Lab Notes · 3
Ask an AI for 30 Blog Titles — What Patterns Emerge?
Whenever I'm stuck on a blog title I ask an AI, but one day I noticed the suggestions all felt strangely similar. To check whether that was just in my head, I had it generate 30 titles for a deliberately plain topic — "the pros and cons of remote work" — and classified every single one.
Method
- Fixed topic: "a blog post about the pros and cons of remote work"
- The request was simply "suggest 30 blog titles" (no other conditions)
- All 30 titles sorted by type, by hand
Results: the 5 title formulas AI loves
| Pattern | Count | Example |
|---|---|---|
| Numbered list | 9 | "7 Reasons Remote Work Is Great" |
| Question | 7 | "Remote Work: Is It Really More Productive?" |
| A vs. B contrast | 6 | "Remote vs. Office: Who Wins?" |
| How-to / guide | 5 | "How to Do Remote Work Right" |
| Confession / experience | 3 | "The Truth from 3 Years of Working Remotely" |
27 of the 30 titles fell into those five formulas. It wasn't a feeling — the patterns really were that strong. Numbered titles ("N reasons/ways…") alone made up a third.
I also experimented with breaking the patterns
Next, I changed the request.
- Adding "no clichéd titles" — fewer numbered titles, more metaphorical ones ("My Living Room Became My Office"). Simple, but it works.
- Assigning a role: "like a newspaper columnist" — full-sentence titles started appearing ("What Remains Where the Commute Used to Be"). The tone changed noticeably.
- Showing 3 well-written example titles first — many titles followed the grain of the examples. This gave the best control.
What I learned
- Without constraints, AI converges on the safest, most common formulas. Given how many such titles are in the training data, that's no surprise.
- But those formulas aren't bad — they survived precisely because they're proven click-getters. If you need safe, use them as-is; if you need personality, shake things up with banned words, roles, and examples.
- My practical workflow: 30 titles with no constraints (to expand ideas) → pick 2–3 directions I like → "10 more in this direction, no clichés." This two-step loop was the most efficient.
The title of this very post is a byproduct of the experiment — it's the question formula. I believe there's a difference between using a formula knowingly and using it blindly.