Using AI to Generate or Troubleshoot Artisan Bread Recipes—What’s Your Experience?

Toast

Hey fellow bread lovers,

I’ve been baking sourdough long enough to trust my hands... but lately curiosity got the better of me, and I asked an AI to help draft a bread recipe. It plopped out something that looked technically plausible—flour, water, starter, salt—plus, oddly, a comment that had me grinning: “If your loaf flattens on day two, blame the yeast, not the baker.”

It wasn’t perfect by any means—hydration was a bit off, timing felt tight—but it gave me a starting point I didn’t have to craft from scratch. Kind of neat, like having a sous-chef who knows code.

So I’m wondering:

Has anyone else tried using AI for baking recipes, whether with a full loaf or a tweak here and there?

Did it spark your creativity, inspire new ideas—or crash and burn spectacularly?

Curious to know if any of you let AI “draft” a recipe and then adapted it with your baker’s intuition?

Let’s share our funniest prompts, weirdest outcomes, or best AI-inspired tweaks. After all, even if the loaf isn’t perfect, the story behind it can be as delightfully crusty as a great sourdough.

— A baker who codes and codes while kneading 😄

I used it a couple of times to make recipes more diet restriction friendly. I plug in the recipe and then explain what changes I want to happen. It does a really good job giving me replacements and gives me fun new ingredients or methods to research. 

As I'm sure we are all aware at this point, AI is far from all knowing. It will often times spit out something that looks good on the surface, but when it comes down to it, the information is off a beat. It always best to use it as a learning reference rather than an absolute answer.

Hope your AI bread executes flawlessly, and you have a bug free day!

I do use chatbots - I refuse to call it "AI" - from time to time but I never ask for a bread recipe.  The only time I would do that is if I have a partly-remembered recipe and wanted to track down the source or the entire recipe.

Mostly I start a dialog, and it often works much like talking something over with a person in that new or remembered things come to mind.  Sometimes I learn something new that's useful. Sometimes I remember things I knew but hadn't thought of for a while or hadn't put together in quite this way.

Here's my initial question in a recent chat:

There is a recipe for rye bread that calls for leaving buttermilk out at room temperature then mixing it with rye flour and leaving it covered at room temperature for another 24 hours. This is an unusual schedule for making bread. What is the name of this kind of rye bread?

Later in the session I asked this:

This recipe came up on the fresh loaf site in connection with food safety. I wasn't sure what the pH would be after mixing in the rye flour, nor whether the pH could be expected to decrease much during the 24 hour subsequent fermentation. I know that a dough made with sourdough starter will have its pH decrease into the food-safe zone fairly quickly but wasn't sure if that would happen with the buttermilk-only dough.

A discussion about pH, buttermilk, and food safety followed.

Another recent start to a chat:

I plan to make sourdough bread including 15% (baker's percent) each of ground pistachios and ricotta. I plan to mix the dough without the additions first and later add the nut/cheese mixture. I don't know how much water the ground nuts will absorb so I don't know how much water to hold back during initial dough mixing. I do know that ricotta contains about 80% water. Guidance for estimating the water-absorbing ability of the ground pistachio powder?

Later the chatbot over-analyzed things and offered to come up with a detailed recipe and procedure.  But I had my own plan.

One more example of a recent chat:

There is a style of sandwich roll or bun that share common features: very light airy crumb, somewhat elastic crumb that won't squish down completely, very thin, crackly yet flexible crust. Varieties include names like hoagie, sheboygen, banh mi, etc. It's not easy to make these buns successfully with all the intended features. What are common best practices for making these buns at home?

This chat was interesting, with the chatbot pointing out differences between many of the common roll types.

I try to pose questions that should be commonly written about (in the training material), so as to get the best chance of correct or useful results.  The newest version of ChatGPT is more neutral to work with and I like that.  It also seems to be better at "understanding" the context and brings in a wider range of apparently relevant material.  I don't expect that arithmetic will come out reliably and I don't ask for that (even though the arithmetic seems to be getting better - I haven't tested it lately).

