#93 - Chat GPT /Ai Calorie Tracking

ChatGPT can tell you your dinner is 300 calories, and that single number can quietly derail weeks of fat loss. We’ve both seen it with clients: AI gives a confident estimate off a blurry, top-down photo, misses what’s underneath the rice or inside the sauce, and suddenly someone is underfuelled, starving by evening, and wondering why progress feels impossible.

We dig into what AI calorie estimation can and can’t do, why prompts matter more than people think, and how misinformation can sneak in when the model pulls from low-quality sources. We also talk about the bigger issue behind the tech: autonomy. Calorie tracking is a skill, portion size is a skill, and the goal isn’t a lifetime of logging or reassurance. The goal is getting good enough that you can eat normally, estimate reasonably, and stay consistent without a “safety blanket” app.

You’ll hear practical ways to make AI more useful for nutrition coaching and self-coaching: describing ingredients and quantities, adding a real-world size reference beside your plate, and doing simple calibration tests where you weigh a meal you control and compare the true number to the AI guess. If you’re using MyFitnessPal, ChatGPT, or any food tracking app, this chat will help you keep your head clear and your expectations realistic.

If this was helpful, subscribe to Gen Pop Podcast, share it with a mate who’s scanning every meal, and leave a review so more people can find it. What’s the weirdest or most misleading calorie estimate you’ve ever gotten from AI?

Got questions? simply email or dm us with those questions

Larry IG https://www.instagram.com/larry_doyle_coaching
Daniel IG https://www.instagram.com/danieldalycoaching
Website https://www.larrydoylecoaching.ie
Coaching with us https://www.larrydoylecoaching.ie/1-1-premium/
Email : info@larrydoylecoaching.ie

Transcript

Welcome To Gen Pop Podcast

SPEAKER_01 0:00

Hey guys, you're listening to the Gen Pop Podcast with me, your host, Larry Doyle. Each week, I'm gonna bring in friends, guests, and experts to help enhance your health, fitness, and wellness journeys. Sit back, relax, and enjoy the conversations.

AI Shows Up In Everyday Life

SPEAKER_02 0:26

I've seen so many people getting into this now who are like sending us in screenshots of, yeah, Larry, I've had ChatGPT uh kind of look at all my food for the week and give me some recommendations based off it. It's a bit of a minefield, a bit of a dangerous element. People are getting into it, Daniel. Let's open the discussion on this.

SPEAKER_00 0:43

Yeah, I think um AI use is just prevalent in every in every sphere now. It's like we were only talking about it before the podcast that like no matter what town or village you go into, you see posters for some fucking raffle or some event that's happening over the weekend, and you can see straight away that the poster is made with ChatGPT. And now, again, we're not gonna sit here and say, like, we don't use ChatGPT or we don't use AI. I think everyone at this stage is using AI because it is a fantastic tool, but we're used it to kind of enhance something that we're already doing, we're used it to kind of fill in a gap. Like we're not, we'll say, you know, uh we're not dependent on graphic designers and we're not people that can actually develop these things ourselves, and like it kind of makes sense for us if we just need something quick to do that, but like we're not like trying to take walk away from other people, or we're not trying to say, right, we're going to use this to kind of replace our job or our role within uh the coaching community.

