How we score recipes

What the Cook Book score is (and isn’t)

The Cook Book score is an AI assessment of how sound, complete and clearly written a recipe is. It checks our recipe against a fixed checklist, then compares it with top-rated versions of the same dish on the web.

It is not a taste test. Nobody has eaten the dish to score it, and it is not a review. It tells you how much you can rely on the recipe as written: are the amounts right, is the method clear, are the times and temperatures realistic?

A score is out of 5, to one decimal place. We only show it when it is 3.5 or more. Recipes below that go on our list to rewrite first.

1. The technical check (60% of the score)

Each recipe is scored from 0 to 5 on five points, with these weights:

• Clear method (25%): Steps are in a workable order and each does one job; every listed ingredient is used in the method, and named consistently; steps give heat level, pan size or visual cues ("until golden", "coats the back of a spoon"); safety points are covered where needed (flambéing, hot sugar, deep fat, food-safe cooking of poultry and mince).
• Ingredient completeness (20%): Everything the method uses is in the ingredient list with an amount and unit; nothing in the list is left unused; ingredients are specific enough to buy (cut, fat %, type of flour or sugar).
• Sound ratios and seasoning (20%): Key ratios are in a normal working range (roux to liquid, flour to raising agent, meat to sauce, sugar to fat); portion sizes match the stated number of servings; seasoning is addressed at sensible points.
• Realistic timings and temperatures (20%): Cooking and resting times are given and are realistic; oven temperatures are given (conventional and fan, and converted for US readers); doneness can be judged (time plus a visual cue or a core temperature).
• No obvious errors (15%): No typos, shorthand or leftover card notes ("toms", "Demi"); no unit mistakes ("1 g onion", butter in ml); amounts, times and temperatures agree between the list, the method and the shopping list.

A known unit or quantity mistake (for example “1 g onion”) caps the score at 3.0 until it is fixed.

2. The web comparison (40% of the score)

We open well-rated versions of the same dish: ones that readers on the publisher’s own site rate 4 to 5 stars. We look first at BBC Good Food, Allrecipes, NYT Cooking, Serious Eats, delicious. and Jamie Oliver, and at recipes by Anne Willan, Marcella Hazan, Rick Stein, Heston Blumenthal, Gordon Ramsay, Laxmi Khurana, Christopher Kimball and Fallow. We use at least 3 versions. Versions with only a handful of ratings count for less.

We then compare ours with them on five points, each from 1 (far behind) through 4 (on a par) to 5 (better):

• Ingredients: Does ours have the ingredients the top versions agree on (or a stated reason not to)?
• Ratios and quantities: Are our quantities in line with theirs, scaled to the same number of servings?
• Technique: Do we use the techniques that make the top versions work (browning, deglazing, reducing, resting)?
• Timings and doneness: Are our times and temperatures in their range, with doneness cues as good as theirs?
• Detail and clarity of method: Is our method as clear and detailed as theirs (make-ahead, swaps, troubleshooting)?

We record every page we compare with: its address, its star rating and how many ratings it had on the day. We never make up ratings or reviews.

3. Putting it together

Cook Book score = 60% technical check + 40% web comparison. On the page it reads, for example: “Cook Book score 4.7/5: an AI check of method and balance, compared with 5 top-rated web versions: on a par, with more detail on method.”

When we improve a recipe, we score it again.

Reader ratings

Every recipe has a Rate this recipe row. Once 5 or more readers have rated it, the readers’ average becomes the headline and the Cook Book score drops to a small note. Before that we show “Tried it? Rate it.”

One vote per recipe per browser, and you can change your vote at any time. To stop the same person voting again and again, we keep a random code in a cookie on your browser and limit the votes coming from one internet connection. We keep only the stars: no name, no e-mail, no IP address (the connection check uses a scrambled code that changes every day and is deleted after 2 days).

What we don’t do

• We don’t publish the Cook Book score to Google as a star rating. Only real reader votes could ever appear there.
• We don’t pay for, invent or copy reviews.
• Steve’s pages are his own stories and are not scored or rated.