Asking AI....

[182042]

Decided to clear out all brewing ingredients and begin ordering ingredients per recipe to keep everything fresh.

Provided AI with a list of everything I had on hand asking AI to create a Pliney the Elder style recipe utilizing everything. AI sent back two interesting recipes one of which I brewed on Thursday 5/28. It's name is Sweeping Double IPA and is visible in Brewers Friend. Intending to post follow-up as it progresses.

 

[Craigerrr]

A few years ago, when AI was.. sorry AI, what is that?, I asked a friend to ask it to create a recipe for a black IPA.
It was an interesting recipe, that I never brewed, but I'm sure it would have been fine.

 

[Brew Cat]

what I would be nervous about is AI will sample all the recipes for say a Pliney type beer and return the results of any beers that may or may not be any good just the most recipes. Often I'll ask AI questions and it comes back with the most common answer not always the best but I guess it's a start to a search
post the recipe I'd be interested to see it
my way would to be get a Pliney recipe I like and then start looking at a substitution chart and using my own experience to make changes
I mean lets face it most beers that are shitty are because you go above the recommended % but not always AI will probably stay in those parameters
Im sure you can tweak your query by inputting recipes you like

between AI and AIO I feel we loose the craft in craft brewing

 

[Dave Y]

I've never used it for a clone, but I have used it to backstop my recipe 'building' and I've found it pretty helpful.

 

[Sunfire96]

AI chat bots are not equivalent to search engines. But they are a useful tool for organizing, analyzing, and otherwise manipulating data and information. A better question for AI is which Pliny clone has the best reviews? Or, which of these Pliny clone recipes can I make with the ingredients I have on hand

 

[182042]

"Sweeping Double IPA" as mentioned my goal is to use up existing inventory. I agree using AI tends to negate the craft aspect. I generally choose a style, research mass recipes select ingredients and build my own in Brewers Friend, even building up RO water. Oddly enjoyable

 

[NorthCoastBrewer]

In the July-Aug 2026 issue of Zymurgy Magazine (AHA member magazine), the "Dear Zymurgy" column (page 9) overviews the approach that Zymurgy is using for AI. It also provides a link to AHA’s AI Policy.

AI chat bots are not equivalent to search engines. But they are a useful tool for organizing, analyzing, and otherwise manipulating data and information.

Agreed - as long as one accepts that responses are (still) statistical word generation.

 

[Zefram]

I've been using AI more to figure out additives for a water profile than anything else. I'm still learning how much of each to get the right combination for a given recipe, so that's been pretty helpful.

 

[Hawkbox]

I'd rather quit brewing entirely than use an expensive chatbot to do it for me.

 

[Donoroto]

Exception: if the chatbot would do a good job cleaning up afterwards…

 

[Mont Y. Märzen]

ChatBots are not 'answer machines'. They even have disclaimers. I can't imagine why folks keep thinking they can ask them questions and get 'answers'. Sure, the output might resemble an answer, and it may appear as if an intelligent human wrote it, but it is simply a generated pile of words with no reliance or connection to understanding, cognition, or anything we'd remotely think of as 'intelligent'. Specifically, they are Turing Machines - designed to fool you into mistaking them for an intelligent human, a novelty device to be sure, like a parlor trick or Magic 8-Ball, nothing more.

Be careful the Dunning-Kruger effect. Using such a device doesn't make you 'clever', not even by half.

At best, this probablisitic word generator might get lucky and accurately 'average' many recipe stats and produce a drinkable beer. But that is an accident. It doesn't *know* what any of this is. It can't even do math. It is simply regurgitating words based on proximity in the source data. If the resulting average that is suggested doesn't make sense, it doesn't know or understand that. It cannot lie, or hallucinate, as it cannot understand any subject, only you can.

But you can easily look at 5–10 recipes on your own and end up in the same place. You can read an article or book on the intended style and you'll learn in the process, rather than just doing what the Magic 8-Ball suggested.

 

[Mont Y. Märzen]

I've been using AI more to figure out additives for a water profile than anything else. I'm still learning how much of each to get the right combination for a given recipe, so that's been pretty helpful.

You are more likely to learn that and learn it faster, by reading about it and brewing it, than asking a ChatBot.

 

[Mont Y. Märzen]

In the July-Aug 2026 issue of Zymurgy Magazine (AHA member magazine), the "Dear Zymurgy" column (page 9) overviews the approach that Zymurgy is using for AI. It also provides a link to AHA’s AI Policy.


Agreed - as long as one accepts that responses are (still) statistical word generation.

I also agree with your point, but I refute they are useful for "organizing, analyzing, and otherwise manipulating data and information", because they are not designed for those purposes. The design is to *fool* you into thinking they are.

An actual app designed for such use would be infinitely better.

 

[Mont Y. Märzen]

AI chat bots are not equivalent to search engines. But they are a useful tool for organizing, analyzing, and otherwise manipulating data and information. A better question for AI is which Pliny clone has the best reviews? Or, which of these Pliny clone recipes can I make with the ingredients I have on hand

But asking which has the best reviews is a search engine type question. You could just see the review counts on your own. It doesn't know what a recipe is, or what a review is, or what 'best' means.

Trying to get it to match your inventory to a recipe is worse. See my other posts in this thread as to why.

 

[Dave Y]

I also agree with your point, but I refute they are useful for "organizing, analyzing, and otherwise manipulating data and information", because they are not designed for those purposes. The design is to *fool* you into thinking they are.

An actual app designed for such use would be infinitely better.