Why Food Photos Mislead AI

SPEAKER_00 1:38

But think of like people at the South Park, like they took our jobs exactly, and uh I think like as you said, we have clients that use AI for the nutrition and whatnot, and like I'm kind of on two minds about it because number one, it's kind of good that they're doing something at least that's better than them just not tracking, but at the same time, people are putting too much faith into it, and like we always kind of use the example that when we tested like AI functions when it comes to nutrition tracking ourselves, that like if you had a bowl of say chicken curry, for example, and like it was a bowl and the rice is underneath the bowl, and you're just taking a picture over the top, it's not going to show the rice, and you're asking ChatGPT, okay, how many calories is in this? It's gonna spit back. Oh, there's about 300 calories in that. It's like, oh fucking hell, there's a lot less calories than that that I thought. This is amazing, I can eat this forever. Then you're trying to figure out why am I not losing weight? I'm tracking my calories, and like ChatGPT is telling me this much, and as we know, the AI overlords are going to take over, it's taking every job, it's amazing, and all this, but it's like again, I'm in two minds about that because I think AI is only as good as the people that are using it, like, and there's a lot of fucking idiots out there that are using AI that are using it basically like Google. That they it and the the funny thing about it is if you put in something to ChatGPT, a question to Chat GPT, it will bring up kind of logos at the bottom or icons of like where it's taking the uh results from, and you'll see like a Reddit popping up there, and it's like, has anyone ever spent five minutes on Reddit? The amount of fucking morons on that thing that are giving their opinion on something, and you're gonna base all of your judgments off of the opinion of some fucking idiot over in the back arsenal in the States. So you have to kind of be realistic with your use of AI and actually learn what you're actually looking to achieve with it. So, like, if you're having a meal, for example, and you know it's gonna be chicken curry, get descriptive with it and say, I had maybe you know three uh cupped handfuls of rice in there, there seem to be like maybe like a pam of chicken in there, maybe there's whatever ingredients in there, and feed that into the AI, and it'll give you a ballpark, but it's always gonna be a ballpark. But if you're just doing the bare bones minimum and you're trying to like offload all of your accountability, tell me the calories in this and upload the picture.

SPEAKER_02 3:50

Like that's so useless.

SPEAKER_00 3:52

Yeah, and like it's the same as like if someone took a picture from a helicopter down at the earth and said, like, right, what's underneath the ground there? It's like I haven't the fucking clue. It's like it doesn't have X-ray vision, it's not gonna be able to tell you every single detail, it's like very surface level, they're giving you a 2D image, it doesn't really have much to work off of. So again, I think people are just like using it to try and make their lives easier, but they're ultimately just like making themselves more lazier because they're like doing less work, they're not like learning the skill. Like, if we were to use AI, for example, to kind of track our meals, we would have a good understanding of like that's not really that accurate. So we could know and understand, okay, that's not actually what is in there, so we can kind of make changes off that. But if someone is coming in fresh, never really tracked calories before, have a vague understanding of calories, and then they're going to look at that and say, like, well, every news article I see and every fucking thing is talk about AI saying it's amazing, it's this and that. They're gonna put all their trust into that. And the scary thing is they probably put more trust into that than they would their actual coaches, because you've probably experienced it as well, where clients will report back saying, like, oh, I asked ChatGBT this and this is what it said, and you'd be like, No, no, no, that's that's not what we're you know looking at doing, or it's not what I would recommend, or whatever. And it's like, but but ChatGBT said it, like it's yeah, why can't we do this, or why should we not do that? So I think you have to be kind of realistic and and with your use of AI um and just kind of be careful of what you're expecting to kind of uh get from

Portion Sizes And Real Autonomy

SPEAKER_00 5:18

it.

