Mouthfeel is a profoundly complex sensory attribute, and that complexity is the primary reason there is no single, standardized test that perfectly correlates with consumer preference. Unlike a quantifiable property such as pH or particle size, mouthfeel is one combined experience made up of several physical and chemical sensations: viscosity, creaminess, astringency, grittiness, melting behavior, and even the sound of a crunch. A machine can measure flow rate or particle distribution, but it cannot capture the subjective, emotional, and often cultural factors that determine whether a consumer finds a texture pleasing. What feels "smooth" to one person might feel "slimy" to another, and preferences for mouthfeel vary widely across demographics, product categories, and intended use occasions.
Expectation shapes preference as much as sensation does. A gritty texture might be a defect in a protein shake but a desired characteristic in a chia-based pudding. A "thin" mouthfeel might be ideal for a sports drink but be perceived as "watery" and cheap in a ready-to-drink meal replacement. Any standardized test would need to account for these context-dependent expectations, which are notoriously difficult to model in a lab setting. Industry attempts, such as rheology (measuring deformation and flow) or texture profile analysis, provide useful engineering data, but they show only moderate correlation with consumer hedonic (pleasure) ratings. Standardized sensory methods do exist: trained descriptive panels working to a shared lexicon under ISO 11036 can score texture attributes with good repeatability. The practical limit is that a trained panel is slow and expensive to run on every reformulation, so formulators keep reaching for a faster instrumental shortcut. The gap persists because machines measure physical forces while humans judge desirability.
Could Such a Test Reduce Reformulation Time?
While a perfect, universal test is not feasible, the development of product-specific, predictive models would reduce reformulation time. Instead of chasing a single "mouthfeel score," manufacturers could benefit from a two-step approach:
- Objective physical characterization: Using tools like rheometers, viscometers, and particle analyzers to generate a "texture fingerprint" of a target product. This creates a baseline for a specific category (e.g., "a successful vanilla protein bar has a firmness range of X and a fracturability of Y").
- Targeted sensory panels: A trained panel that aligns descriptive terms (e.g., "grainy," "slippery," "dense") with the physical measurements. Once a correlation is established for a given product category, formulators can quickly adjust ingredients to hit the desired physical markers, cutting trial-and-error iterations.
In practice, a formulator working on a new gummy supplement could use such a model: if previous successful gummies required a specific gel strength (measured in grams of force), a new batch hitting that target is far more likely to have the preferred "chewy but not tough" mouthfeel. This eliminates weeks of making multiple batches and waiting for consumer tests. At KorNutra, we focus on establishing these product-specific correlations for our clients, ensuring that every batch of a new supplement formula is not only technically correct but also aligned with the sensory expectations of its target audience.
Machine learning prediction of mouthfeel
A parallel line of work skips the single-instrument question entirely. Instead of asking one machine to reproduce a human judgment, researchers train models on paired data: physical measurements on one side, panel or consumer scores on the other. The model learns the mapping the way a formulator does, across many examples rather than through one equation. No single equation captures it. A 2025 study from Purdue, published in Food Research International, built a sensory-biased autoencoder that links rheological properties of liquids to perceived texture, using sensory scores as a bias during training so the model's output stays anchored to human perception rather than drifting into pure physics. The aim is to let developers predict texture from a few physical readings instead of running full panels for every change.
Texture datasets are small and often not public, so models are only as good as the data behind them. For a gummy manufacturer, the highest-value move is still internal: log gel strength, moisture, and springiness alongside panel scores on every batch, category by category, and let the correlation library accumulate. The standardized test, if it arrives, will likely be a trained model over that kind of proprietary data rather than a single new instrument on a lab bench.
The absence of a single universal test reflects the deep, personal nature of texture perception. The smartest path forward is to build a library of tailored, objective parameters that serve as reliable proxies for consumer preference within each product category. This targeted approach is where the time-savings and formulation efficiency lie.