Ad banner audit (Meta/Display)

On Meta, your creative has roughly 0.5 seconds to justify itself before a user scrolls past.

We didn’t pick a safe product shot for this test. We chose a Duolingo creative. Why? Because it is the ultimate test of whether AI understands context, not just pixels.

Duolingo is famous for “breaking” aesthetics. Using memes and melting owls to stop the scroll.

It’s an algorithm trap. We wanted to see if the models could look past rigid design rules to appreciate the hook, or if they would get hung up on the “deformed” mascot.

If your virtual auditor doesn’t get the viral intent, its advice will tank your CTR.

Prompt used: Act as a senior performance marketer and evaluate this creative for readability and CTA strength. Identify one critical error lowering the CTR and propose a specific visual fix to eliminate it. Justify this change with attention psychology, limiting your entire response to a maximum of three sentences.

The results:

• GPT: The biggest CTR-killing issue is that the green background, white text, and white CTA all blend together, creating almost zero visual hierarchy and making the “Install Now” button feel invisible. Add a high-contrast CTA bar (e.g., deep blue or dark purple) with bold white text to create an immediate focal point that the eye snaps to due to contrast bias. This single change redirects attention from the floating elements toward a clear action anchor, sharply increasing click likelihood.
• Gemini: Critical Error: The ad suffers from false affordance, as the three language options visually mimic clickable UI buttons, creating a “decoy effect” that splits user attention and competes with the actual “Install Now” CTA. Visual Fix: Remove the button styling (white containers and drop shadows) from the language list, presenting them instead as a simple, non-interactive list of flags with text or a unified graphic. Justification: This change reduces cognitive load by eliminating the expectation of interactivity within the image, preventing user frustration from non-functional clicks and channeling focus directly to the primary conversion point.
• Grok: The critical error tanking CTR is the extreme visual competition between the massive Duolingo logo and the tiny, low-contrast “Install Now” CTA, causing users’ attention to stick on the brand mark instead of the action button. Attention psychology shows that when a non-interactive decorative element (the logo + owl) dominates the visual weight and occupies the strongest focal area, users’ gaze rarely escapes to the actual CTA. Fix: drastically shrink or remove the Duolingo logo from the right half and replace it with a much larger, high-contrast (e.g., bright yellow or red) “Install Now” button that occupies at least 30-40% of the lower canvas, instantly redirecting focal attention to the single desired action.
• Claude: The critical error is CTA invisibility—the “Install Now” button bleeds into the white background with minimal contrast, making it nearly invisible and violating the principle of visual salience that drives eye-tracking and click behavior. Fix: Make the CTA a high-contrast green button matching Duolingo’s brand color with white text, creating a clear visual hierarchy that leverages the Von Restorff effect (isolation principle) to make the action element pop against the neutral background. This change would dramatically improve CTR by ensuring the conversion point is the most attention-grabbing element in the visual field, as pre-attentive processing prioritizes high-contrast color differentiation.

The verdict:

GPT takes the gold. It correctly diagnosed that in a fully clickable environment like Meta, the key to high CTR is strong contrast and a “Faux CTA”—a visual target the user subconsciously wants to tap.

Gemini slips to second place. Its advice to remove the button styling was actually harmful here. Elements that mimic UI on a clickable image usually increase interaction, not decrease it.

Grok offered a solution that might work mechanically, but it would destroy brand visibility.

Claude comes in last. It suggested changing the color of the platform’s native system button, which is technically impossible.

Thumbnail audit

Treating a thumbnail like a standard photograph is a fast track to zero views. On YouTube, your creative is a micro-billboard. It has fractions of a second to win the battle for attention.

We tested whether AI can distinguish between an “aesthetic” image and a click-magnet.

If a model can accurately judge emotions and composition, it becomes a powerful tool for boosting CTR—and determining if the algorithm gives your video a chance in the first place.

Prompt used: Act as a YouTube Growth Strategist and audit this thumbnail for CTR potential. Identify the single biggest distraction or weakness, propose a specific visual edit to fix it, and briefly explain why this change will stop the scroll. You must provide your answer in a maximum of three short sentences.

The results:

