Why does a YouTube video still flop after testing 25 different thumbnails?
by Alessandro · updated 4 November 2026
Testing 25 thumbnails will not fix a video if the packaging points the algorithm at the wrong audience. On Risk Takers, I packaged a SALT interview with a massive guest around the midterms and tried over 20 different thumbnails. All of them failed, because my audience wants market talk, not politics, whatever picture or title sits on top of it.
In short
- Thumbnail testing cannot fix a mismatch between the topic and what your audience actually watches for
- A big guest does not guarantee views if the packaging points at the wrong interest
- Packaging is roughly half the job, but trying 20-plus versions is still guessing at the angle, not the image
- A guest telling you the conversation was great does not mean the video will perform
- Letting the guest talk, with open questions, is a separate skill from packaging, and the one that keeps guests coming back
The interview that should have worked
I did an interview at SALT with Anthony, a political commentator who had already been on the show once before. It's done fucking terribly. One of my worst shows in the last 10 uploads, and that's against a guest who is, by any normal measure, massive.
I really thought I'd nailed it. The last episode I ran, packaged around the midterms, pulled over 100,000 views, so leaning into the midterms again for a guest who talks politics for a living seemed obvious. It wasn't. Going into it I was piece by piece solving what the packaging and the audience actually wants, and this piece of the puzzle turned out to be wrong.
What 25 thumbnails actually proved
I trialed 25 different thumbnails on this one. Tried a lot. None of them hit. When asked whether it was the thumbnail or the angle of the conversation, my answer is always the same: it's always the thumbnail, it's the packaging, and packaging is fifty percent of the job.
But 20-plus failed attempts on the same video tells you something past the usual advice to just test more thumbnails. If the topic itself is wrong for the audience, no version of the image fixes it. I tried that many packages and they all failed, which is the point: the testing wasn't the problem, the premise under the testing was.
Reading the real mismatch
My read on it, once I stepped back, was that my audience wants market talk, not politics, and I'm not a political expert. Everything is related in the sense that politics drives regulation and regulation drives what's possible economically, but that connection wasn't enough to carry a video that was, in substance, a conversation about the midterms rather than markets.
The timing made it worse. We filmed before the midterms had actually been decided, even though prediction markets already had it at 75 to 80 percent for the Democrats. We were discussing the reasons the Democrats were likely to win that turned out to be the same reasons they did win, political donations, Citizens United, labour's shrinking share of GDP against a rising share going to asset owners. All of it held up. None of it was what the channel is packaged for.
Why the conversation still worked, even though the video didn't
YouTube is phenomenal at putting the right video in front of the right people, if it's packaged right. So I still don't think the video itself was bad. After we filmed, Anthony told me I ask great questions, the second time he's said that after being on the show twice. That's not a line he needed to repeat.
If someone found this video for the first time with an interest in markets, or in my perspective on this from outside the US, they would have learnt something, which is essentially the point of doing it. The conversation was insightful. The packaging just didn't carry it to anyone who was going to care.
The hostage negotiator lesson I use in every interview
What I apply to every conversation, learned from a hostage negotiator, is to let the person talk, and to ask open-ended questions rather than questions that box someone into two possible answers. The example I give is asking whether the midterms are important from the perspective of data centres or fiscal spending, when the guest might have had something completely unrelated to either. A better question is just: why are these midterms important?
People generally like talking to me because of this, not because I'm the most interesting person in the room but because I'm the most interested. You've met that person at a party, the one who asks you questions and keeps you talking and makes you feel interesting, and you leave liking them without learning much about them. That's a skill, and it's the one that gets guests like Anthony to come back a second time and say the questions were great, independent of whether the video performs.
What this costs in time
Twenty-five thumbnails is not a quick afternoon. It's a full packaging cycle run against one episode. None of that time bought the view count. The guest, the research, the questions and the packaging attempts are all separate jobs, and failing at the last one doesn't mean the earlier ones were wasted, it means they're sitting on a video that nobody who'd value it is finding.
Who this applies to
This is for anyone running a channel with an established audience who tries a guest or topic slightly outside what that audience already comes for. A guest's size, a topic's apparent relevance, and even your own confidence going in are not signals you can trust over what the audience has already shown you it watches for. If your last upload in one lane did 100,000 views and the next one in a different lane does a fraction of that, don't keep iterating on the thumbnail. Iterate on whether the topic belongs on the channel at all.
Questions people ask
Does a bigger guest always mean more views?
No. A massive guest with a strong, complimented conversation still underperformed because the topic, politics rather than markets, didn't match what the audience on this channel actually watches for.
How many thumbnails should you test before changing the topic instead?
There's no fixed number, but trying over 20 and getting nothing back is a sign the packaging isn't the problem. If every version of the image fails the same way, the issue is likely the subject underneath it, not the design.
Is packaging really half the result on YouTube?
That's the stated split: packaging is fifty percent. The other half is presumably the content and the fit with the audience, which is what seems to have gone wrong here even with the packaging effort maxed out.
What makes an interview question open-ended rather than closed?
An open question doesn't hand the guest two predefined answers to choose between. Asking whether the midterms are important from a data centre or fiscal spending angle boxes the guest in; asking simply why the midterms are important leaves room for an answer you didn't anticipate.
If the conversation is good but the video flops, was the interview wasted?
Not necessarily. The guest said the questions were great, for the second time, and the substance held up against what later happened. The interview can succeed on its own terms while the video still fails to find an audience.