Surbee gave me inaccurate or misleading information. What's going on?
Why AI-generated surveys, analysis, and answers can be wrong, and what to do about it.
Why AI-generated surveys, analysis, and answers can be wrong, and what to do about it.
Surbee is built on large language models. Like all AI systems of this kind, it can sometimes produce information that sounds confident and well written but is partly or completely wrong. That is why you'll see the note "Surbee can make mistakes. Double check important info." next to the chat box.
This article explains why it happens and how to work around it.
Large language models generate text by predicting what is most likely to come next, based on patterns learned from very large amounts of data. They are not looking up facts in a database, and they have no built-in sense of what is true. That produces a few predictable failure modes:
None of this means the model is being deliberately misleading. It has no intentions. It's producing the most plausible-looking continuation of your prompt.
The parts of Surbee that use AI, and so are subject to this, include:
Your raw response data is not AI-generated. Counts, individual answers, and export files come straight from what respondents submitted. It's the interpretation layer on top that can be wrong.
Always review a generated survey before you publish it. Read every question and every answer option. Check that the response types are right, that required questions are the ones you intended, and that any conditional logic sends people where you expect. Use the preview to run through the survey yourself, including the paths you did not design for.
Verify anything important against a primary source. For claims about the world, check a source you trust. For claims about Surbee, check the help center or the docs.
Be specific in your prompts. Tell Surbee your audience, your goal, the number of questions you want, and any constraints such as reading level or required topics. Vague prompts leave more room for the model to guess.
Ask it to show its work. Asking why a question was worded a certain way, or which responses support a summary, makes it easier to spot where an answer went off track.
Sanity check the numbers yourself. If an insight cites a percentage, confirm it against the response counts on your dashboard.
Feedback genuinely helps us improve the product, so please send it.
If an answer is not just wrong but harmful, unsafe, or violates our acceptable use policy, please report it the same way and flag it as such.
Surbee is not a substitute for a professional. Do not rely on it for medical, legal, financial, or other advice where being wrong carries real consequences, and do not present AI-generated content to respondents as verified fact without checking it first. If your survey touches a regulated area, have a qualified person review it before you publish.