Best Diet for Type 2 Diabetes: The Complete Beginner’s Guide (2026)

This article’s title asks which diet is best for type 2 diabetes. That question has a real answer, and the answer is small.

Fifty-six randomised trials covering 4,937 adults with type 2 diabetes compared nine named eating patterns: low-carbohydrate, Mediterranean, vegetarian, high-protein, moderate-carbohydrate, low-fat, low glycaemic index, Palaeolithic, and usual care. Every one of them lowered HbA1c. The reductions ranged from 0.47 to 0.82 percentage points.

Read that range again, because it is the whole prize. The gap between the best-performing named diet and the worst-performing named diet is roughly a third of a percentage point of HbA1c. Choosing correctly between them matters, and it matters about that much.

Meanwhile, in a different trial, the proportion of people whose diabetes went into remission ranged from 5% to 70% — and the thing that separated 5% from 70% was not which foods they ate. It was how many kilograms they lost.

That is the honest shape of this subject, and it is almost the opposite of how pages like this one are usually written. The question in the title has a modest answer. The question next to it has a very large one. This guide answers both, in that order, and spends most of its length on the second.

It also keeps what the previous version of this page got right — the plate method, the food lists, the practical structure — because a correct answer you cannot follow is worth nothing, and a plate is easier to follow than a percentage.

The Nine Diets, Ranked — and How Little Separates Them

The ranking below comes from a network meta-analysis, which is the design that lets trials of different diets be compared indirectly when few studies test two diets head to head. SUCRA is a probability score for how likely a treatment is to rank among the best; it is a ranking tool, not an effect size, and a high SUCRA on a crowded field can sit on a small difference.

Two things in that table deserve more attention than the rankings.

First, the top two swap places depending on which number you measure. Low-carbohydrate ranked first for HbA1c; Mediterranean ranked first for fasting glucose, and the authors concluded in favour of Mediterranean overall despite low-carbohydrate winning the HbA1c ranking. When a ranking reverses on a different outcome from the same dataset, the honest reading is that the two are close, not that one is the answer.

Second, the bottom row is the real finding. Every named pattern worked. The band is 0.47 to 0.82 points. If a page tells you one of these diets is the diet for diabetes, it is converting a third of a percentage point into a headline.

There is a newer analysis worth knowing about, and it is odd in an instructive way. A 2025 network meta-analysis of 18 trials and 1,687 patients compared twelve nutritional interventions and ranked, in first place for HbA1c, a digital dietary model (SUCRA 84.6%; standardised mean difference −1.06, 95% CI −2.11 to −0.01), with medical nutrition therapy behind it. Neither of those is a diet. They are ways of delivering dietary advice.

That result should be read cautiously — the confidence interval only just excludes zero, the paper reports standardised rather than real-world units, and two other interventions had larger pairwise effects than the top-ranked one without the paper explaining why. But it points at something the whole literature keeps suggesting: how the advice reaches you may matter as much as which pattern it describes. That is also what the 2026 ADA standards act on, with a grade-A recommendation to refer people to a registered dietitian nutritionist rather than a recommendation to follow any particular diet.

The Bigger Number: Remission, and What It Formally Means

The previous version of this page never used the word remission. That is the largest omission in it, because remission is the outcome with the big effect size attached — and because it has had an agreed definition since 2021, which means it is a target you can check rather than a claim you have to trust.

In August 2021 the American Diabetes Association, the Endocrine Society, the European Association for the Study of Diabetes and Diabetes UK issued a joint consensus report defining it.

Two features of that definition carry most of its practical weight.

It requires being off glucose-lowering medication. An HbA1c of 6.2% on metformin is a good result and it is not remission, and the distinction is not pedantry — it changes what the number tells you about the underlying condition.

And it requires rechecking every year, which tells you what the panel expected: that remission is a state you can leave. The word chosen was remission, not cure and not reversal. The sections below show why that word was the right one.

What Remission Actually Looks Like Over Five Years

The trial that produced the numbers is DiRECT, run in UK primary care. It matters here because almost everything written about diabetes remission online rests on it, usually quoting only its best year.

