Last July we wrote a book called PATH. Prospect, assess, test, harvest. Brian handed the whole manuscript to ChatGPT as context and built officialolsons.com off of it. Then one day he asked it to confirm that its suggestions followed PATH protocol. It said yes, of course. So he asked it to define the protocol. It came back with persistence, accountability, truth, and honesty, completely confident, and not even close. The book was in memory. Attached. It built the website off it. And it still made the acronym up. That is where this episode starts, and Brian says the answer up front so nobody has to wait for it: this does not mean stop using AI. We use it every day, all day, and we are not going to be apologetic about it. The question Robin Joy keeps returning to is the sharper one. When somebody says they got the right answer, right based on what? There are two camps and they have the same problem. One camp has seen the slop and decided AI is garbage. The other camp is the new seller who posted a question in a Facebook group at eleven at night, got told to just ask ChatGPT, did exactly that, had an answer in six seconds, and has no idea what is wrong with it. Neither camp knows what is happening inside. You do not need a computer science degree for this. You need one thing. Brian ran the demonstrations himself. He asked a model to think of a famous person and played twenty questions against it. At the end it named Marie Curie, then admitted it never had a name in mind at the start. It had looked back at the answers it gave and found someone who fit them. It knew one fact about her, the two Nobel Prizes. It kept the fact. It never had the page. Then he ran 200 fresh sessions, in two blocks of 100, each one a cold window with no context, and asked for a random number between one and 100. Ninety-four percent of the answers came back 73 or 74. Six distinct numbers total on the first pass, five on the second. That is not random. That is the model handing back what the crowd says when it is asked to sound random. Humans do the same thing, by the way, and land on 37. The heart of it is a coffee cake. Brian's mother's recipe, made every Christmas morning, sour cream and shredded apple. Ask anybody what temperature a cake bakes at and you get 350, because most cakes are 350 and the answer is not wrong. The recipe card in her handwriting says 325, and longer, and the lower slower bake is the entire reason it comes out with that texture. Ask a thousand cookbooks and you get a perfectly good coffee cake that is not the coffee cake. Everybody has a 325 somewhere. It is the thing the average does not have, which makes the average exactly the wrong place to go looking for it. Everything we teach is off-road on purpose. Slower movers instead of saturated ones, 50 a month instead of 5,000, 325 for 60 minutes instead of 350 for 40. Then the defensive half, four checks you can start using on your very next prompt: Ask for sources, every time. Read the sources yourself, not the summary, because a summary is somebody else's interpretation. Confirm the model read it the way you would have read it. And the fourth one, the one almost nobody does, ask it to explain its criteria, then open a separate session and ask again. If the criteria moved between sessions, they were never criteria, and you cannot depend on them. Robin Joy widens the fourth check past AI entirely, and this is the part of the episode that outlives the AI conversation. A tool tells you an ASIN scored 94. Based on what? Ninety-four out of 10,000 is nothing. Ninety-four out of 100 is pretty good. A hundred what? Somebody emails offering a list of 500 vetted brands. Define vetted. Ask that of your tools, of anyone you listen to, and of us. The back half is the other side, what happens when it does know you. Brian handed our Claude an ASIN and asked whether it passed the three-step check, and it refused, because it did not have the Keepa chart or his break even number and it was not going to guess. Then it ran the pre-flight inspection on its own, found a brand problem on the listing, and told him not to bother. It also flagged that the item is a sprinkler nozzle and it is September, so the capital protection he was measuring might be an off-season low. That is not a smarter model. That is somebody who has cooked in your kitchen and knows your oven runs hot. Robin Joy closes the loop on how you build that, using the coaching follow-up emails she writes after every one-on-one. She used to correct the draft, send it, and save the corrected version somewhere the model never saw, which meant making the same corrections forever. Now she feeds the corrected version back. The changes get fewer every month. She still reads every word before it goes to a client, and she always will. Be honest about the timeline, though. This is a couple of years of content, six years of Brian's coaching files, five of Robin Joy's, the podcasts, all of it organized and corrected as things changed. This is not a weekend project. It compounds. And here is the tie-in to becoming undeniable. The average business is deniable by definition. There are ten thousand of them and they all made the same cake. Nobody else has your frameworks, your numbers, your history, or your mother's 325 on a recipe card. You have to build that. A tool can make you faster. It cannot make you undeniable. That part is still yours. In this episode: (00:00) The book is called PATH, and ChatGPT built our website from the manuscript(02:00) It had the book in memory and still invented the acronym(03:00) Two camps, and why we are not apologetic about using AI every day(05:00) The slop, and the new seller who got an answer in six seconds(06:00) The one thing you need to know, and no, you do not need a CS degree(07:00) Twenty questions against a model that never picked a name(09:00) Marie Curie: it kept the fact, it never had the page(11:00) It is not lying. It is finishing the pattern, and a plausible answer beats a blank(12:00) 200 fresh sessions, pick a number between 1 and 100(13:00) 94% came back 73 or 74, and why humans land on 37(15:00) Mom's coffee cake, and why you would guess 350(17:00) The recipe card says 325, and it bakes longer(19:00) Everybody has a 325, and what a cold window thinks PATH means(20:00) Everything we teach is off-road, on purpose(21:00) Check one: ask for sources, then read them yourself(22:00) What kind of source is it? Primary versus forum(23:00) Check four, the one nobody does: explain your criteria(25:00) Tools that hand you a score you never set(27:00) Why the Modern Builders Stack shows its work on all 800-plus sources(28:00) No hacks. If it cannot scale, it is not a solution(29:00) The other side: it refused to guess without the Keepa chart(30:00) The pre-flight inspection caught a brand problem on its own(32:00) Somebody who has actually cooked in your kitchen(34:00) Teaching it, one coaching follow-up at a time(37:00) The dog, the stick, and training the human(39:00) The average business is deniable by definition(41:00) A tool can make you faster. It cannot make you undeniable(42:00) Feynman, and who those four checks are really there to catchThe next time a tool hands you a number, ask it based on what. If you cannot get an answer, you just learned something about the tool. Run your own numbers. The Profit Tune-Up is free. Upload your Amazon or Seller Board data, or even a Seller Board screenshot, and get the same review Brian runs in a live coaching session. officialolsons.com Everything else lives in one place. The newsletter, the coaching, the community, and the rest of the tools are all at officialolsons.com (all O's, no E's). New episodes weekly. If this one helped, leave a rating. It is the single easiest thing you can do for us. The New Era Reseller is part of the Modern Builders Network, hosted by Brian and Robin Joy Olson. Become undeniable.