I have a standard prompt I usually include:

Your ground rules:
1. Do not praise my remarks or try to give positive or fawning feedback.
2. If you do not know something, or make assumptions, say so.
3. Do not restate my remarks before making your own responses.
 

TomP

Tom, those conditions are interesting.  My preferred chatbot calls itself an AI Assistant, so what are your objections to using "AI"?

 

Yes, I regularly use "them" for recipes, outlining the ingredients and/or quantities I want. No funny stories, but once it left out an important ingredient. Caught it before I used the recipe.

They can call themselves whatever they like, but for me what we have now is nowhere near a general artificial intelligence.  For me, to be intelligent they need to have problem-solving skills and be able to assess or evaluate the quality (and common-sensibleness) of their responses. Until then, the term "AI" is hype that I wish to avoid.  So I say "LLM" or "chatbot". Although I suppose I won't be able to avoid "AI" eventually.  

I gotta admit, though, that especially with the new model of ChatGPT, the responses are sounding surprisingly cogent and conversational.

One feature of these things that I find interesting is that they may "know" something, but they don't know where the information comes from.  I'm like that too.  If I ask the chatbot for a reference it will run an internet search, but it can do that much faster than I can.  Sometimes the references are good, sometimes they aren't very on-point.

I recently had a long session with Grok about English muffins, which resulted in three interesting recipe options. I am now making quite good English muffins, so you could say it was successful. I did have one failed batch initially, because I tried the most extreme option first. 

If you offer background input to the AI, like your baking specialties and experience, it can help tailor the response you receive. So for example, I am primarily a  panettone and shokupan baker, so Grok framed the English muffin conversation with that in mind. 

I did find that AI has some blind spots because of course it cannot really perceive organoleptically, so you must take its opinion with a grain of salt, so to speak. 

I have been using ChatGPT for a few months, and Claude more recently, mainly as an education tool where I want to drill deeper into the 'why' for certain techniques.  I have generally found it very helpful, but it requires an intense amount of reading and skepticism.  The more I use these tools, the more challenging I get to the bot to justify its answers and clarify what it is saying.  I now have a much better understanding of how sourdough starter works,  what the impact is of different feeding ratios and different flours.  Just this week it explained to me why using bread flour for a new starter was a bad idea, and why Rye flour performs the way it does in building a starter.  

In the early days of using it for bread recipes, i had an absolute fail with the recipe it came up with,  (92% hydration), and learned that I must always check its math.  No exceptions.  But on the flip side, it taught me to always hold water back in reserve when mixing dough, and learning what the 'feel' of a dough is, and what my own limit is on shaping high-hydration dough. 

OpenAI did a very good job recently of explaining the 'why' of steam in baking.  The reason we got on this topic was that I asked if there was any downside to simply opening a 550F pre-heated oven and spraying water with a pump sprayer into the oven box.  (turns out its not a great idea.  Likely void the warranty.)

Most entertaining - I did come up with my own recipe for a Thai curry ice cream.  

One more - it was bang on converting from instant yeast to a combination of sourdough discard/active dry yeast.  I was pleasantly surprised at the result.  I'm done with sourdough crackers, now I use my discard to make more bread.  

And kudos to @tpassin for the ground rules.  That is a true time saver.  

 

Those ground rules make for a real improvement, don't they?  Glad they were useful for you. I got an earlier version directly from ChatGPT by asking it to suggest ground rules for the kind of conversations I wanted to have (and what I didn't want included). Later I refined them to the form they have now.

What reason did the chatbot give for claiming that bread flour for a new starter is a bad idea?  Bread flour as opposed to All-purpose, bread flour as opposed to some whole grain flour, or something else?

TomP

from ChatGPT:

Why early-stage starters can struggle on bread flour only

1. Nutrient availability

  • Bread flour is basically just refined endosperm.
  • Compared to whole wheat or rye, it has far fewer minerals (like magnesium, iron, zinc), vitamins, and trace elements that microbes need to thrive.
  • Yeast and lactic acid bacteria (LAB) rely on these micronutrients to multiply and establish dominance in the early days.