The design: 306 adults, randomised by GP practice rather than individually, aged 20 to 65, diagnosed within the previous six years, BMI 27 to 45, not on insulin. The intervention was a total diet replacement of 825 to 853 kcal/day for three to five months, then stepped food reintroduction, then monthly maintenance support, with a defined plan for regain. Control was usual care.

Read down that first column and you have the honest story of dietary remission: real, large, and decaying. Forty-six per cent at one year is a genuinely remarkable result for a dietary intervention. Ten per cent at five years is what was left of it.

The row most people never see is the fourth one. Measuring remission as a yes-or-no state on one day hides what actually happened: the intervention group spent 27% of five years in remission against 4% for controls. Years without diabetes are the thing being bought, and a single-year snapshot undercounts them in both directions.

And then the row that changes how the whole page should be read: serious adverse events were lower in the intervention group — 15 against 25 over two years, and 9 against 22 in the second year alone. The trial’s own investigators reported that the weight-loss group had less than half the serious health problems of the usual-care group, whether or not they were in remission. The control arm’s events included two strokes, a toe amputation, a ruptured aortic aneurysm and a sudden death; the intervention arm had one non-fatal heart attack.

So the benefit was not confined to the people who got the headline outcome. That is the single most useful sentence on this page for anyone who loses weight, improves, and still needs their medication.

Remission Is a Function of Kilograms, Not of Food Choices

This is the table that answers the title’s question better than any diet ranking does. It is the 24-month remission rate in DiRECT sorted by how much weight people had actually lost — across the whole trial population, intervention and control together.

That is a dose-response, and it is a steep one: roughly a thirteen-fold difference between the bottom band and the top. Nothing in the diet-ranking table comes close to that spread. The amount of weight lost is the variable that moves remission; the eating pattern is the vehicle you use to lose it.

Two honest caveats, because this is the most quoted figure in the field and the most often quoted without them.

This analysis was post hoc — the trial was not designed to test weight bands, and the groups were formed after the fact by how much weight people happened to lose. It describes an association within a trial, not a randomised comparison of losing 5 kg against losing 15 kg. Nobody was assigned to a band.

And the bands were hard to reach and harder to hold. In the intervention group, 24.2% achieved a loss of 15 kg or more at 12 months. At 24 months only 11.4% still had it, and 24.2% were holding 10 kg or more. Among those who did hold 10 kg or more, 64% were in remission. At five years, the people still in remission had kept an average of 8.9 kg against 6.1 kg for the supported group as a whole.

So the practical reading is not “lose 15 kg.” It is this: the first ten kilograms are where the remission curve turns steep, and keeping them is the whole task. Which eating pattern gets you there is a question about you — your cooking, your household, your budget, what you can still be eating in two years — and that is exactly why the guidelines decline to name one.

The 2026 Numbers That Carry an Evidence Grade

Most diabetes-diet advice online is ungraded. The ADA’s Standards of Care attach a letter to each recommendation — A for the strongest evidence, then B, C and E for expert consensus — which lets you see which figures are supported by trials and which are judgement calls. These are the ones that bear on eating.

Four of those deserve comment.

The fiber figure is not 25 to 38 grams. The previous version of this page gave that range, which is a general adult nutrition target lifted from population guidance. The diabetes standard is relative to how much you eat: at least 14 g per 1,000 kcal. On an 1,800-calorie intake that is about 25 g; on 2,500 calories it is about 35 g. The generic range brackets the right answer by accident, and the anchored version tells you something the range cannot — that fiber scales with the rest of the plate.

Carbohydrate reduction is a grade B “consider,” not a prescription. The standards do not set a carbohydrate number for type 2 diabetes, and they point at limiting processed foods as an effective way to reduce intake — which is a different instruction from counting grams.

The 800 to 1,000 kcal plans come with a warning attached, and that warning applies to DiRECT. The trial’s own intervention was 825 to 853 kcal/day, delivered in primary care with structured medical support and a protocol for restarting medication. Reading about 46% remission and attempting that unsupervised is the single most dangerous misreading available on this subject, which is why it has its own section below.

