What I Learned From Posting My AI Outputs Publicly

3 months of posting my AI outputs on social media, what worked, what flopped, and the 3 things I learned about sharing work in public.

The experimentFor the last 3 months, I have been posting my AI outputs on social media. The outputs are things I would normally generate for myself or for clients: a prompt I refined, a script I built, a debugging session I went through, a system I designed. Each one is a real thing I made with the help of a model, and each one is posted as a screenshot or a code block or a short writeup.I started the experiment because I was curious. I had been using AI tools heavily for about 2 years, and most of my use was private. I would generate something, use it, move on. The outputs were invisible. The only feedback I got on my AI workflow was the speed at which the work got done, which is a private metric and a misleading one (speed is not quality).Posting the outputs publicly changed the feedback loop. The work was no longer private. Other people could see it, react to it, ask about it, and tell me when it was bad. The feedback was free, fast, and not filtered through anyone I work for. The feedback was the most useful AI-related data I have received in 2 years.What workedThree things worked. The first was posting the prompt, not just the output. When I posted a script, people asked "what prompt did you use to write it?" When I posted the prompt, the question turned into "I would not have written that prompt the same way, but I see why it worked." The prompt is the part that teaches. The output is the part that demonstrates. Without the prompt, the output is a magic trick. With the prompt, the output is a method.The second was posting the failures, not just the wins. I had been trained by social media to post the polished thing. Posting the polished thing is what the platforms reward. Posting the failures is what taught me the most. The failures are where the model broke, where the prompt was wrong, where I had to re-do the work. The failures are the part that makes the wins look real, because the wins have a cost. Without the failures, the wins look like the model did it, and the model did not do it; I did it, with the model.The third was replying to every comment for the first 7 days. The first 7 days are when a post gets most of its reach. Replying to every comment in the first 7 days kept the post in the algorithm, kept me in the conversation, and gave me a feedback loop I would not have had otherwise. The replies are also the part that surfaces the good follow-up questions, which become the next post. About 30% of the posts on this site were prompted by a comment on a previous post.What floppedThree things flopped. The first was posting the output without context. I tried posting a script with the caption "made this with AI, pretty cool right?" and got almost no engagement. The caption was lazy. The post was lazy. The lesson: the post is the part people engage with, not the output. The post is the teaching. The output is the example. Without the teaching, the example is just code.The second was trying to be clever. I posted a few posts that were "look at this clever prompt trick I figured out." The posts were clever. The posts were also useless. Nobody could reproduce the trick in their own work. The lesson: cleverness is not the goal. Usefulness is the goal. Cleverness is a write-off if the cleverness does not make the output more useful.The third was posting too often. I tried posting daily for 2 weeks. The first few days got good engagement. By the end of the 2 weeks, the engagement had dropped by about 70%, and I was running out of things to post that were not filler. The lesson: posting is a cadence, not a sprint. The cadence that works for me is 2-3 times a week, with at least one substantial post (a real how-to, a real example, a real failure) per week.What I learnedThree things. The first is that the AI workflow I was proud of in private is not the same as the AI workflow I am proud of in public. The private workflow had shortcuts I did not realize were shortcuts until I had to explain them publicly. The public workflow is more explicit, more documented, and better. The act of explaining the workflow to other people is the part that made the workflow better.The second is that the AI community on social media is generous. I expected to be eviscerated for posting failures, or for posting prompts that were not as clever as other people's prompts. The opposite happened. People who had better prompts told me what made them better. People who had been stuck on the same problem told me what worked for them. The community is generous because the work is hard, and the people who do the work know the work is hard. The generosity is real, and it is the part I did not expect.The third is that posting the AI outputs is the single best marketing for the prompt pack. The pack has been live for about 6 months. The first 3 months, when I was not posting, the pack made about $14 x 8 in sales. The second 3 months, when I was posting, the pack made about $14 x 47. The pack is the same pack. The price is the same. The marketing is the public posting, and the public posting is what the pack is built to be marketed by.What the experiment does not proveIt does not prove that posting AI outputs is a good strategy for everyone. I have a niche (small-business marketing managers, developers, prompt tinkerers) and a voice (practical, direct, teacherly) that work for this kind of content. Other niches and other voices might not work. The experiment proves it works for me. The experiment does not prove it works for you.It does not prove that the social media lift in the prompt pack sales is sustainable. The lift is correlated with the posting, but correlation is not causation, and the prompt pack might be lifting on its own without the posting. The only way to know would be to stop posting for 3 months and see if the sales drop. I am not going to do that, because the posting is also useful for me (I learn from the comments, the failures are good content, the public feedback is the part I would miss). The lift is correlated. The lift is also free. The lift is the part I would keep even if the sales dropped.What I would tell past-meIf I could go back and tell past-me one thing, it would be: post the failures, not just the wins. The failures are the part that makes the wins real. The failures are the part that makes the teaching work. The failures are the part that makes the community generous. The wins are easy to post. The failures are hard. The failures are the post.I would also tell past-me: reply to every comment for the first 7 days. The first 7 days are the part that determines whether the post reaches anyone. The replies are the part that determines whether the post is worth the reach. The replies are not optional. The replies are the part that converts reach into a conversation, and a conversation is the part that produces the next post.Who this is forAnyone who uses AI tools and is considering posting the work publicly. The case is strongest if you are in a niche where the work is the marketing, which is most niches. The case is weakest if the work is private, which is some niches (e.g. medical, legal, financial). For the public-niche case, the posting is the marketing, the teaching, the feedback, and the practice. The posting is the cheap part. The posting is the lever.The honest partI have posted about 60 outputs publicly in the last 3 months. Of those 60, about 10 did really well (5,000+ impressions, 50+ comments, 100+ likes). About 20 did OK (500-5,000 impressions, 5-50 comments, 10-100 likes). About 30 did not do much (under 500 impressions, under 5 comments). The distribution is what I expected: most posts do not do much, some do OK, a few do really well, and the really-well ones are the ones I learn the most from.The posting is also the part of the work I was most resistant to starting. I had a mental model that posting was "selling" and selling was bad. The mental model was wrong. The posting is teaching, and teaching is good, and the teaching is the part I should have been doing all along. The posting is the part that makes the AI workflow public, and the public part is the part that makes the AI workflow better.Want a prompt pack to share publicly?The Blog Writing Factory is built to be shared publicly. The prompts are written, the outputs are written, the failures are written, the teaching is written. The pack is the system. The posting is the marketing. The marketing is the part that does not feel like marketing, because the marketing is the teaching, and the teaching is the work.Buy on Gumroad. $14

Send me the rough edges if you try it. I read every message.