I post about AI tools on TikTok to an audience of over a million people. That means if I say something wrong, a lot of people hear it wrong.
So I have a research process. And AI is now a core part of it — not to write my scripts (I do that myself) but to make sure what I’m saying is actually accurate, current, and interesting.
Here’s the full workflow.
Step 1: Idea Generation
Ideas come from everywhere — my feed, my DMs, newsletters, podcasts. But I use ChatGPT to help me qualify ideas before I spend time on them.
My prompt, roughly:
“Here’s a topic I’m thinking about for a 60-second TikTok: [topic]. Who cares about this? What’s the angle that would make someone stop scrolling? What’s the most surprising or counterintuitive thing about it?”
I don’t take ChatGPT’s output as gospel — I’m looking for angles I hadn’t considered. It’s brainstorming with a very fast thinking partner.
Step 2: Research
Once I’ve got an angle I like, I go to Perplexity Pro.
Perplexity is my research starting point because it cites everything. I can ask:
“What do users actually think of [AI tool]? What are the main complaints? What do power users love about it?”
And it pulls from Reddit threads, YouTube reviews, newsletters, and tech journalism — all cited so I can click through and verify. I’m not relying on Perplexity’s summary as fact; I’m using it to rapidly identify what I should go read directly.
For bigger topics, I’ll also check Google Trends to see if search interest is rising or falling. There’s no point making a video about a tool that peaked 8 months ago.
Step 3: Fact-Checking
Here’s where people cut corners and get burned.
Before I make a claim on camera, I verify it. I use Claude for this:
“I’m about to say that [AI tool] has a free tier with X features. Based on what I’ve shared, is that still accurate? What should I double-check?”
Claude is good at flagging where my information might be stale or incomplete. Combined with a quick visit to the tool’s actual website, this takes about 5 minutes and has saved me from embarrassing corrections multiple times.
Step 4: Script Structure
I outline in Notion AI, but not the way you’d expect.
I dump everything — my research notes, my personal experience with the tool, the angle I’m going for, and what I want the viewer to feel at the end — into a Notion doc. Then I ask the AI Q&A:
“What’s the most compelling hook I could use for this video based on these notes?”
Again — I’m not using the output directly. I’m mining it for one good idea. Then I write the script myself.
My TikTok voice is mine. The AI helps me find the angle. The words are still me.
Step 5: Final Check Before Filming
Last step before I hit record: I re-read the script and ask myself three questions:
- Is every factual claim I’m making something I personally verified?
- Is there anything here that could be misunderstood or out of date?
- Is this actually useful, or am I just filling time?
If I can’t confidently answer yes-yes-no, I don’t film it yet.
Why This Matters
There’s a temptation in this space to move fast and fix mistakes in the comments. I get it — the algorithm rewards volume. But my audience trusts me because I’m careful, and that trust is worth more than an extra post.
AI lets me move faster without cutting corners. That’s the point.
— Deco