Professional · Article 23

Dietary & Alternative Baking

A reference article · 3,800 words · 78 cited sources · last reviewed October 2026 · fact-checked October 2026, 45 claims

Nearly every failed substitution starts the same way, by treating an ingredient as one thing. Flour, egg, and sugar are each doing five or six jobs simultaneously. Take one out and you have not made one hole, you have made a bundle of them.

Usage rates are given as percentages of flour weight where the literature supports it, with volume equivalents in parentheses. Two claims in this article are flagged inline as directional rather than settled.

The governing principle

Every dietary adaptation is a function-replacement problem. A conventional ingredient is never doing one job, and it is never doing it alone.

Wheat flour supplies structure, but its gluten also traps fermentation gas, retains moisture, and contributes to browning. A whole egg binds, but it also aerates, emulsifies, moistens, enriches, and sets on heating (Sahi, 2021; Figoni, 2011). Sucrose sweetens, but it also tenderises, holds water, promotes spread, feeds browning, and depresses the freezing point (BAKERpedia, n.d.-d).

The recurring failure of naive substitution, swapping almond flour one-for-one for wheat, is that it treats the ingredient as monolithic. The professional move is to break the ingredient into its functional roles, then source each role from the best-matched replacer, accepting that one ingredient out often means three or four ingredients in.

Why gluten is hard to replace

Wheat’s behaviour comes from two storage proteins. Glutenin supplies elasticity, the resistance and spring. Gliadin supplies viscosity and extensibility, the flow and stretch. Hydrated and worked, they link through disulfide bonds and weaker interactions into gluten: a continuous viscoelastic network that holds expanding gas cells as they stretch, with thin liquid films at the bubble surfaces helping to stabilise them late in proof, when the gluten matrix itself begins to break up (Wieser, 2007; Gan et al., 1995).

No other common cereal protein does this. Remove gluten and you lose the one ingredient that provides structure and gas retention at the same time. Every characteristic gluten-free defect follows from that loss, compounded by the low protein and high starch of the replacement flours: no gas-trapping film, so batters rise and collapse; no cohesive matrix, so crumb fractures; starch-heavy systems that retrograde quickly, so staling is fast (Gallagher et al., 2004; Culetu et al., 2021).

The three-tier flour system

Contemporary gluten-free baking abandons the search for a single wheat substitute in favour of an engineered system with three tiers.

Tier 1, starches, for lightness and tenderness. White rice, tapioca, potato starch, and cornstarch are low in protein and high in gelatinising starch (Hager et al., 2012). They lighten texture and set structure, but alone they bake gummy and flavourless (America’s Test Kitchen, 2014). Tapioca adds chew and browning, potato starch adds moisture retention, cornstarch adds tenderness (Bob’s Red Mill, n.d.-a).

Tier 2, whole-grain and protein flours, for flavour and body. Sorghum, teff, buckwheat, brown rice, oat, and millet supply the flavour, colour, fibre, and protein that pure starch lacks (Hager et al., 2012). Sorghum and brown rice are neutral workhorses; teff and buckwheat are assertive and mineral-rich; oat flour adds soft moisture. When baking for celiac disease, buy every one of these flours labelled gluten-free, oats above all, because naturally gluten-free flours are often contaminated in milling (Thompson et al., 2010; 21 C.F.R. § 101.91, 2020). These contribute some structure through their own proteins and fibre, but never an elastic network.

Tier 3, hydrocolloids, the network mimic. This is the functional heart of the system. Since no flour supplies a viscoelastic film, one is engineered from long-chain polysaccharides that hydrate into viscous, film-forming gels. They do gluten’s two key jobs: raising batter viscosity so gas is trapped, and binding particles so crumb coheres (Culetu et al., 2021; Fetriyuna et al., 2026).

Xanthan gum, a microbial polysaccharide from Xanthomonas campestris, is cold-soluble and hydrates into a highly viscous, shear-thinning solution, though it clumps unless dispersed through the dry ingredients first (Fetriyuna et al., 2026). It thickens the aqueous phase, stabilises gas bubbles, and provides cohesion, which makes it the general-purpose choice. Overdose it and you get a heavy, gummy, slick crumb, which is one of the most common gluten-free failures (Torres-Pérez et al., 2024).

