Sabitlenmiş Tweet
Today, I’m excited to introduce Valid AI Creative.
We've built a system to make AI ads that actually convert.
Brands like Replit, Tolans, HoneyHomes, and more already trust us.
Now we're opening this up to the world.
English
Rahul Bhardwaj
307 posts
Rahul Bhardwaj
@rahulb_co
Founder and CEO @ Valid, Ex-Meta, Ex-Amazon.
San Francisco, CA Katılım Temmuz 2020
215 Takip Edilen1.4K Takipçiler
I keep hearing three objections to AI creative. The data doesn’t support them.
→ “AI creative feels inauthentic—audiences can tell.”
Meta beta tests saw +10% CTR and +8% CVR with AI creative. It works.
Caveat: for categories where realism is the product (apparel, skincare, beauty), authenticity matters more.
→ “The compliance risk is too high.”
FTC rules don’t change for AI vs. human ads—just disclose when AI is used.
→ “We’ll adopt when the tech matures.”
It already has. Look at TikTok’s “AI-generated” label: plenty of ads are fully production-grade.
And as platforms reward creative variety, it’s getting tough to keep up without AI volume.
Which objection are you hearing most?
English
After managing $450M+ in ad spend and making probably 15,000 ads+…
These are the non-negotiable rules I run Fraggell with.
(This will probably help you out too)
English
Valid’s ad stack runs on AI models that could lose key capabilities overnight. That’s unsettling.
The obvious response is to hedge: contingency plans for every deprecation, an architecture built to survive any platform going dark. I think that’s the wrong bet.
I’m betting on a Moore’s Law for model quality: costs drop, access expands, and capability keeps improving.
Facebook had bad months—algorithm shifts, advertiser pullbacks. The best teams didn’t build bunkers; they kept shipping. I see AI platform risk the same way. Regressions will happen, but they’re noise on an up-and-to-the-right curve.
Over-hedging slows you down: fewer experiments, slower deployment, getting passed by teams that keep moving.
If you’re building AI-native: how model-agnostic are you, really?
English
@Jake_Joseph Let’s go! Zoe has more views/post than any creator we’ve ever made.
Can’t wait to see how clients benefit from that.
English
Rahul Bhardwaj retweetledi
Zoe racked up millions of views last week.
She's not real.
Backed one of the best teams in the business, now anyone can book her for ugc.
Who's brave enough to try it?
Book below. 👇🏼
Jake Baumann@Jake_Joseph
Old people on Facebook are absolutely cooked. GPT Image 2 + Seedance 2
English
We just launched a new type of influencer.
Quick turnarounds, no revision limits, and content that actually converts.
Creators engineered to convert.
"Zoe" will take a look at your ad account, come up with a concept, make it, and send it over.
Get AI-UGC that actually works today.
Jake Baumann@Jake_Joseph
Zoe racked up millions of views last week. She's not real. Backed one of the best teams in the business, now anyone can book her for ugc. Who's brave enough to try it? Book below. 👇🏼
English
We gave Claude Code one job: generate and analyze ad creative ideas. Then we got out of the way.
Audiences churn through creative faster than teams can ship and test. So we built a system that generates and tests at volume.
Claude Code outputs 70 concepts, runs them in parallel, then flags the 4–5 with real statistical lift. Those are the winners.
The real value comes after: patterns.
Claude extracts what the winners share—and catalogs the losers too. Because the system doesn’t reset. Every run grows two living datasets: what drives performance, and what doesn’t.
I’m obsessed with the losers. If you ignore them, your “new ideas” are just recycled guesses with a new coat of paint.
Winners + losers together are what unlock performance at scale.
How are you structuring creative testing—and are your losers informing the next round?
English
Kling 3.0 in 4k mode is so much fun.
Every product shot looks like a +100k motion graphic.
GPT Image 2 + Topaz Upsale + Kling 3.
English
@SmokeySnipe It’s a flywheel. Clients who join us from ads often submit referral as well.
English
@rahulb_co Bold claims but if 65% come from referrals why are you posting this on twitter to strangers lol
English
65% of our clients come from referrals. Here’s why.
We build systems that learn.
Every dollar makes the next one more efficient.
We’re an AI-enabled agency that learns faster, lowers CPA, and scales spend aggressively.
English
Want to hang out with other mobile app founders crushing it right now? If you’re going to MAU Vegas next month, Valid is co-hosting the Consumer Apps Happy Hour.
May 19, 7:30 to 10:30 p.m. @Superwall, @Braavocapital and @Adjustcom are co-hosting with us.
You’ll get plenty of slide decks during the day sessions. This is just drinks, small bites, and a room full of people who build & grow apps for a living.
The sessions will no doubt be awesome, but I find the best insights come from unfiltered conversations with your peers. That’s why we keep hosting these gatherings. You mention a challenge, and it turns out the person you’re talking to solved it last quarter.
If you're going to be at MAU and work in consumer apps, you need to be at this exclusive event.
Want to join? Apply to RSVP at the link or feel free to DM me.
👉 luma.com/mau_happy_hour
English
That video hit ~700K impressions in 24 hours.
Now I'm considering something dumb.
Or genius?
You tell me. 😂
Jake Baumann@Jake_Joseph
Old people on Facebook are absolutely cooked. GPT Image 2 + Seedance 2
English
Who here has the absolutely best "Careers" page?