IDK how much experience you have with the various AI clients, but I interface with them daily or almost daily in my job. If they're designed to fool us, they've done a great job of fooling me and about 25,000 cow-orkers, about 1/3 of whom also use an AI agent regularly to multiple times / day.

If you feed the right ai client the proper prompts, it is capable of composing code in various languages, it's capable of selecting the most efficient ways to accomplish routines and subroutines, and it's capable of also doing exactly what you indicate.

While you can certainly read reviews of recipes on your own or pick out 5-10 that may be on point, the AI client can do it faster.

Now, AI can be wrong. Depending on the model and currency of code it can feed you bad data, incomplete data, and sometimes you need to 'challenge' it on the source.

Sadly, AI clients are useless at brew day cleanup. That, my friends, is what wives are for

 

[Donoroto]

Sadly, AI clients are useless at brew day cleanup. That, my friends, is what wives are for

Ah well. A guy can hope.

 

[Semper Sitientum]

I've been using AI more to figure out additives for a water profile than anything else. I'm still learning how much of each to get the right combination for a given recipe, so that's been pretty helpful.

I use it for this purpose as well. I don’t really care to learn why I need X amount of gypsum in Y recipe. Plus, personally, I find the BF water addition calculator a little wonky.

 

[Mont Y. Märzen]

I use it for this purpose as well. I don’t really care to learn why I need X amount of gypsum in Y recipe. Plus, personally, I find the BF water addition calculator a little wonky.

I haven't had any issue with it since I learned it won't suggest amounts of salts for me. I have to choose them and tweak until it shows I've hit my desired target. It will however suggest an acid amount if you select the proper type with the proper concentration based on your grain bill.

The only thing I find a bit buggy about it is if I make changes later, it doesn't always want to update the amounts it feeds back to the recipe. Doing a save-rinse-repeat dance though usually unsticks it.

I have some friends that complain about the pH estimates and the acid suggestions, but they don't realize you should spot check your pH of your strike water on brew day and adjust if needed. Defaults in profiles are just estimates. The calc works fine with accurate numbers.

 

[Mont Y. Märzen]

IDK how much experience you have with the various AI clients, but I interface with them daily or almost daily in my job. If they're designed to fool us, they've done a great job of fooling me and about 25,000 cow-orkers, about 1/3 of whom also use an AI agent regularly to multiple times / day.

If you feed the right ai client the proper prompts, it is capable of composing code in various languages, it's capable of selecting the most efficient ways to accomplish routines and subroutines, and it's capable of also doing exactly what you indicate.

While you can certainly read reviews of recipes on your own or pick out 5-10 that may be on point, the AI client can do it faster.

Now, AI can be wrong. Depending on the model and currency of code it can feed you bad data, incomplete data, and sometimes you need to 'challenge' it on the source.

Sadly, AI clients are useless at brew day cleanup. That, my friends, is what wives are for

Yep, they are doing a great job of getting people to think they are designed for various tasks other than novelty parlor tricks. And they are getting better and better at it. I have several friends and family that are fooled. Developers and tech shops who jumped on the bandwagon are already pulling back and questioning that move. The net is littered with articles and blog posts on that very subject. Rationality is setting in surprisingly fast with humans, considering the hype, but it remains to be seen if that will be fast and pervasive enough.

You might ask yourself if we've reached the point that this doesn't matter. Except once you give up on that question, you turn over your agency to make decisions to it. That is always very dangerous, and not just for you. I have no fear of these bots. My worry is people trusting them as intelligent and turning over critical decisions to them.

I don't really care if a bot can fake a recipe faster than I can. Speed isn't a virtue here. If that is the goal, just pick any standard style recipe from a book, magazine or the web and brew it. Learning how to brew a style is my goal. It isn't like any knowledge gained isn't applicable to other styles either. As noted above, devs and engineers are starting to challenge the notion of speed. Yes, you can get it to regurgitate more than you can in less time. But having to tweak your prompting in the meantime should be counted as part of the time-cost and most folks are not. Then there is the time spent checking the result, cleaning up after its mess, etc. I'll just learn malts and hops and yeasts and process, and practice brewing, and have fun making my own beer, thanks. If I wanted speed, I'd just buy a craft beer off the shelf. (which I also do!)

It can't be wrong. It doesn't understand or comprehend anything, at all. It can refine the way it generates text, but that doesn't mean it is learning. If what it spits out isn't correct, then you simply got bit by probability. Knowledge or intelligence don't even enter the picture. And lets clear up something very fundamental to these things:

You are *not* even having a conversation with it. You are *not* challenging it or questioning it, or anything of the sort. You are prompting it with text and it is *appearing* to reply. What you see is not a reply. It is the algorithmic result of assembling words from a set of words based on your words.

 

[Zefram]

You are more likely to learn that and learn it faster, by reading about it and brewing it, than asking a ChatBot.

I respectively disagree. Asking AI what "recipe" of additives will achieve the desired water profile is considerably more efficient than "guessing and checking".

I read the Palmer & Kaminski book on water and, while way over my chemistry competency, it did help me better understand what profiles are best suited for certain styles AND what each individual additive can contribute to that profile. My point of using AI is to find the best amounts of each to achieve the desired profile.

Nerd fact: I'm a former math teacher and tried to set up a system of equations (a matrix, really) to find the perfect combination of additives for a desired profile. I found out real quick there isn't an exact answer in most cases. I would tinker with the result to make it better, but I found the AI solution a better approximation to start with.