Psyllium husk, the mucilaginous seed husk of Plantago ovata, is extraordinarily hydrophilic. Hydrated, it forms a cohesive, extensible gel that is uniquely valuable in gluten-free yeast bread, because it produces something closer to real dough handling. It holds water, gives the batter enough body to shape, and retains fermentation gas, and its water-holding slows staling: Fratelli et al. (2021) measured crumb 65 to 75% softer than a psyllium-free control across three days of storage, with the slowest staling at the highest dose (Belorio & Gómez, 2022; Culetu et al., 2021). The dose matters in both directions. The 3 to 6% working range comes from Modernist Bread (Myhrvold & Migoya, 2017); published trials run from about 2% (Ren et al., 2020) to 17% of flour weight (Fratelli et al., 2021), and more psyllium brings more gumminess and crust hardness along with the structure (Torres-Pérez et al., 2024).

Guar gum, a galactomannan from the guar bean, hydrates cold to high viscosity and is used at slightly higher levels than xanthan (Tahmouzi et al., 2023; Bob’s Red Mill, 2010). It performs well in cold or unbaked applications, and some bakers find it less prone to the slick mouthfeel of excess xanthan.

HPMC, or hydroxypropyl methylcellulose, is a modified cellulose with an unusual defining property: thermally reversible gelation. It gels on heating and liquefies on cooling, the opposite of most hydrocolloids. In gluten-free bread that is powerful, because as the loaf heats, HPMC sets a structure that stabilises expanding gas cells through oven spring, then relaxes as it cools (Fetriyuna et al., 2026). It appears mostly in commercial formulas, where oven-set gas retention is the limiting factor.

Three further levers

Higher hydration. Gluten-free flours and starches absorb water differently from wheat, and the finished product needs water both to gelatinise starch and to keep the crumb from turning dry and sandy. These batters run wetter than wheat doughs, pourable or pipeable rather than kneadable. Adequate hydration also gives the hydrocolloids the water they need to form their films at all (Ren et al., 2020).

Added protein, usually egg. Egg is the unsung structural agent here. Its proteins coagulate on heating to set a matrix, the white adds structure and lift, and the whole egg emulsifies and enriches. Many successful gluten-free cakes carry a higher egg load than their wheat equivalents precisely because egg coagulation partly substitutes for the missing scaffold (King Arthur Baking Company, 2015). This is also why recipes that are simultaneously gluten-free and vegan are the hardest case in this article: they remove both structural systems at once.

Mixing to disperse, not develop. With no elastic film to build, mixing aims to hydrate and distribute. Overmixing does not build strength, it just levels the batter. A brief rest before baking improves hydration of the flours and gives psyllium’s slow-swelling mucilage time to take up water (America’s Test Kitchen, n.d.-b; Ren et al., 2020).

What eggs actually do

Vegan baking removes the animal-derived ingredients carrying some of the most functionally dense roles in the kitchen. The disciplined approach is to tabulate the jobs and replace per job (Sahi, 2021).

The table makes the lesson explicit: no single vegan replacer covers all six functions (Halm et al., 2024; King Arthur Baking Company, 2021). A binding egg in a cookie is a different problem from an aerating egg in an angel cake, and they take different answers.

Aquafaba

Aquafaba, from the Latin for bean water, is the viscous liquid from cooking or canning chickpeas. Its capacity to whip into a stable, egg-white-like foam comes from what the beans leach during cooking: soluble, mostly low-molecular-weight proteins; saponins, which are surface-active glycosides; and solubilised starches and carbohydrates (He et al., 2021; Stantiall et al., 2018).

Each component does a job. The proteins provide the film surrounding air bubbles. The saponins are natural surfactants that lower surface tension and promote foaming. The leached carbohydrates raise viscosity, which stabilises the foam and slows drainage. Together they reproduce the foaming and emulsifying behaviour of egg white (He et al., 2021).

Whipped with an acid stabiliser and sugar, it produces foams for meringue, macarons, mousse, and nougat (Fuentes Choya et al., 2023). Unwhipped, it emulsifies and enriches batters (Mustafa et al., 2018). Its limits are real: the foams hold up less well to heat than egg white, batch composition varies with bean variety and cooking time and dilution, and it contributes no protein-set structure comparable to a baked custard.

Matching the other replacers to the job

Flax and chia eggs work by mucilage, soluble fibre that hydrates into a viscous, sticky gel (Puligundla & Lim, 2022). One tablespoon of ground flaxseed or whole chia in three tablespoons of water, rested until gelled, is an excellent binder for cookies, muffins, pancakes, and dense quick breads (King Arthur Baking Company, 2021). In a batter they do not aerate, so they cannot lighten a cake or hold a whipped foam, and they add flecks and a nutty note.

