online and downloaded as finished PDF documents.
The process generally starts with the original Word template. I drag the document directly into the chat and provide Codex with specific instructions about what I want it to become.
These instructions are not simply, “convert this into HTML”. I explain how the form should behave, how it should look, which sections need to remain editable and how the final output will be used by students and assessors.
For example, I might ask it to:
- reproduce the structure and wording of the original assessment template
- convert response areas into expandable text fields
- create checkboxes, drop-down menus, tables and declaration fields
- add buttons for saving, printing or downloading the completed form
- apply SpecTraining’s colours, fonts and branding
- make the form responsive so it works on different screen sizes
- generate a clean PDF version when the learner has finished
- ensure the file can be uploaded and hosted as a webpage
Codex then produces the HTML, CSS and JavaScript required to create the application. In some cases, it can also help organise the supporting files, improve the layout and troubleshoot problems when something does not behave as expected.
One of the biggest benefits has been the speed at which I can move from an idea to a working prototype.
Traditionally, creating an interactive assessment form would require a web developer, a detailed specification and several rounds of revisions. With Codex, I can upload the existing template, describe the outcome and begin testing a functional version almost immediately.
The ability to refine the form through conversation is also extremely useful. I can ask it to widen a response box, change the placement of a button, improve the PDF formatting, restructure a table or apply the same design approach to another assessment document.
This makes the development process far more accessible for someone who understands assessment design and user experience, but is not necessarily an expert programmer.
However, the process has not been completely smooth.
The first version is rarely the final version. A form may look correct on screen but behave differently when exported to PDF. Tables may split across pages, text may be cut off, fields may not expand correctly or the downloaded document may not match the appearance of the webpage.
Codex can also misunderstand instructions, particularly when the original Word template contains complicated tables, merged cells, repeated headers or unusual formatting. Sometimes fixing one problem creates another.
There is also a risk that the technology becomes more important than the assessment itself. A visually impressive form is not automatically a valid assessment tool. The instructions, evidence requirements, mapping and assessor guidance still need to be reviewed by someone who understands competency-based assessment.
Another lesson has been the importance of giving precise instructions. Codex performs much better when I explain the purpose of the form, the learner journey, the required functionality and the expected output. General instructions tend to produce generic results.
Overall, the experience has been extremely promising.
ChatGPT 5.6 and Codex have allowed me to take assessment templates we already use and begin transforming them into more modern, interactive resources. The process still requires testing, judgement and persistence, but it has significantly reduced the barrier between identifying an improvement and actually building it.
For me, this is where AI becomes genuinely useful. It is not replacing assessment expertise or web development. It is allowing someone with a clear idea to design, test and improve digital solutions far more quickly than would previously have been possible.