Test Data Bottleneck Unplugged – API Test Automation with AI

The cool thing about testing is the learning. In the world of API test automation, some of us are excited to learn that the system actually does what it needs to. Some of us are excited when we find a bug. Some of us like debugging failures, analyzing what we know should be, vs what we see.

But I’ll tell you what we don’t like. What slows down our API test automation flow. The preparations. Obviously, they are essential to what we want to learn, but they’re a bottleneck.

It can be going through multiple screens to get to where we need to start exploring. Or, more often than not, it’s about setting up the right data in the database, filling it up with exactly what we need for our tests to run. Or preparing JSON files.

It’s tedious, and because humans do it, it is error-prone. We can prepare the wrong data, or if we’re not careful, use the right data in a wrong way, leading to flaky automated checks.

So how can we speed up our API test automation setup? By delegating the data creation. Let’s turn to our genie-in-residence!

Using AI for test data generation is great at handling the menial steps, which is usually safer, and a lot faster.

What can I generate today?

Let’s look at a couple of examples. In our Bigger Better Bookstore^TM, we have APIs that can help us do everything with books. When we call these APIs in our tests, it’s usually with data. Let’s see how leveraging AI can help generate these examples for us.

I even refer it to the right schema. We need to give the schema to our genie, so it can create the data in the right format. But once you did, you get lovely JSON data you can save for your test suite.

Since our AI companion doesn’t care about how much work it needs to do, we can ask it to generate data for all our API test cases. And of course, if it already suggested cases by itself, we can refer to those.

Of course, great API testing isn’t just about single API calls; it’s about sequences and workflows. When we test flows, some of the data we need is dynamic. It cannot all be pre-generated. But it can still be useful to have most of the JSON in place, and we’ll fill in the blanks.

For example, for this common automation sequence:

We can ask the AI genie to create the data for each API call. Now, the ID returned for the created book is dynamic, and the next APIs rely on it. So when the ID appears in the generated body of the requests, we’ll need to parameterize that in our automation script. But all the rest of the data will appear there, ready to go.

Beyond Inputs: API Test Automation Stepping Up

Now, input data is not the only data worth creating. There’s also assertion data.

Our GET book details API returns everything the book was created with. To properly validate this in our tests, we need to check the response. We can use our AI genie to create a sample of that expected data, and we can compare them.

For multiple sets of inputs, we can ask the AI to create their assertion counterpart. This may not be the exact data that’s going to be returned. Some of it is dynamic, some may be inferred, and some plain guesses. But it gives me a solid baseline. In my tests, I can visually compare the actual result and the pre-generated assertion data and confirm it’s what I expected.

While I’m at it, why not ask the AI dude to create a script that compares the actual and expected result, omitting dynamic data like IDs or timestamps, and focuses on the important parts?

Take it a step further, and ask it to turn the whole flow into an automated test that can run all the time. That’s the real power: moving from AI for test data generation to a complete, ready-to-run API test automation script.

Data preparation is not limited to APIs’ request and response bodies. You can generate data, images, scripts—but YOU need to approve all of them. Otherwise, we can’t trust the results.

In the end, effective testing is about the learning. By embracing AI for test data generation, we not only get to that learning stage more quickly, we can build more comprehensive test suites in our spare time.

But the generation tasks need to be delegated properly. That’s real delegation, not sending the genie on an errand without checking what it brought us back.

Now it’s your turn to talk to the genie

Besides creating input data, what other tedious testing tasks would you offload to an AI assistant? Let’s brainstorm some ideas in the comments.