Commercial egg replacers are typically starches, usually potato or tapioca, plus leavening and often a hydrocolloid; many now add plant protein, and a few add emulsifiers such as mono- and diglycerides. They bind and provide lift in structured baked goods, since that is what they are engineered for, but none reproduces egg’s coagulating protein set or makes a meringue-type foam, and the starch-only types contribute no fat or emulsification (Halm et al., 2024).

Applesauce supplies moisture and some binding along with bulk and mild sweetness. Too much makes the crumb dense and gummy, and it adds no leavening or structure. Banana does the same with assertive flavour, so it only works where banana is welcome. Silken tofu brings moisture, richness, protein, and some binding, which suits dense fudgy items like brownies and cheesecakes, but it is heavy and non-aerating (King Arthur Baking Company, 2021).

Dairy, and the lamination problem

Plant milks differ from cow’s milk in protein and fat, and those differences show up in the bake. Cow’s milk contributes lactose, which browns through Maillard reactions, plus milk proteins for browning and structure and fat for tenderness (BAKERpedia, n.d.-c). Most plant milks carry less protein, so baked goods brown less and set differently; sugar is less predictable, since oat and rice milks pick up maltose and glucose when their starch is hydrolysed. Soy and pea milks, at roughly 7 to 10 g of protein per cup, behave closest to dairy; oat sits around 3 g, and almond and rice, at about 1 g or less, least so (University of Utah Health, 2023; U.S. Department of Agriculture, Agricultural Research Service, 2019).

Vegan butter is where lamination gets difficult. Margarine-type fats, which come in dedicated puff-pastry grades, are at least 80% fat (BAKERpedia, n.d.-b). Vegan butter is a similar emulsion of plant oils and water, but its water content and plasticity range differ from dairy butter, and formulations vary widely from brand to brand. Successful croissants depend on butter staying pliable but solid across a narrow temperature window so it rolls into continuous unbroken layers. Many vegan butters are either too soft, merging into the dough, or too brittle, shattering and puncturing the layers (Brod & Taylor, n.d.). Higher-fat block versions formulated for lamination perform best, but the margin is narrower than with dairy butter.

For cream, coconut cream whips once it has been thoroughly chilled so that its highly saturated fat is solid, the same requirement dairy cream has, where solid fat around the air bubbles is what holds the foam (McGee, 2004; America’s Test Kitchen, n.d.-a). Cashew cream provides rich emulsified body for fillings. Aquafaba whips into light toppings, though it holds less well than dairy cream.

Sugar does at least seven jobs

Sucrose is the most functionally underestimated ingredient in the pantry. Beyond sweetness it retains moisture by binding water, which keeps crumb soft and extends shelf life. It tenderises by competing with flour proteins and starch for available water, limiting gluten development and starch gelatinisation. It aerates, because in the creaming method the sharp edges of granulated crystals cut air pockets into solid fat, nucleating the bubbles that leavening later expands. It browns: it caramelises, and once heat and acid split part of it into glucose and fructose, those reducing sugars feed Maillard reactions. It depresses the freezing point, which matters in frozen work. It preserves, by lowering water activity. And it promotes spread, melting and dissolving during the bake to govern how far a cookie flows (BAKERpedia, n.d.-d; Figoni, 2011).

Because sugar does all seven, you cannot remove it or swap it one-for-one without affecting structure, browning, moisture, and spread at once. That is the source of most reduced-sugar disappointment.

Allulose is a rare monosaccharide roughly 70% as sweet as sucrose with minimal glycaemic impact (Chung et al., 2012; Tani et al., 2023). It is functionally the closest alternative, because it bulks like sugar, dissolves like sugar, and browns through the Maillard reaction. It browns faster than sucrose, so formulas usually need a lower temperature or a shorter bake to avoid over-colouring (Lee et al., 2020). The FDA said in 2020 that it would not object to allulose being left out of the total and added sugars lines on the Nutrition Facts label, though it still counts toward total carbohydrate (U.S. Food and Drug Administration, 2020). That is an enforcement position, not a change to the rule: in 2026 a federal appeals court held that allulose is a sugar under the labeling regulation, with the FDA itself arguing as much in its brief, so a “sugar-free” claim on an allulose product is now open to challenge (Franco v. Chobani, LLC, 2026).