2. Enzyme activity

  • Whole grains (esp. rye) are rich in amylases — enzymes that break down starch into simple sugars (maltose, glucose, etc.) that microbes can eat.
  • Refined bread flour has much lower enzyme activity. Early on, before the microbial population is strong, this can mean not enough food gets unlocked fast enough for the microbes to flourish.
  • That’s why rye is famously “turbo-charging” — it brings its own enzymatic toolbox.

3. Microbial diversity

  • The flour itself brings in microbial “seed stock.” Whole wheat and rye carry way more naturally occurring yeasts and bacteria on the bran/germ surfaces.
  • Bread flour is more sterile (bran/germ stripped away, roller-milled, sometimes bleached), so there’s a smaller microbial population entering the game.
  • With fewer players, the starter can take longer to stabilize or might stall before the right balance is achieved.

4. pH and buffering

  • Early on, acidification is crucial — LAB lower the pH to suppress unwanted microbes.
  • Whole grain flours buffer better (thanks to bran minerals and proteins), so LAB can thrive without the pH dropping too fast.
  • On pure bread flour, there’s less buffering capacity, which can let the environment swing in ways that inhibit the beneficial microbes.

In short:

  • Bread flour = “clean” but nutrient-poor → slow to kick off a microbial ecosystem.
  • Whole wheat/rye = nutrient- and enzyme-rich → fuels early microbial explosions, helps LAB/yeast establish quickly.

That’s why you’ll often see starter guides recommend rye or whole wheat for the first week, then transition to bread flour once the culture is stable.

That's interesting. I agree with some of the response but not all.  I just had my own little chat about this.  The use of the term "bread flour" is misleading.  It really means "any white refined (roller-milled) flour that might be used for making bread". The claim that rye flours are very high in yeasts isn't right.  Whole grain rye flours are especially rich in LABs but not in yeasts. Most of those organisms are found in the bran and you can just use the bran at the beginning. Later you will want to add flour for its starch

Most refined unbleached flours in the US contain diastatic malt or amylase, which will help bread down the starch in the flour so there will be an adequate supply of nutrition for the microbes. Quick acidification is desirable to suppress undesired organisms. A paste of water and bran that I tested held the pH above 6 for some time, which is the opposite of what you want, apparently because of the bran's buffering ability. BTW, water + bran alone starts out at a rather higher pH than water + flour.

What was effective in rapidly lowering the pH of the bran solution was to mix in some live-culture yogurt.

My experience, in agreement with the above, is that you can rapidly create a working starter using white refined, unbleached, malted flour and an acidifying agent. Pineapple juice and brine from salt-fermented pickles were very effective, and water plus live-culture yogurt should also work well.
OTOH, if the white flour has been bleached or isn't malted (or have amylase added) and even more so if you don't use an acidifying agent, then using bran or whole grain flour is undoubtedly going to be very helpful.

  • Whole grains (esp. rye) are rich in amylases — enzymes that break down starch into simple sugars (maltose, glucose, etc.) that microbes can eat.
  • Refined bread flour has much lower enzyme activity. Early on, before the microbial population is strong, this can mean not enough food gets unlocked fast enough for the microbes to flourish.
  • That’s why rye is famously “turbo-charging” — it brings its own enzymatic toolbox.

2. Is partially inaccurate / misleading and is an example of a particular weakness of AI, in that it doesn't have oversight and so bundles together facts viewed myopically.

Wholewheat is indeed richer in enzymes, however amylase activity as measured by the amylograph will demonstrate little difference between wholewheat flour and bread flour made from the same batch of grains. They will have more or less the same falling number.

Rye typically has higher amylase activity compared to wheat.

Amylase activity is related to the grain itself, growing conditions and how sound the wheat is. Milling to wholewheat or white (bread) flour has little bearing on starch conversion rate.

Wheat growers aim for high falling numbers for bread wheat and millers are unlikely to accept grains which give a falling number much below 300. However for the purposes of bread-making bakers prefer more amylase activity. In the US, additional amylase is often added to the flour while in Europe malt is added in the bakery.