Especially oriented towards eng hiring
I think Ramp does a great job. But they are later stage. Looking for Seed/Series-A references.
English
Rahul Bhardwaj retweetledi
Old people on Facebook are absolutely cooked.
GPT Image 2 + Seedance 2
English
We ran 2,000+ Sora renders in 30 days. We would have moved off it anyway.
OpenAI isn't shutting down Sora because of GPUs. They're shutting it down because they're losing.
Here's what Sora did well and why it's no longer needed:
Sora's superpower was pre-visualization. The framing, the motion, who the actor should be, how they deliver the lines, what they should look like. Sora was able to do this better than Veo/Kling.
Sora's resolution and realism was never on par with Veo/Kling. But it's shot selection far surpassed them.
Then Seedance showed up and ate Sora's lunch.
Kling level accuracy with better-than-Sora shot composition.
IMO, OpenAI isn't shutting down Sora because they need to reallocate GPUs. It's because ByteDance (Seedance) and Kuaishou (Kling) have the lead and are running away.
Credits to @Jake_Joseph for making this video. GPT Image 2 + Seedance 2.
English
We put typos in our ads on purpose. And it works.
Not for every brand. But for the ones where performance is the priority, deliberate imperfection is a real lever.
I posted an example of this two weeks ago - the video ad that pronounced “hazelnuts” as HAH-zelnuts. We did that on purpose.
We’ve done a ton of A/B testing of these intentional “mistakes” and uncovered a few specific tactics that outperform every time:
→ Typos in text overlays: This gets people to pay more attention. A small error in the copy makes the viewer pause, which increases dwell time.
→ Mispronounced words: Pronouncing “avocado” incorrectly or pronouncing “chocolate” wrong can increase engagement. The viewer hears the error and instinctively wants to correct it, so they leave a comment.
(We’ve actually homed in on the specific mispronounced words that drive the biggest spike in engagement. I have data on which ones.)
→ Cut-off edits: If you cut off half of a word in the edit, that can result in an uptick in engagement. The incomplete word creates an unresolved loop. The brain wants to finish it, so the viewer keeps watching.
→ Incorrect captions: Viewers who are reading along catch the error and stay engaged to see if it continues. Or like the mispronounced words, they leave a comment to call it out.
Perfectly polished creative optimizes for aesthetics. Performance creative optimizes for attention. Make sure you know which goal you’re chasing.
Has perfectionism ever stifled your ad performance?
English
If you aren't writing yet, I highly recommend you start writing anything. For me, it’s dreams, goals, and journaling. It can be whatever. It organizes and amplifies your thinking, and is extremely useful. Try it. 📝
English
@MaxxifiedLian Yeah this is the biggest tell rn. There's some tricks you can do to make it better.
Biggest gap imo is that the models don't handle inflection well yet. We're working on that.
English
@rahulb_co By the way the words are spoken you will know it's AI
English
AI content is going to make Instagram fun again.
Our AI ads already stop the scroll 54% of the time.
Because with AI, we have no creative limits.
Today, the feed is an endless stream of "here's how I went from x to y", "a day in the life", a selfie video from the car. You get the point.
It's not due to laziness. Making any video is really hard.
So creators stick to one format. Then copycats follow.
Suddenly and everything looks the same.
AI-video will reset the bar for what's worth watching.
The only limits on what you can make is... creativity.
The feed will become more diverse. More individualistic. More interesting.
Even on our team. Yes we make AI ads.
But we don't talk about AI. We talk about ideas, script, hooks.
English
@Joonzzy Can you share what creatives you tested on?
It’s much harder to pull off than “traditional” creative. But when you pull it off we’ve found it converts exceptionally well.
English
@rahulb_co Our studies show the conversion rate on these types of ads are at least 40% lower than regular ads. It’s just based on 2 TikTok shop brands tho.
English
Rahul Bhardwaj retweetledi
Our AI ads stop the scroll 54% of the time.
Because there are no creative limits.
We opened up our AI creative at Valid to help brands do exactly this.
The old content playbook is dead.
Rahul Bhardwaj@rahulb_co
AI content is going to make Instagram fun again. Our AI ads already stop the scroll 54% of the time. Because with AI, we have no creative limits. Today, the feed is an endless stream of "here's how I went from x to y", "a day in the life", a selfie video from the car. You get the point. It's not due to laziness. Making any video is really hard. So creators stick to one format. Then copycats follow. Suddenly and everything looks the same. AI-video will reset the bar for what's worth watching. The only limits on what you can make is... creativity. The feed will become more diverse. More individualistic. More interesting. Even on our team. Yes we make AI ads. But we don't talk about AI. We talk about ideas, script, hooks.
English
We're adding 10 new clients this month.
If you're serious about spending on Meta or TikTok and need better creatives, we should talk.
Learn more and apply here: valid.co/creatives
Rahul Bhardwaj@rahulb_co
AI content is going to make Instagram fun again. Our AI ads already stop the scroll 54% of the time. Because with AI, we have no creative limits. Today, the feed is an endless stream of "here's how I went from x to y", "a day in the life", a selfie video from the car. You get the point. It's not due to laziness. Making any video is really hard. So creators stick to one format. Then copycats follow. Suddenly and everything looks the same. AI-video will reset the bar for what's worth watching. The only limits on what you can make is... creativity. The feed will become more diverse. More individualistic. More interesting. Even on our team. Yes we make AI ads. But we don't talk about AI. We talk about ideas, script, hooks.
English