Erythritol is a polyol about 60 to 70% as sweet as sucrose, largely non-glycaemic (Regnat et al., 2018). Two behaviours define it in baking: a pronounced cooling sensation on the palate, from its negative heat of solution, and limited solubility, so at high replacement levels it can recrystallise as products cool and turn gritty (Regnat et al., 2018; BAKERpedia, n.d.-a). It does not brown, and it can crystallise out of frostings. At partial replacement the penalty is small: in one Danish-cookie trial, cookies grew lighter in colour as erythritol rose, but hardness and moistness did not change even at full replacement, and tasters liked cookies with up to 50% erythritol as well as the all-sucrose control (Lin et al., 2010).

Xylitol is nearly as sweet as sucrose and provides real bulk and some moisture retention, though like erythritol it is not a reducing sugar and does not brown, and it can still crystallise (BAKERpedia, n.d.-e; American Society of Baking, n.d.-b). Note that xylitol is highly toxic to dogs (Murphy & Dunayer, 2018; Xia et al., 2009).

Stevia and monk fruit are high-intensity sweeteners, hundreds of times sweeter than sugar and used in trace amounts (Orellana-Paucar, 2023). They contribute no bulk, no moisture, no browning, and no spread, so they leave large functional gaps and cannot stand alone. They must be paired with a bulking agent. Commercial one-to-one baking blends are mostly erythritol or allulose, or in some brands maltodextrin, carrying a trace of the intense sweetener, which is worth knowing before you buy one.

Keto and low-carb

Keto baking removes wheat flour and most sugar simultaneously, so it inherits the whole gluten-free structural problem plus the sugar-function problem, and it solves the flour side with two flours that behave nothing like wheat.

Almond flour is high in fat and protein, with minimal starch and no gluten. It gives richness, moisture, and a tender slightly dense crumb, and browns readily thanks to its protein and fat. It forms no network and no continuous structure, so it produces flat, fragile, crumbly results unless structure comes from elsewhere, typically eggs and a binder. By weight it runs close to wheat flour, about 80 to 90 g per 100 g; by volume it cannot be swapped cup-for-cup, since a cup weighs about 96 g against 120 g for all-purpose flour, and its high fat and lack of starch change how it behaves (Almond Board of California, 2025; King Arthur Baking Company, n.d.).

Coconut flour is the defatted, dried, ground residue of coconut meat, and its defining property is extreme absorbency (American Society of Baking, n.d.-a). Its high fibre content, which in isolated form holds many times its own weight in water (Zheng et al., 2018), soaks up several times its weight in liquid, so a little goes a long way, and formulas must supply large amounts of liquid and characteristically many eggs. Coconut-flour cakes commonly use six or more eggs per cup of flour, supplying both the moisture and the coagulated protein structure the flour cannot (Arrowhead Mills, n.d.). Used carelessly it bakes dry, dense, and sandy.

Since neither flour builds a network, keto bakers borrow the gluten-free toolkit for binding and gas retention, most often psyllium husk and sometimes xanthan, alongside heavy reliance on egg (Rao et al., 2022).

The portable framework

Any adaptation, however unfamiliar, works with the same five steps.

One, disaggregate the ingredient into functions. List every job it does in this formula: structure, aeration, moisture, tenderising, browning, binding, emulsification, coagulation, preservation, spread.

Two, rank those functions by importance to the result. In an angel cake, aeration and set dominate. In a shortbread, tenderness and spread dominate. In an artisan loaf, gas retention and structure dominate. Replace the load-bearing functions first.

Three, source each function from its best-matched replacer, accepting that one ingredient out may need several in.

Four, rebalance the whole system. Substitutions shift hydration, pH, fat, and set temperature. Expect to adjust liquid, since gluten-free and coconut flour both need more; leavening, to compensate for lost creaming or foam; time and temperature, since allulose and almond flour brown faster; and mixing method, since gluten-free is a batter.

Five, read the failure mode diagnostically. Crumbly usually points to insufficient binding. Gummy usually points to an under-bake, or to excess hydrocolloid or fruit purée. Pale points to a missing browning substrate, sugar or protein. Dense points to lost aeration, though too much fat or too little liquid or egg produce it too. Dry and fast-staling points to lost humectancy. Each defect narrows the search to a few unfilled functions (America’s Test Kitchen, n.d.-b; King Arthur Baking Company, 2021).

The baker who internalises this stops asking what replaces butter, egg, sugar, or flour, and starts asking which of that ingredient’s jobs matter here, and what the best tool is for each. Function first, ingredient second, is the whole discipline.