SPEAKER_02 5:18

Yeah, and again, like the interpretation of it, like I had a client send me a meal that he had in the restaurant or the canteen in work yesterday, and it like he had like two fillets. I know you this was like one is like your typical is like what is a portion of lasagna? It's like fucking what is lasagna, yeah. Yeah, what is a fillet of hake? Because I know if I cut a slice of bread myself, I'm a greedy bastard, and it's gonna be like a big door wedge, it's not gonna be like a thin slice. So, like, what is the fillet? What is a standardized portion of this stuff? And ultimately our job, and it's probably a shit business model, but is to increase the autonomy in the individual that they actually can sustain it themselves, that they don't need a reliance on tracking, they don't need a reliance on coaching all the time. But if you're constantly relying on Chat GPT, if you're constantly relying on my fitness paddle, if you're constantly relying on something to give you all the answers, and you're not paying any attention to those answers whatsoever, to even question it and say, I don't think that's pretty accurate. Like, even do something like put a can of your Diet Coke or your monster or fucking something beside the plate so it has a size context beside it. Tell it that I have put a 330 mil can beside it, I've put a 500 mil can beside it. Here's my iPhone, my work phone, my whatever beside the plate. And all of a sudden now we can kind of calibrate that a little bit more to what a portion size might be in relation to that. But to come back to the client, he had given uh he had two pieces of hake. Again, did a bit of a uh Chat GPT on it, it came back as two fillets of Hake. It came back that there was oil, there was like the meal was two pieces of hake. There was green beans, there was a cucumber, and some green lettuce and stuff like this as well. It came back that there was like 40 grams of carbs, 20 something grams of fat, and again, there was like 60 something grams of protein within it all. And it gave them like, I don't know, it was like 500 and something calories as the total intake. It was literally two half pound sized versions of hake, some cucumber, some greens, some whatever, and it was a little bit of sauce on it. It's not physically possible to have like 20 something grams of carb or 40 something grams of carb or 20 something grams of fat within that meal. So if you're not questioning this, if you're not spotting it, I ultimately came back and said, right, based off this, I think there's about because again, he's a very good client that way, he will be diligent to a degree with it, and he'll put his work phone beside it to give me context and reference. I want him to come back with his guesstimations to me when he's sending these across, as opposed to saying, Hey, Larry, what's the calories in this? I've taught him that I want you to come back with a guesstimate of what you think is the calories within this, so we can see where he's thinking and see what his calibration is with it. Then, based off my experience of seeing like literally thousands and thousands of meals and helping clients with them all, I'll know that it's not gonna be that close, or I know it's gonna be very close, or I know it's up and down, or whatever. Now all of a sudden we came back and it was about 280 to 300 calories that were actually in this meal, given the size, given the reference, given what I knew was in the meal, given the ingredients. Now all of a sudden, like he's got multiple hundred calories left over that is not gonna be like way underestimating his calories, and now on like 1200 calories for the day. Because if he does this in every single meal and he's out by about 40%, he's gonna be vastly under eating, he's gonna be vastly underfueled. Chances of him overeating later on that day are now huge because he's gonna be starving, he's not gonna be like, I stuck to my calories for the day, but I'm fucking ravenous. It's like now all of a sudden he realizes I could actually have more food during the day and I'm probably not gonna be eating the shit out of the house by the time I get to it later on the day. But because we pay that little bit of attention to detail, now he knows in his head, he's taking