And the only grade-A instruction about who should design your eating plan names a person, not a diet. For a page titled “best diet,” that is worth sitting with.

The Plate Method, Kept — and What It Is Actually Doing

The plate method survives this rebuild unchanged in its proportions, because it is the most useful thing on the original page. What it needed was not correction but an explanation of its mechanism, since knowing what each section is for is what lets you adapt it to food the method was never written for.

One honest note about the method itself: it is a portion-control heuristic, not an intervention with its own trial evidence. No network meta-analysis ranks “the plate method” against Mediterranean or low-carbohydrate eating, because it is not that kind of thing. What it has going for it is that it produces a meal consistent with the graded eating-pattern recommendation — vegetables, legumes, lean protein, whole grains, nuts and seeds, with refined grains and ultra-processed food crowded out — while requiring no numbers, no app and no weighing.

For a beginner that is worth more than a better-ranked diet, for a reason the whole page keeps circling: the pattern you can still follow in two years beats the pattern that scored a tenth of a point higher in a twelve-week trial. Most of the trials behind the rankings were short. Your diabetes is not.

What Each Pattern Asks of You

Since the measured differences between patterns are small, the sensible basis for choosing is what each one costs you to run. This table is the one the title’s question should have led to.

The bottom row is not a platitude. The diet rankings come from trials that mostly ran weeks to months; the remission data comes from a trial that followed people for five years and watched most of the effect decay as weight returned. Across both, duration is the variable that separates a result from a headline — and duration is the one thing no meta-analysis can choose on your behalf.

If you want a single starting recommendation: begin with the plate method, built out of Mediterranean-style ingredients, and get a referral to a registered dietitian nutritionist — the one grade-A instruction about how your plan should be made. That combination is unexciting and it is what the evidence actually supports.

What This Page Got Wrong, and What It Was Missing

The previous version of this guide was not misleading. Its food lists were sound, its plate method was right, and nothing in it would have harmed a reader. What it had were gaps of a particular kind: the figures it gave were real but unanchored, and the outcomes with the largest evidence behind them were absent.

The fiber figure was generic. 25 to 38 grams a day is population nutrition guidance. The diabetes-specific standard is at least 14 g per 1,000 kcal, which ties fiber to intake instead of handing every reader the same number.

Remission was never mentioned. The word did not appear. An article about eating for type 2 diabetes that omits the outcome with a formal consensus definition, a five-year trial behind it and a thirteen-fold dose-response across weight bands has left out the most consequential thing in its own subject.

Nobody was told to see a dietitian. Referral to a registered dietitian nutritionist is a grade-A recommendation in the 2026 standards — stronger evidence than any single eating pattern carries. It was not in the article, and “see a doctor if symptoms worsen” is not the same instruction.

No named eating pattern appeared at all, despite the title promising the best diet. Not naming one is defensible; what is not defensible is promising a comparison and never making it, which leaves the reader to assume one was being withheld.

The weight-loss figure stopped at 5 to 10%. That matched the guidelines as far as it went, but the standards say above 10% brings greater benefit including possible remission, and the trial data show the curve turning steeply somewhere around ten kilograms.

And two housekeeping faults worth fixing on the live page whatever else changes. The reference list contains a National Institutes of Health entry with no URL at all — an institution named as a source with nothing to check, which is the same pattern that appears in several other articles on this site. And two internal links to other articles on this site are listed under the References heading, where a reader reasonably expects independent sources.

When This Is Not a Diet Problem

This section exists because of a specific, foreseeable harm. The 46% remission figure travels further than the conditions attached to it, and someone reading it will try the diet behind it on their own.

The DiRECT intervention was 825 to 853 kcal a day, and it was delivered with medical supervision. The 2026 standards grade very-low-calorie plans of 800 to 1,000 kcal/day as appropriate only for carefully selected people with close monitoring. The trial had a protocol for stopping and restarting glucose-lowering medication, and 39.5% of the intervention group were back on medication at two years by that protocol. Attempting that eating pattern while continuing your usual doses is how people end up in hospital, not in remission.

