How to Scale Ad Creative With AI Without Scaling the Waste

To scale ad creative with AI without scaling the waste, spend the new volume on more tested variants, not on cheaper published ones, and make every asset pass a quality and compliance check before it gets a dollar. The lever is worth pulling: in NCSolutions' meta-analysis of roughly 450 CPG campaigns, creative accounted for 46% of incremental sales on social campaigns and 49% across all formats, ahead of every media factor combined. And AI creative does perform. In a field study by NYU Stern and Emory researchers, ads generated entirely by visual AI lifted click-through rates by up to 19% over ads built by human experts.

The catch sits on the other side of the screen. Half of US consumers told Gartner in October 2025 they would rather give their business to brands that do not use generative AI in consumer-facing content. Both findings are real, and this post is about the pipeline that survives both.

Why is ad creative the highest-leverage place to put AI?

Because creative decides more of the outcome than anything else you buy. NCSolutions analyzed 18 features across five keys of advertising effectiveness (creative, brand, targeting, reach, recency) over almost 450 campaigns, and creative won by a wide margin.

Contribution to short-term incremental sales	All formats	Social campaigns
Creative	49%	46%
Media (targeting + reach + recency)	30%	28%
Brand factors	21%	26%

Source: NCSolutions, via MarketingCharts (2023).

Creative's share has not moved since the previous edition of the same study, while brand climbed from 15% to 21% and reach fell from 22% to 14%. Targeting, the thing most marketers name first, contributes 11%. Nielsen's earlier work found media's effect on sales rising from 15% to 36% over eleven years with creative on top throughout.

Does AI-generated ad creative actually work?

Yes, with a specific and counterintuitive condition: it works best when AI creates the whole asset rather than touching up yours.

NYU Stern's Anindya Ghose, PhD student Hyesoo Lee, and Vilma Todri and Panagiotis Adamopoulos at Emory ran lab experiments plus a four-month field study on Google Ads across three kinds of visual ad: human expert-made, human-made then modified by AI, and fully AI-generated. Their paper, The Impact of Visual Generative AI on Advertising Effectiveness, reports that fully AI-created ads raised click-through by up to 19% against the human expert benchmark, while AI-modified human ads showed no significant improvement at all.

That second result is the useful one. Most businesses work the AI-modified way: take the hero shot from last year's photoshoot, ask a model to swap the background, ship it. That buys you nothing measurable. Let the model start from the product and the brief instead.

Which models can actually do this is a separate question, covered in AI image generation in 2026 and, for motion, AI video in 2026.

What happens when you tell people the ad was made by AI?

The same field study measured it, and the answer is uncomfortable: disclosing AI involvement was associated with a 31.5% decrease in click-through relative to the human expert ads.

Set that next to the survey evidence. IAB and Sonata Insights asked 505 US Gen Z and Millennial consumers and 104 ad executives about exactly this between October 2025 and January 2026. In The AI Ad Gap Widens, 73% of those consumers said knowing an ad was created with AI would increase or make no difference to their purchase likelihood. IAB reads that as disclosure having more upside than downside.

So people say disclosure is fine and behave as though it is not. Two things are still safe to act on.

The perception gap is widening, and it is generational. IAB found 82% of ad executives believe Gen Z and Millennial consumers feel positive about AI-generated ads, against the 45% who actually do; that 37-point gap was 32 points in 2024. Among Gen Z alone, 39% report negative sentiment, nearly double the 20% of Millennials, and 30% call brands that use AI for ads "inauthentic."

And disclosure is drifting from a choice to an obligation. The rules landing on customer-facing AI are in US AI chatbot disclosure laws in 2026. Build for the version where you have to say it.

What does the extra volume actually buy?

Not cheaper ads. More shots on goal.

This is where most AI creative programs go wrong. Cost efficiency became the top cited benefit of AI in advertising in 2026, named by 64% of advertisers in the IAB study, up from fifth place in 2024. IAB is blunt about the risk: advertisers focused on cost efficiency alone "should know that it can't come at the sacrifice of quality."

Do the arithmetic instead. A clinic that could afford four concepts a quarter was never really testing; four data points across three months tell you almost nothing. The same clinic generating forty variants in a month can run a real test. The constraint moves off the camera and onto judgment: which three of the forty deserve budget, and how you tell before you have paid to find out.

Gartner's number matters commercially here, not just ethically. In the same survey of 1,539 US consumers, 68% said they frequently wonder whether the content they see is real. An audience in that state punishes generic output, and unfiltered volume is a generic-output machine.

What does a working AI creative pipeline look like?

Six steps. Numbers 3 and 5 are the ones people skip.

1. Brief from real customer language. Pull objections out of your actual DMs and calls, not a persona document. The hooks that work are usually sentences a customer already wrote.
2. Generate from the product, not from an existing ad. Per the NYU findings, let the model compose the whole asset from your catalog images and the brief.
3. Score before you spend. Check every asset for hook strength, unsupported claims, and category compliance before it is eligible for budget. This filter is what turns volume into testing rather than noise.
4. Publish in small, comparable sets. One variable per set. A thirty-asset dump into one ad set teaches you nothing.
5. Measure to booked revenue, not clicks. A 19% click lift that produces no extra bookings is just a more expensive way to lose money. Send real conversion events back to the platform so bidding optimizes against outcomes.
6. Feed the winners back into step 1. The hook that booked appointments is your next brief.