a screenshot of that physically in his mind that I know what a portion of Hake looks like. I know what those calories are like. But if he's constantly just inputting those into my into ChatGPT or MyFitnessPal or whatever it might be, and getting these random fucking spurious answers coming back, he's no autonomy in his approach. He's just constantly, he's just data entering, like he's just mining for data all the time. There's no learning, there's no skill set. Like fat loss is a skill. Tracking calories is a skill. You are going to need tools like AI, like MyFitnessPal, like whatever, to get to a certain point of proficiency. But the goal should always be to get to the point of proficiency. It shouldn't be to the point of dependence. And when we've got a massive dependence on the reassurance from AI, from MyFitnessPal, from the weighing scales, from whatever. Like we'll see this all the time. People coming in, yeah, I've got a five-year streak on my fitness pal, but they don't know the fucking calories that are in the most basic foods. And those five years of entries have been the same shit every day. It's like, grand, because if someone's that regimented with tracking on my fitness pal, you can be guaranteed that they're eating a lot of the same shit all the time, but they haven't a clue. If we were to take away my fitness pal from them, they wouldn't survive a minute because they'd freak out because they don't have this safety blanket. Every entry you make accurately with your data, with your food, is another step forward to never needing it again. And this is something I'll give people that all the time. Yes, try learn from AI. Yes, question it to say, is that actually the best answer you can give me? Is this based off something else here? Like what is the portion size you're giving me? What do you think is uh how many grams do you think is in this meal? And then get a reference point. Again, if you get familiar with looking at how many grams is like 200 grams of meat, it doesn't matter really what the meat is, it's gonna look very similar on a plate. If you know that a portion of rice or a packet of rice is this much on my plate, if it takes this much at the bottom of the bowl, if it looks X, Y, and Z way, you can be pretty damn sure that it's gonna be similar every other time you look at it. So again, it's learning from that and giving yourself the reassurance that there isn't just these like magical ingredients that are magically going to be scanned by AI and picked up. Again, it's gonna give you like either very high outlier or very low outlier answers in terms of like way more calories or way less calories that are actually in it. Very rarely will it give you an accurate response, right? Um, I have seen a bit accurate before, but more often than not, I'll get clients sending in and be like, Larry, I sent this in, like, what do you think here? And I'm like, no, it's not remotely near it whatsoever, because again, there's no reference, there's no ability for that to understand what's actually there. It can only, it's like the prompts, and like this is like where people like the best way to learn ChatGPT is learn how to give it better prompts. So for you to get better answers with your caloric estimations is to give it better prompts. So again, having the reference point beside it saying this is my Yeti cup that's beside it, and this is a X amount of, it's a 0.9 liter Yeti cup, cool. It's going to be able to look up the dimensions of that and dimensions of the plate and the dimensions of the food that's in it and give a really good accurate response then as well, too. But if you just give it a random bowl of shit that's heaped up on top of it, or like, here's my burrito in tinfoil, it's like the fucking AI doesn't know what the fuck is in that. So again, like it's it's it's using a little bit of common sense to see if can you get a better answer, but then also question it that little bit more and say, is this actually in line with you know what we think is within the calories? But then learn from that and know that you never need to use it again if you've done that one time. Just pay attention, and it means you're getting way more food freedom, it's way more tracking freedom, way more less reliability on

