How to turn your comments into a month of content ideas with AI

For about two years I brainstormed content ideas the hard way: staring at a blank document, trying to guess what my audience wanted, and picking whatever sounded clever at 11pm. The posts were fine. They were also, in hindsight, mostly about what interested me.

Then I did something obvious that I should have done on day one. I copied every comment, DM and reply from the previous three months into a document, handed it to an AI, and asked it to tell me what people kept asking. I got 34 ideas out of one session, and the six I published first outperformed everything I’d written that quarter.

Here’s the workflow, exactly as I run it.

Why comments beat brainstorming

Your comment section is the only research source where people volunteer their problems in their own words, unprompted and for free. Keyword tools tell you what strangers type into a search bar. Your comments tell you what your audience is stuck on, in the phrasing they’d actually use, with the emotional temperature attached.

That last part matters more than people think. “How do I edit faster?” and “I spent my whole Sunday editing one video and I want to cry” are the same topic, but only one of them tells you the headline.

The catch is volume. Once you have a few hundred comments across a few platforms, you can’t hold the patterns in your head. You read the last twenty and think you know what people want. You don’t, you know what the last twenty said. That’s the exact job AI is good at: not inventing, just clustering.

Step 1: Gather the raw material

Set a timer for twenty minutes and collect everything you can into one plain text document. Don’t clean it, don’t organise it, don’t skip the ones that seem trivial.

• YouTube, TikTok and Instagram comments. Scroll back three months. Copy in bulk, ignore the emoji-only ones.
• DMs and replies. These are gold, because people ask things in private they won’t ask publicly.
• Email replies. If you run a newsletter, your reply-to inbox is the highest-signal source on this list.
• Support or sales questions. Every “does it work with…” is a post.
• Your own saved notes of questions people asked you on calls or in real life.

I keep the raw dump in a Notion database with one row per mining session, so I can look back at what people were asking six months ago and see how it shifted. A plain text file works just as well on day one.

Step 2: Cluster, don’t brainstorm

This is the part everyone gets wrong. If you ask an AI for content ideas, it will happily generate twenty generic listicles that could belong to anyone. You will use none of them.

Instead, paste the dump and ask it to do analysis:

Below are real comments and questions from my audience. Group them into themes by the underlying question being asked, not by surface wording. For each theme give me the recurring question in plain language, how many comments map to it, and three verbatim quotes. Do not add themes that aren’t represented in the text.

That last sentence does a lot of work. It gives the model permission to report a small number of themes rather than padding to a nice round ten.

What comes back is usually humbling. My biggest cluster wasn’t the sophisticated stuff I liked writing about, it was people asking, over and over, some version of “which one should I buy”. Which is why half my calendar is now comparisons.

If you’re new to prompting for real work rather than autocomplete, the AI content workflow I copy every single week covers the wider system this slots into.

Step 3: Turn themes into angles

A theme is not a post. “People are confused about AI voice tools” is a theme. You need angles, and this is where you can let AI generate, because now it’s working from real constraints.

For each theme, I ask for three angles at different levels: a beginner explainer, a specific comparison, and a workflow post. Then I pick whichever one I have something genuinely useful to say about, and cut the rest.

• The explainer answers the question directly for someone who just arrived.
• The comparison serves the “which should I buy” crowd, which is most of them.
• The workflow shows your process, which is the one that builds trust.

Keep the verbatim quotes attached to each angle. When you sit down to write, opening with a real question someone actually asked is the fastest way past a blank page.

Step 4: Load the calendar

Now you have thirty-odd angles tied to real demand. Drop them into whatever you plan in, assign dates, and stop deciding what to write on the day you write it.

I sort mine by two things: how often the question came up, and whether I can answer it properly. Frequent plus confident goes first. Frequent but I’d have to research goes in the middle, so I have time. Rare but I have a strong opinion goes in as a palate cleanser.

If you don’t have a planning system yet, how to build a 30-day content calendar with AI is the companion piece to this one. That’s where these ideas go to become an actual schedule.

Step 5: Close the loop

The workflow only compounds if you tell people you heard them. When I publish something that came out of a comment, I reply to the original comment with the link. It costs nothing and it does two things: it makes that person far more likely to comment again, and it publicly demonstrates that commenting here gets you answers.

More comments means better mining next month. That’s the whole flywheel.

My take

The reason this works isn’t the AI. It’s that most of us have been guessing when the answer was sitting in our notifications the entire time. The AI just makes a pile of 400 comments readable in ten minutes instead of unreadable forever.

Run it once, properly, with real data and a clustering prompt rather than an ideas prompt. Even if you throw away half the output you’ll come away with more useful direction than a month of brainstorming. Then put the good ones in a queue with real dates attached, because an idea without a publish date is a note, not a plan. I keep mine in Notion alongside the raw dumps, and when the calendar feels thin I don’t panic, I go mine again. If you want the wider repurposing angle, turning one blog post into 10 pieces of content is what I run on whichever of these lands best.