Reducing carbohydrate while taking insulin or a sulfonylurea can cause hypoglycaemia. Those medications lower blood glucose by a fixed mechanism that does not know you have changed what you eat. If you are on either, the dose conversation comes before the dietary change. This is the most common serious mistake in self-directed diabetes dieting and it is entirely preventable.

The trial population was narrower than most readers. Aged 20 to 65, diagnosed within six years, BMI 27 to 45, not on insulin. If you fall outside that — longer duration, insulin-treated, a different BMI — the remission figures were not measured in people like you. They may still apply. Nobody has shown that they do.

Good control on medication is a success, not a failure. The strongest single finding in the five-year data is that the weight-loss group had less than half the serious health problems of the usual-care group whether or not they reached remission. If you lose ten kilograms, improve your HbA1c and still need metformin, you have bought most of what was on offer. The remission label is a diagnostic category, not the prize.

And if food has stopped being a practical question, that is the thing to raise. Diabetes puts eating under permanent evaluation, and for some people that turns into counting that will not stop, fear of eating in front of others, or distress that is not about blood glucose any more. The 2026 standards tell clinicians to screen for diabetes distress and to refer to behavioural health professionals when they find it. If that describes you, say so at your next appointment — it is a recognised part of diabetes care, and a stricter meal plan will make it worse rather than better.

Frequently Asked Questions

So what is the best diet for type 2 diabetes? Across 56 trials and 4,937 adults, low-carbohydrate ranked first for HbA1c and Mediterranean first for fasting glucose, and every named pattern lowered HbA1c by 0.47 to 0.82 percentage points. The spread between them is about a third of a point. The best diet is therefore the one you will still be eating in two years, and the 2026 standards reflect that by naming a person to help you build it — a registered dietitian nutritionist, grade A — rather than naming a diet.

Can type 2 diabetes be reversed by diet? It can go into remission, which has an agreed definition: HbA1c below 6.5% sustained for at least three months without glucose-lowering medication, rechecked yearly. In DiRECT, 46% of the intervention group reached it at one year, 36% at two years and 10% at five. The word the consensus panel chose was remission, not cure, and the five-year figures show why.

How much weight do I need to lose? The standards set at least 5–7% as the goal and say above 10% brings greater benefit including possible remission. In DiRECT, 24-month remission ran at 5.2% among those who had lost under 5 kg, 28.8% at 5–10 kg, 60.0% at 10–15 kg and 70.0% above 15 kg. Those bands are a post hoc analysis, not a randomised comparison, but the direction is unambiguous: the first ten kilograms are where the curve turns.

Can I eat fruit? Yes. Whole fruit is on the list of foods the 2026 standards tell you to emphasise, alongside non-starchy vegetables, legumes, whole grains, nuts and seeds. Fruit juice is not — the recommendation to replace sugar-sweetened beverages names juice explicitly. Whole fruit sits in the carbohydrate quarter of the plate; it is portioned, not avoided.

How much fiber should I eat? At least 14 grams per 1,000 calories, which is the diabetes-specific standard. On an 1,800-calorie intake that is about 25 g; at 2,500 calories, about 35 g. From minimally processed, nutrient-dense sources rather than a supplement.

Do I have to count carbohydrates? Not for type 2 diabetes managed without fixed insulin doses. The standards suggest considering reduced carbohydrate intake for some adults (grade B) and point at limiting processed foods as an effective way to do it. Consistent carbohydrate timing and amount is recommended specifically for people on fixed insulin doses.

Is the low-calorie diet from the news something I can just start? No. The trial used 825–853 kcal/day with medical supervision and a protocol for adjusting medication, and the standards grade plans of 800–1,000 kcal/day as suitable only for carefully selected people under close monitoring. If it interests you — and the results justify the interest — the route is asking whether you are a candidate, not starting on your own.

