Midjourney for Native Ads on Taboola & Outbrain (2026)

Every media buyer I talk to wants the same shortcut right now: point Midjourney at a prompt, generate a stack of gorgeous images, and feed them straight into Taboola and Outbrain. We ran that exact experiment as an A/B test against the creative system we've refined over the past decade, and the AI images lost. Badly.

That doesn't mean AI image tools are useless on native traffic. It means most people are using them in the one role they're worst at. Below is what the test actually showed, and where a Midjourney image still earns its slot in your account.

I'm Marcel Sattler, founder of native-advertising.net, and since 2015 I've deployed more than $100M across Taboola, Outbrain, Newsbreak, MGID, Yahoo Native, Mediago, and RevContent in DTC, lead-gen, and affiliate. So when I say we tested this, I mean we ran it through the same A/B framework we use to vet every new creative angle before it touches a client budget.

Do Midjourney images work as standalone Taboola and Outbrain ads?

No. As of our December 2023 test, a Midjourney image used as a standalone native creative is not competitive with our best-practice approach on Taboola or Outbrain.

We didn't guess at this. We took our best-practice ads — the templates we've built over the last decade — and A/B tested them against ads built purely on Midjourney output. Same offers, same accounts, same audiences. The Midjourney standalones produced poor performance across the board, with click-through rates well below the control.

The reason is counterintuitive. Midjourney images are too good. Copy a decent prompt off any website and you get something that looks like a polished stock photo — clean lighting, perfect composition, magazine-grade aesthetics. That is exactly the wrong signal for a native feed. People scroll past ads that look like ads, and a flawless image reads as a brand placement, not as editorial content worth a click.

Why "too perfect" kills your click-through rate

Native traffic on Taboola and Outbrain lives inside content feeds on sites like USA Today. The creatives that win there don't look professional. They look native — like a photo someone snapped, a slightly ugly before/after, or a curiosity gap your brain can't ignore.

A Midjourney render fights that instinct. The texture on AI people still isn't natural; skin and hands give it away if you have a trained eye. On a small feed thumbnail most users won't consciously clock that it's AI, but they read it as "advertisement," and the click never comes.

Here's the test I want you to run yourself. Scroll any Taboola or Outbrain feed for two minutes and watch what your own thumb stops on. It won't be the immaculate product shot. It'll be the weight-loss ad showing a shrinking belly, the messy listicle thumbnail, the stitched-together image that plays on curiosity. Those are the units pulling the highest CTRs, and not one of them looks like it came out of Midjourney.

If you want a deeper breakdown of what a winning unit actually contains, our perfect native ad framework walks through the components piece by piece.

What actually works on the feed instead

The ads winning on Taboola and Outbrain right now share a few traits, and none of them require a render engine:

• Curiosity over beauty. The same girl holding an Amazon package against different backgrounds beats a clean product-only shot every time. The product alone is boring; the curiosity gap converts.
• Listicle thumbnails. Slightly busy, slightly low-fi listicle images consistently outperform polished single-product creatives.
• "Ugly" before/afters. Weight-loss and health angles built on raw, unglamorous visuals catch attention far better than any AI picture. They look far from awesome — and that's the point.
• Stitched composites. A few pills, a quick GIF, a couple of stock photos pulled together in Photoshop. Simple, fast, and effective.

None of this takes long. A workable native creative in this style is honestly a couple of minutes of assembly work. The skill isn't in the rendering — it's in knowing which curiosity angle to lead with for your offer and vertical.

If you're running DTC or dropshipping, this is the difference between scaling and stalling — see how we apply it on /solutions/ecommerce. For lead-gen advertisers, the same curiosity logic drives our work on /solutions/lead-gen.

Where Midjourney does belong in your native workflow

This is the part most people miss. Midjourney isn't worthless on native — it's just a component, not the finished product.

Use AI images for the pieces you can't easily source anywhere else:

• As a background. Generate a scene that doesn't exist as a stock photo, then layer your real subject, product, or curiosity element on top of it.
• As one element in a composite. When you need a specific object or setting that no stock library or Google image search will give you, Midjourney fills the gap inside a larger stitched creative.
• For product mockups. You can take a generated bottle or package, swap the label, and turn it into a private-label visual — useful for concept and iteration.

The pattern that beat the AI standalones in our own feed audits combined approaches: real photos, different backgrounds, the same model with an Amazon package, and one AI-generated element working together. The AI didn't carry the ad. It supported it. That's the role to give it on Taboola and Outbrain affiliate, lead-gen, and DTC campaigns.

Midjourney is techy to set up — but that's not the bottleneck

People assume the hard part is the tooling. It isn't. Midjourney runs through a Discord server, which sounds intimidating, but any 10-minute YouTube walkthrough gets you set up. Prompting takes practice — I'm not a prompt expert, and there are people writing far better prompts than I can — but you can produce genuinely high-quality images quickly.

The bottleneck is strategy, not setup. A perfect prompt that produces a perfect image still produces a losing native ad if you drop it in as a standalone. The hours you'd spend chasing the ideal prompt are better spent understanding why the ugly curiosity creative converts and the beautiful one doesn't.

And there are dozens of tools in this space, not just Midjourney. In the 365 days before this test, a wave of AI image tools hit the market — some general-purpose, some dedicated to ads, with Meta building its own AI creative tools for Facebook. We tested several and concentrated on Midjourney because it's the most famous and has the deepest library of public prompts to pull from. The conclusion held across the tools we tried.

Will AI images work for native ads soon? Yes — but not yet

I'm bullish on where this goes. The textures are getting more natural every quarter, and the "too perfect" tell that hurts these images today is exactly the kind of flaw the next generation of models fixes.

When I recorded the original test in December 2023, I predicted that a year out we'd be five steps ahead, and that AI images might become viable standalones for Taboola and Outbrain. The trajectory has borne that out — the gap keeps closing. But the principle that matters for your budget hasn't changed: test it against your control before you scale it. If a prompt or a tool genuinely beats your best-practice creative in a clean A/B test, use it. Until it does, treat AI output as a component and let curiosity-driven creatives carry the click.

If you've found a prompt or setup that beats a classical native creative head-to-head, I genuinely want to see it. Until then, the data says combine, don't replace.

Watch the full breakdown

Is your account a fit for the same play?

If you're spending real money on Taboola or Outbrain and your creatives lean on clean, AI-perfect images, you're likely leaving click-through rate on the table. The fix isn't more rendering horsepower — it's a creative system built on curiosity, tested against a control, and refined the way we've done it across $100M+ in native spend.

Book a strategy call at /contact and we'll look at your current creatives and tell you where AI helps and where it's hurting you. If you want proof first, our case studies show what the best-practice approach does at scale, and you can browse every breakdown like this one in /resources.