AI psychological tests: choosing a reliable one and reading results

CareOps editorial team · November 7, 2026 · 7 min read

Mental wellness information and service guidance published by CareOps.

These days, if you ask ChatGPT or Gemini to “test whether I'm depressed,” a few questions appear in seconds. Apps and websites also offer plenty of quizzes labeled “AI psychological test.” But under the same label, some results are worth reflecting on and some are just for fun. Here is how to tell them apart, and how to read your results together with AI.

Which test fits which concern is covered in free psychological tests, and the differences among depression tests in depression tests compared. This post focuses on how to sort through tests labeled “AI.” If you'd like to try one now, pick from our free psychological tests, no login needed.

What is an AI psychological test? #

The phrase “AI psychological test” is used in two ways. One is an on-the-spot quiz: an AI makes up questions during a conversation, asks them and interprets your answers. The other is a published psychological scale, built by researchers and published in a journal, which an app administers as written while an AI helps you read the result. Both look similar, since you answer questions and get a result, but how far you can rely on the result differs.

How are published psychological scales different? #

A scale is built by asking many people the same items and checking how well the scores match an expert interview. The PHQ-9 for depression, for example, was completed by 6,000 clinic patients, and for 580 of them the answers were compared with an interview by a mental health professional. A score of 10 or more identified major depression with 88% sensitivity and 88% specificity, and scores of 5, 10, 15 and 20 marked mild, moderate, moderately severe and severe levels (Kroenke et al., 2001).

The GAD-7 for anxiety was completed by 2,740 adults at 15 US primary care clinics, and 965 of them had an interview with a mental health professional within a week. At the cutoff, sensitivity was 89% and specificity 82% (Spitzer et al., 2006). The ECR short form (ECR-S), which looks at attachment in close relationships, shortened the original long test to 12 items and was checked across six studies; its two dimensions, anxiety and avoidance, showed the same validity as the original (Wei et al., 2007).

You can take all three on CareOps right away, no login needed: the PHQ-9 depression check, the GAD-7 anxiety check and the ECR-S attachment check. Each one lists the original scale's authors and year.

How do I choose a reliable AI psychological test? #

Rather than the “AI” label, look for the five things below on the screen. If all five are there, the result is worth reflecting on.

1. A scale name and source: does it have a name like PHQ-9 or GAD-7, and does it say who published it and when?
2. Fixed items: do the same questions come up in the same order when you retake it? If questions change each time, scores can't be compared.
3. A time frame: does it say which period to think about, such as “over the past two weeks”?
4. Score ranges: is the rule for what each score means made public?
5. How answers are kept: does it explain where your answers are stored and whether you can delete them?

Can I ask ChatGPT to make me a psychological test? #

General-purpose AI such as ChatGPT, Gemini and Claude can quickly write questions that fit your situation, and organizing your answers can help you think more widely. But questions made this way have no set scoring rules or cutoffs, so it's hard to gauge where you are with a number or compare with a few weeks later. A good split is this: when you need a score, take a published scale, then bring that result to the AI and talk it through.

How do I read my test results together with AI? #

1. Look at the items that scored high, not only the total: the same score of 10 leads to different conversations depending on whether sleep and focus or loss of interest scored high.
2. Connect it to daily life: ask the AI to help you sort out when and in what situations the high-scoring items actually show up.
3. Follow the trend, not one result: retake the same test every few weeks and look back on what changed and what happened in between.
4. Leave out names and contact details: when you paste results into a general-purpose AI, include only scores and items and remove anything that identifies you.

For example, you could say (a made-up example): “I took the PHQ-9 depression test and scored 12. The sleep and concentration items were high. Help me sort out which scenes in my daily life this connects to.” A result is a starting point for looking at where you are, not a diagnosis. If you're wondering about a condition or medication, check with a psychiatrist. More ways to start a conversation with a general-purpose AI are in ChatGPT counseling prompts.

How has your mood been lately? The PHQ-9 depression check #

The PHQ-9 looks back on your mood and daily life over the past two weeks in nine items. It takes two to three minutes, and the result shows which items scored high.

How does CareOps Maeumi read test results with you? #

CareOps Maeumi is an AI counseling app informed by established approaches such as cognitive behavioral (CBT) and acceptance-based (ACT) techniques. When you take a test in the app, Maeumi briefly explains what your score means right under the result card, and you can keep asking for more detail or for something to try. When you start, a short intake and initial check-in look at where you are, and the conversation continues with the approach that fits. Your first session is free. Get the app

This guide shares mental wellness information. If you are in immediate danger, contact emergency services where you are. Support and contact information.