Dec 25,2025 
Many people turn to language models, hoping they can play the part of their users. People do so to avoid long recruitment timelines, the costs behind user research studies, and the difficulty of finding their target demographics.
Gemini offers a feature called Gems that lets you build a persistent persona. You give it a name, a backstory, instructions on how to respond. The model then carries those traits through your conversation. It becomes a synthetic user, a stand-in for the people you hope to reach.
This approach has limits. The persona you create depends entirely on the prompts you write and the patterns Gemini learned during training. But it can still be useful for quick feedback, early-stage concept testing, and pressure-testing your assumptions before you bring in actual humans.
Here is how to set one up and what to keep in mind as you do.
Getting Started with Gemini Gems
Go to gemini.google.com and look for the option to explore Gems. Click on it, then select New Gem. You will see a space to name your creation and write the instructions it should follow.
The name matters less than the instructions. This is where you define who the synthetic user is supposed to be. Think of it as writing a character brief for an actor. The more specific you are, the more consistent the responses will feel.
You can also upload files to give your Gem additional context. Product documentation, feature lists, previous user feedback, anything that helps ground the conversation in something concrete. The Gem can reference these documents when you ask questions, which keeps the responses tied to real information rather than generic guesses.
Writing Instructions That Work
The quality of your synthetic user depends on the instructions you provide. Vague prompts produce vague outputs. If you tell Gemini to be a “tech-savvy millennial,” you will get responses that feel like they were pulled from a marketing slide deck. Generic in, generic out.
Instead, break down the persona into specific elements. Define the task you want feedback on. Add context about the situation this person would find themselves in. Specify the format you want responses to take.
For example, you might write something like this: “You are a 34-year-old project manager at a mid-sized logistics company. You have tried 3 different project management tools in the past 2 years and felt frustrated by onboarding complexity. You prefer software that does not require team-wide training sessions. When reviewing a new tool, you focus first on how quickly you can set up your first project.”
This kind of specificity gives Gemini something to work with. The responses will still be generated from patterns in training data, but they will at least be filtered through a consistent lens.
Structuring Your Prompts
Gemini responds well to structured formatting. Using markdown or XML-style tags helps the model distinguish between your instructions and the actual content you want it to engage with.
If you are sharing a product description for the synthetic user to react to, wrap it clearly:
<product_description>
Your product text goes here.
</product_description>
This prevents confusion. The model knows where your instructions end and where the test material begins.
After receiving a response, consider asking the model to critique itself. A simple follow-up like “Does this response feel authentic to the persona you were given?” can surface inconsistencies. The model may notice when its answers drift into generic territory.
What Gemini Gems Can and Cannot Do
A synthetic user built this way can give you directional feedback. It can help you spot awkward phrasing in your product copy. It can identify features that might confuse someone unfamiliar with your industry. It can role-play objections you might hear on a sales call.
But it cannot replicate the reasoning of a specific person. Gemini generates outputs based on statistical patterns. When you ask it to be a frustrated project manager, it draws from everything it learned about project managers, frustration, and software preferences. The result is a kind of averaged response, plausible but not grounded in any single human mind.
This matters when you need to predict how a particular audience will actually behave. Category averages miss the outliers, the edge cases, the people who respond differently because of their unique history.
Choose Evelance When You Need Something Closer to Human Accuracy
Evelance’s personas are grounded in real human behavior. Rather than constructing personas through prompt engineering, Evelance models internalize behavioral data until it becomes part of how they respond. Each persona carries identity, memory, and situational context that shapes their reaction to your specific design.
In testing, Evelance predicted how real people would respond with 89.78% accuracy. The personas flagged the same concerns, valued the same features, and expressed the same hesitations as their human counterparts.
Evelance is a different platform category. You are not asking a language model to play a character. You are working with a model that has absorbed enough about a type of person that its outputs track with what those people actually say and do.
For early exploration, Gemini Gems can be useful. For product decisions that hinge on accurately predicting human response, the gap between prompted personas and modeled behavior becomes harder to ignore.
Choosing the Right Tool for the Task
If you need quick feedback on messaging, a rough check on user flow, or a way to pressure-test your assumptions before a real user study, building a synthetic user in Gemini can save time. It is accessible, fast, and good enough for many early-stage questions.
If you need to trust the feedback, if you are making decisions that will shape development timelines or feature priorities, then you need something that gets closer to how real people actually respond. That is where Evelance fits.
Both tools have their place. The question is which kind of feedback your current decision requires.