Better Prompts And Calibration Tests

SPEAKER_02 12:36

it. Because again, AI, I strongly believe, and there's more again, studies and research coming out on this, that for the first time in like generations we're getting dumber as a race. And AI has a lot to fucking blame on that. And again, like you'll see the posters for the local GA match, and it's made by GA or it's made by AI, and there's like a player with like four legs on it because they don't even bother their whole to fucking reference it and say, Does that actually look like it should be? Or there's like misinformation. People are putting in something, they're asking it for a quick response, they're not even proofreading it or checking it then based off what's coming back out. So we're getting super lazy.

SPEAKER_00 13:21

Let's go. One of the things that I find is a bit concerning about AI use is I have to be very careful with the wording I use here. Don't go on L elderly, the elderly group of people that are using AI use. They're also the

When People Trust AI Too Much

SPEAKER_00 13:36

people that when Google they discovered Google were typing www.google.com into Google to get onto Google. So it's like they're also now using ChatGPT and stuff, and like they're the ones that are going to be like thinking like, oh, this is the fucking this is the bees' knees, this is gonna answer every question they ever have, and it's going to be like the new doctor for me, it's gonna be the new trainer for me, it's gonna be the new fucking everything for me. So like if imagine if you went into the doctor's surgery and you walked in, you said, All right, doctor, I have this problem, whatever. And next team went out to ChatGPT, we're like, would you be happy with that happening? But like that's the level of spat your ass it's after happening. 100% it has like, and I think everyone has kind of heard that story of or has experienced themselves of the doctor using Google, but for some reason that's not you know good enough. People wouldn't have, you know, people will be complaining if the doctor used Google, but ChatGPT seems to be like the thing that like if they use that, then like no, it's that obviously is going to be giving more correct answers, but like ChatGPT for the most part is going to be getting most of its answers from a lot of places you could find on Google. It's just pulling it into one place and giving you like an actual paragraph as opposed to links where you now have to go and do the work yourself and kind of determine is that actually the correct information that I'm seeing in this link, or will I have to go to the fourth page on Google? Or so I think I think like you used the example there of one of your clients like putting in the Hake, and I kind of overestimated it. Like I actually done that exact same thing on Monday with clients that I used chicken curry. I got a picture of BBC Good Foods and I just put it into AI and I gave three different scenarios. So I put it into AI and I said, give me the calories for this meal. Didn't say what it was, just said I'll leave it up to them to figure it out. And they said it was like seven to eight hundred or seven to nine hundred calories. Then I told them what it was, so I was like, it's coconut chicken curry or something, and it said, okay, that's eight hundred and fifty calories. Then I actually gave it the ingredients and the quantities from the website, it said it was like 500 calories, but the BBC Good Foods website said it was actually 400 calories. So it's like there was actually four different numbers that I got there from three from ChatGPT, one from the website. Like, right, how does the say uneducated person who comes to tracking calories or estimating wherever determine like where does this meal actually lie? Because like that, if ChatGPT said it was 800 calories, that could be like double the amount of calories that actually is in that meal. And if somebody's tracking that, they're gonna be like, I'm fucking starving all the time, I don't understand why, and all this kind of shit. And some people might look at that and say, But if it's overestimating, isn't that a good thing for fat loss? Because ultimately you're eating less. It's like, but more often than not, that's gonna cause someone after four days to fucking binge because of like, what's the point? I'm constantly starving, I can't eat any foods, all this kind of shit. So that's where like using Chat GPT can be good for kind of educating you to a degree if you use it correctly, but I think you still, as with everything, have to do the work, you have to, you know, ask us questions, lever off, leverage off of coaches, and just kind of be realistic with things and kind of say, like, does that actually seem real? Does that actually seem like it is accurate at the end of the day? So I think ChatGPT can be a good side tool, but I don't think it should be, you know, taking all the uh the work out of things for you.

Use AI As A Tool

SPEAKER_02 16:43

I think it's a bit of a wrap-up on this one. You're gonna find two specific different cohorts of people who will get those references and ranges of answers. So you're gonna have the person who is more likely struggling with weight loss, struggling with adherence, they're gonna be like, it's telling me between four and eight hundred calories. It's definitely four hundred calories. Yeah. Even though it might be six or it might be eight, it might be thousands. And then you have the person who's really anxious, they're really wound up, they're really struggling with their stress management, they're really struggling with keeping enough fuel coming in, they're gonna be like, yeah, there's definitely, oh, like ChatGPD said 800, so I'm just gonna put down a thousand to be safe. It's like they're gonna go to two different polar ends and opposites of it, and this is like where people really struggle. So it's getting better, better at understanding what's actually in the food is gonna reduce that ultimate stress for those, and like getting better at estimating for those who are like kind of weight loss deniers or like calorie deniers is gonna give them a little bit better chance to actually lose some weight. So our takeaway from this is use it. It's a tool, don't rely on it. Try and get yourself up to speed that little bit more. Does this actually make sense? I tell you what I'm gonna do. I'm gonna weigh out my food when I had control over it. I'm gonna know the exact portions that are in this, I'm gonna weigh them out, I'm gonna put them into my fitness panel, I'm gonna trek and I'm gonna track everything that's in it, but then I'm gonna get a calibration based off that with ChatGPT and say, I know there's 550 calories in this meal. Don't tell my uh don't tell ChatGPT that. But then put it in a picture. Just say, tell me the calories that are on this plate. You'll see exactly then how accurate or inaccurate it's gonna be. You can do these little calibration tests from time to time, and you're gonna get a really good gauge of what the hell's going on and what's not going on as well, too. Uh, Daniel, it's been an awesome episode. Loads of takeaways from it with this. Hopefully, people have realized to use it, use it as a tool, but try to create an independent approach from that, that you're not overly reliant all the time on utilizing all these. Again, it's an amazing tool. I'm gonna keep using it, but I'm not gonna be like getting it to create this dependence on it all the time to uh try and hit my goals because again, ultimately, we're not gonna be able to hit them, we're not gonna sustain them, we're not gonna be able to go long term. And if uh God forbid we can't log into my uh into uh Chat GPT someday, we're actually gonna know what the fuck to be able to do with it all. Uh, Daniel, been an amazing episode.

SPEAKER_00 18:55

Talk to you guys soon.