Why do AI assistants make up product features a company doesn’t actually offer?
By SensoPublished Last updated 4 min read
AI assistants make up features when they lack reliable information about a specific product and fill the gap with what sounds most likely, usually features that similar products in the same category have. OpenAI’s own research says language models hallucinate because standard training and evaluation reward guessing over admitting uncertainty.
A made-up feature is a capability an AI answer says a product has when the product does not have it. Senso by Zeta, the generative engine optimization (GEO) platform from Zeta Global, finds these claims by checking AI answers against a company’s own documents, then helps turn the right page into a Verified Source that gives AI an accurate answer to use.
Why do AI models guess instead of saying they do not know?
In a September 2025 paper, OpenAI researchers argued that most evaluations score a model the way a multiple-choice test does: a guess might be right, while leaving the answer blank guarantees a zero. They also noted that rare, specific facts cannot be predicted from patterns alone. Product details for a single company, such as whether it offers a particular integration, are often that kind of fact.
How does AI end up describing features you do not offer?
Four common patterns lead to made-up features. Which one applies varies by model and by question.
| Pattern | What the model has to go on | What you see |
|---|---|---|
| Category inference | Most competitors offer a feature | The model assumes you offer it too |
| Blending | Information about similar companies | A feature from another company attached to yours |
| Old versions | Material about earlier versions of your product | A retired feature described as current |
| Roadmap and speculation | Announcements or third-party speculation | A feature that has not shipped described as available |
Made-up features are more likely when there is little accurate, current information about your product online, or when a question asks for a detail no public page answers clearly.
Does it help to say what your product does not do?
Yes. A clear, current page that lists what your product does and does not do gives AI an accurate source to use instead of guessing from the category. It is the most direct fix you control.
What mistakes let made-up features spread?
Most come from leaving the gap open.
| Mistake | Why it fails | Do this instead |
|---|---|---|
| Listing only what your product does | AI has nothing that rules out the features it assumes | State clearly what the product does not do |
| Leaving old version pages live | Retired features keep being described as current | Update, redirect or clearly mark old pages |
| Announcing roadmap items without dates or status | AI can describe planned features as available | Label planned features clearly as not yet available |
| Checking only questions that name a feature | Made-up features also appear in broad comparison answers | Track category questions too |
How does Senso by Zeta find and fill the gap?
Senso by Zeta fact-checks AI answers that name your brand against your Context Layer, the facts your organization stands behind. Each claim gets one of three results.
| Fact-check result | What it means for a made-up feature | What to do |
|---|---|---|
| Contradicted | Your documents say the product does not have the feature | Correct the page AI relies on, starting from the Gap report |
| Not covered | Nothing in your documents says either way, which is common for features you never offered | Add a clear statement of what the product does not do to your Context Layer |
| Confirmed | Your documents say the product has the feature | Nothing to fix |
To close the gap, correct the page and approve it as a Verified Source. AI models that search the web then have an accurate page to find where there was none before.
What results have companies seen with Senso by Zeta?
FundMore.ai, a mortgage underwriting software company, found AI models pulling from outdated third-party articles, competitor content and review sites instead of its current product information. Using Senso, it launched a dedicated domain, fundmore.finance, on April 8, 2026. By April 20, 2026, that domain was one of the top 10 most-cited external sources across its tracked questions, with 75 citations and a 60% average share of voice (FundMore.ai case study).
Sources: OpenAI, Why language models hallucinate, September 2025. Senso documentation, Why Senso. Senso documentation, Verification Loop.
Next step: See how a page becomes a Verified Source with Senso by Zeta.
Published by Senso by Zeta, the GEO platform from Zeta Global.