Step 5 is the honest limit of every ad-side optimization, ours included. The buying-side version of this argument is in how to use AI for Meta ads without burning budget.

Why is click-through the wrong finish line?

Because a click on a Meta ad for a clinic, a salon, or a Shopify brand usually becomes a message, and an unanswered message is a lost sale however good the creative was.

In the Entagl Response Velocity Study (2026) we looked at 32,581 conversations across 1,247 businesses in nine countries. Conversations answered within 60 seconds converted at 35.1%; at one to twenty-four hours, 7.1%. In 5,201 competitive multi-vendor inquiries we could link to a purchase, 78.4% of sales went to whichever merchant replied first.

Put those next to the 19% click lift. You can win the creative and still hand most of the demand it produced to whoever answers faster. That argument, at length: why DMs drive revenue.

Where does Entagl fit in this?

Entagl runs the whole loop rather than one leg of it, which is the only reason step 5 is answerable.

Studio generates product images and e-commerce video from your real catalog, and every asset is analyzed before it can be promoted: hooks, claims, category compliance flags, and a readiness score for paid distribution, with industry-specific rule packs applied. Talking-head and avatar video with consent-based voice cloning is available and newer, so pilot it rather than building a campaign on it.

Ads Co-Pilot watches the Meta account and files proposals with its reasoning attached: a budget change, a pause, a creative rotation. It proposes; a human approves or rejects. Each proposal records who decided and when, and stale ones expire rather than firing late. There is an autopilot, but it is off by default, opt-in per ad account, and bound by a hard daily spend ceiling, and it can never auto-approve a brand new creative. The default posture is a proposal a person signs off.

Receptionist answers what the ads produce, across WhatsApp, Instagram, Facebook Messenger, Telegram, web chat, email, and an API, in 100+ languages, and books into a real calendar. Coordinator calls to confirm and recover no-shows, already knowing the chat history. Bookings post back to Meta through the Conversions API, so the next ad dollar bids on customers, not clicks.

What this evidence does and does not prove

These numbers get quoted loosely, so:

• The NCSolutions meta-analysis is CPG campaigns. Creative's roughly-half share is directional for a med spa or a Shopify store, not a measurement of one.
• The NYU and Emory study measures click-through, not revenue. A click lift is not a profit lift.
• Gartner's 50% is a stated preference, not observed purchase behavior. The IAB disclosure figure has the same limitation, which is exactly why it disagrees with the field experiment.
• Nobody has published a clean read on whether the advantage holds up over quarters as audiences see more AI creative. Assume decay and keep measuring.

FAQ

Does AI-generated ad creative perform better than human-made ads?

In the NYU Stern and Emory field study, ads generated entirely by visual AI raised click-through by up to 19% over ads made by human experts, while human-made ads edited by AI showed no significant improvement. Fully AI-created assets outperformed in that setting and half-measures did not. It has not been shown to hold for revenue, or across every category.

Do I have to disclose that an ad was made with AI?

It depends on your jurisdiction and category, and the requirements are expanding. Healthcare and financial services draw the most pressure, and IAB found consumers rank pharmaceutical and political ads highest for disclosure importance. Assume you will need to say it, and build creative that works when you do rather than creative that depends on nobody knowing.

How many creative variants should a small business test at once?

As many as you can judge properly and fund to a readable result, which is usually far fewer than you can generate. One variable per comparable set. If you cannot put enough spend behind each variant to separate winners from noise, you are publishing, not testing.

Will AI creative damage my brand?

It can, and the mechanism is quality, not the technology. Gartner found 68% of US consumers frequently wonder whether the content they see is real, and IAB found 30% of Gen Z describe brands using AI for ads as "inauthentic." The businesses getting hurt are the ones shipping unfiltered volume. Gate every asset on brand and compliance rules before it gets budget and the risk drops sharply.

The one thing to change this week

Add a gate. Pick the three checks that matter in your category (hook, claim accuracy, compliance), apply them to every asset before it gets a dollar, and measure the survivors against bookings rather than clicks.

If you want to see what that looks like with the creative, the ads, the replies, and the bookings in one place, book a 30-minute demo. We will walk your actual ad account and your actual DMs.

---

Sources: NCSolutions five keys to advertising effectiveness, via MarketingCharts (2023) · Nielsen, advertising effectiveness (2017) · Ghose, Lee, Todri & Adamopoulos, The Impact of Visual Generative AI on Advertising Effectiveness (SSRN, Nov 2025) and the NYU Stern research summary · IAB and Sonata Insights, The AI Ad Gap Widens (Jan 2026) · Gartner consumer survey (March 2026) · Entagl Response Velocity Study (2026). Figures accurate as of September 2026.