• GPT: The biggest distraction is the long tweet text, which forces viewers to read instead of instantly understanding the message. Replace it with a short, bold phrase like “Is Siri UNDERRATED?” in a large, high-contrast font. This creates immediate clarity, triggers curiosity, and makes the thumbnail readable in under 0.2 seconds—dramatically improving stop-the-scroll performance.
• Gemini: The text-heavy tweet creates visual clutter that distracts from the creator’s expression and forces the viewer to read too much. Replace the screenshot with short, massive text like “SIRI IS GOOD?” directly on the background. This reduces cognitive load, ensuring the viewer instantly grasps the concept and stops scrolling without the friction of reading.
• Grok: The single biggest weakness is the cluttered Twitter screenshot overlapping the face and text – it pulls attention away from the emotional reaction and makes the thumbnail feel messy. Completely remove the Twitter bubble and replace it with bold, high-contrast yellow text directly on the image saying “SIRI ISN’T THAT BAD”. This instantly turns the thumbnail into a clean, curiosity-driven reaction shot with a controversial statement that forces people to click to see why he’s smiling.
• Claude: The biggest weakness is the competing focal points—the person’s face, phone, Siri logo, and tweet all fight for attention, creating visual clutter that dilutes the main message. Remove or dramatically shrink the tweet overlay and make the Siri logo 2-3x larger with a subtle glow effect, positioning it as the clear centerpiece next to the expressive face. This creates a cleaner visual hierarchy that immediately communicates “shocking Siri revelation” without requiring viewers to read text, increasing the likelihood they’ll stop scrolling to learn what the controversy is about.

The verdict:

GPT takes the win. It delivered the most specific copywriting fix, instantly building intrigue while lowering cognitive load.

Gemini takes a strong second place with a very similar approach.

Grok missed the mark. It claimed the graphics obscured the creator’s face when they actually didn’t.

Claude finishes last. It focused on aesthetics at the expense of the text hook, and its feedback was the least concise.

In short: GPT understood CTR dynamics best, prioritizing the speed at which the brain processes a thumbnail.

E-commerce “Hero Section” audit (UX & trust)

Your Hero section has one job: build instant trust and kill hesitation. It is your digital storefront. If it is cluttered or confusing, visitors bounce without a word.

This was the critical test. Why? Because that first screen is the biggest lever for converting cold traffic—especially on mobile.

We tested if AI could look past pretty pixels to judge actual credibility and offer clarity.

Prompt used: Act as a conversion-focused UX Designer and audit this e-commerce Hero section to identify the primary friction point affecting trust or clarity. Propose one specific layout or copy improvement to immediately reduce buyer hesitation. Explain your reasoning in a maximum of three short sentences.

The results:

• GPT: The biggest friction point is low clarity of the actual offer — the headline announces a “Shopping Event,” but the hero never shows what products qualify or how much the Gift Card is worth. This forces users to scroll before understanding the value. Add a small qualification bar under the headline (e.g., “Up to $200 Gift Card on select Mac, iPad, and AirPods”) or show three product cards with Gift Card values. This instantly anchors the offer, reduces uncertainty, and strengthens motivation to click “Shop.”
• Gemini: The lack of specific value in the offer creates ambiguity; “Get an Apple Gift Card” forces users to guess the potential reward, lowering motivation to click. Proposed Improvement: specific quantitative copy in the subheadline, such as “Get up to a $200 Apple Gift Card with an eligible purchase. Displaying a specific monetary figure creates an immediate value anchor, eliminating the hesitation caused by uncertainty. This clarifies the incentive’s worth upfront, making the call-to-action significantly more compelling than a generic promise.
• Grok: The hero promises an “Apple Gift Card with an eligible purchase” but hides the crucial details (how much gift card? for which products? minimum spend?) behind the footnote and a separate page. This lack of immediate clarity triggers distrust and hesitation during the biggest shopping weekend of the year. Replace the vague line with a bold, high-contrast sub-headline directly under the main headline: “Get up to $250 Apple Gift Card with select Macs, iPads, iPhones & more · Ends Dec 1”
• Claude: The headline “The Apple Store Shopping Event” is vague and doesn’t communicate clear value—users must read the subheadline to understand they’re getting a gift card, creating an unnecessary cognitive load. Change the headline to “Get Up to $200 Apple Gift Card” (with actual amount) and move “with eligible purchase” to smaller text below. This immediately answers “What’s in it for me?” and leverages the proven power of leading with concrete value rather than burying it in secondary copy. Eye-tracking studies show users scan headlines first and often abandon before reading body copy, so front-loading the tangible benefit (free gift card amount) eliminates the “what is this?” moment and accelerates decision-making.

The verdict:

Claude takes the win. It diagnosed the issue at a strategic level: users scan headlines first.

“The Apple Store Shopping Event” wastes the most valuable real estate on the screen because it offers zero immediate value.

Claude moved the concrete promise (“Get a card worth $X”) straight to the headline, drastically shortening the decision time. The other models just focused on tweaking the smaller sub-text below.

Gemini came close with great “price anchoring,” but Claude thought like a Senior Product Designer—questioning the structure, not just the words.

We have the new Creative Director

GPT proved unrivaled in tactical execution and click psychology, sweeping most rounds.

Gemini held a solid second place, while Claude—despite some early technical hiccups—actually demonstrated the best strategic thinking in the finale.

Grok performed the weakest, often missing details or suggesting changes that would actually damage the brand.

In short: GPT dominates “here and now” sales effectiveness, while Claude shines in deep information structure design.