What if I do everything right and still need medication? Then you are in the position most of the DiRECT intervention group ended up in, and the data say you still gained most of the benefit. Serious adverse events over the first two years were 15 in the intervention group against 25 in usual care, and the investigators reported fewer than half as many serious health problems in the weight-loss group whether or not remission was achieved.

The Bottom Line

This guide opened with the answer to its own title: a third of a percentage point of HbA1c is what separates the best-ranked eating pattern from the worst-ranked one. That number has been sitting there the whole way down, and it is worth turning over now that the rest is in view.

A third of a point is small enough that the choice is genuinely yours. Not in the evasive sense that pages like this one usually mean, but literally: the evidence does not have a strong preference between Mediterranean, low-carbohydrate, vegetarian and the rest, so the deciding factors can be the ones only you know — what you can cook, what your household will eat, what your budget carries, what you will not resent in a year. Choosing on those grounds is not settling for second best. It is using the only information the trials could not supply.

The number that is not small is the one attached to kilograms. Remission at two years ran from 5.2% in people who had lost under 5 kg to 70% in those who had lost over 15 kg — a dose-response no eating pattern comes near. That is where the effort belongs, and it is also the part the guidelines say should not be attempted alone: at least 5–7% as the goal, above 10% for greater benefit, sixteen or more counselling sessions in six months, and a referral to a dietitian as the single strongest recommendation about how any of this gets designed.

And then the finding that should be read last, because it changes what counts as success.

Five years after DiRECT began, 10% of the intervention group were still in remission. Most were not. They had regained weight, most were back on medication, and by the trial’s primary outcome they had not kept what they gained. Their serious health problems over that period still ran at less than half the usual-care group’s — in remission or not. Fifteen serious adverse events against twenty-five across two years, nine against twenty-two in the second year alone. The control arm’s list included two strokes, an amputation, a ruptured aneurysm and a sudden death.

So the honest closing claim of this page is not that the right diet reverses diabetes. It is that the weight you lose buys protection that does not expire when the remission does — and that this holds whichever of the nine patterns you used to lose it.

This article is for general informational purposes and is not a substitute for individual medical advice. Do not change or stop any diabetes medication on the basis of an article. If you take insulin or a sulfonylurea, talk to your doctor before reducing carbohydrate intake. Very-low-calorie diets require medical supervision.

References

• Schwingshackl L, et al. A network meta-analysis on the comparative efficacy of different dietary approaches on glycaemic control in patients with type 2 diabetes mellitus. European Journal of Epidemiology, 2018 — 56 trials, 4,937 adults, nine dietary patterns
• Effects of 12 nutritional interventions on type 2 diabetes: a systematic review with network meta-analysis of randomized trials. Nutrition & Metabolism, 2025
• Lean MEJ, et al. Durability of a primary care-led weight-management intervention for remission of type 2 diabetes: 2-year results of the DiRECT open-label, cluster-randomised trial. The Lancet Diabetes & Endocrinology
• 5-year follow-up of the randomised Diabetes Remission Clinical Trial (DiRECT) — Diabetes UK summary of the extension study
• 5-year follow-up of the randomised Diabetes Remission Clinical Trial (DiRECT) of continued support for weight loss maintenance in the UK: an extension study. The Lancet Diabetes & Endocrinology
• Riddle MC, et al. Consensus Report: Definition and Interpretation of Remission in Type 2 Diabetes — ADA, Endocrine Society, EASD and Diabetes UK, 2021 (announcement of criteria)
• Consensus report: definition and interpretation of remission in type 2 diabetes — PubMed record
• Standards of Care in Diabetes — 2026, American Diabetes Association: graded recommendations summary
• The American Diabetes Association Releases “Standards of Care in Diabetes — 2026”
• 5. Facilitating Positive Health Behaviors and Well-being to Improve Health Outcomes: Standards of Care in Diabetes — 2026. Diabetes Care 49(Supplement_1):S89
• Summary of Revisions: Standards of Care in Diabetes — 2026. Diabetes Care 49(Supplement_1):S6
• American Diabetes Association — Food and Nutrition
• CDC — Type 2 Diabetes Management