AI in eCommerce Marketing: What's Real, What's Hype, and What We Actually Use

I run a marketing agency. We use AI tools every single day. And I'm going to tell you something that might sound strange coming from someone who's bullish on technology:

Most of the AI hype in eCommerce marketing is exactly that — hype.

Not all of it. Some AI tools have genuinely changed how we work. They've made us faster, more efficient, and able to deliver better results for our clients. But the gap between what AI vendors promise and what AI actually delivers in day-to-day eCommerce marketing is massive.

I'm writing this because I'm tired of the noise. Every SaaS tool has slapped "AI-powered" on their landing page. Every agency is claiming they use "proprietary AI" to 10x results. Every LinkedIn post is breathlessly announcing that AI will replace marketers by next Tuesday.

None of that is helpful. What's helpful is an honest breakdown of which AI tools actually move the needle in eCommerce marketing, which ones are underwhelming, and how to think about AI investment as a brand.

Here's what we've learned after testing dozens of AI tools across 150+ eCommerce clients.

What's Actually Working: AI Tools We Use Daily

1. Klaviyo's AI Features

Verdict: Genuinely useful. Not magic, but genuinely useful.

Klaviyo has been steadily adding AI capabilities, and several of them have become core to how we manage email/SMS for our clients.

What works well:

Predictive Analytics — Klaviyo's predicted customer lifetime value (CLV), predicted next order date, and churn risk scores are legitimately good. They're based on your store's actual purchase data, not generic models.

How we use them:

• Segment high-CLV customers for VIP treatment (better offers, early access, priority support)
• Trigger replenishment emails based on predicted next order date (instead of arbitrary timing)
• Identify at-risk customers based on churn risk and trigger win-back flows before they're gone

The predictive models aren't perfect — they work best with stores that have 6+ months of data and a decent repeat purchase rate. But they're accurate enough to meaningfully improve targeting.

Subject Line AI — Klaviyo's subject line assistant suggests alternatives based on your campaign content. It's not going to write a Pulitzer-winning subject line, but it's a solid brainstorming partner. We use it as a starting point, then human-edit.

Send Time Optimization — Klaviyo's Smart Send Time analyzes each individual subscriber's open behavior and sends the email when they're most likely to open it. For campaigns, this consistently lifts open rates by 5-10%.

What doesn't work well (yet):

AI-generated email copy — Klaviyo's AI can draft email body copy. It's... fine. It reads like AI-generated marketing copy — technically correct, emotionally flat, and indistinguishable from every other brand's AI-generated copy. We always rewrite it. The AI draft saves maybe 10 minutes vs. writing from scratch. Not transformative.

AI product recommendations — The recommendations engine is decent but not dramatically better than rule-based "frequently bought together" logic for most catalogs. It shines with large catalogs (1,000+ SKUs) where manual merchandising isn't practical.

2. Ad Creative Generation

Verdict: The biggest time-saver in our workflow.

This is where AI has made the most tangible difference in our day-to-day work. Generating ad creative variations — images, copy, and video — used to be the biggest bottleneck in our paid media process.

What we use:

For ad copy: We use AI writing tools to generate 10-20 ad copy variations in minutes. Not final copy — first drafts. A copywriter reviews, edits, and picks the best 5-6. The AI is good at generating volume. The human is good at picking winners and adding brand voice.

Before AI: 3-4 hours to write 10 ad variations. Now: 45 minutes (including review and editing).

For image creation: Tools that generate and edit product images — background removal, lifestyle context placement, color variations — have gotten remarkably good. We can take a flat product photo and generate 20 lifestyle-context variations in 30 minutes.

For video ads: AI video tools can generate short-form video ads from product images and copy. The quality is... okay. Good enough for testing concepts. Not good enough for hero creative. We use them for rapid testing — run 10 AI-generated video concepts, see which angle resonates, then produce the winner professionally.

The real impact: We can test 3-5x more creative variations than we could two years ago, at the same budget. More creative testing = faster learning = better performance. That's the genuine AI advantage in paid media.

3. Search and SEO Tools

Verdict: Useful for research. Unreliable for content creation.

AI has genuinely improved the research phase of SEO work:

• Keyword clustering: Tools that group hundreds of keywords into topical clusters in seconds (instead of hours of spreadsheet work). This is a real time-saver.
• Competitor analysis: AI-powered tools that analyze competitor content, identify gaps, and suggest content opportunities. Saves significant research time.
• Technical SEO auditing: AI that identifies technical SEO issues and prioritizes them by impact. Good for large sites with thousands of pages.

What doesn't work for SEO:

AI-generated blog content. I know this is controversial. Plenty of people are publishing thousands of AI-generated blog posts and claiming great results. And yeah — some of them rank. For now.

But here's what we've seen across our clients: AI-generated content that ranks tends to rank for low-competition, low-value keywords. For competitive, high-value keywords — the ones that actually drive revenue — human-written, expert-level content consistently outperforms AI-generated content.

Google's helpful content updates have gotten increasingly good at identifying and deprioritizing content that reads like it was written to rank rather than to help. AI content, even when well-prompted, tends to be comprehensive but not insightful. It covers the topic but doesn't add original perspective, real experience, or genuine expertise.

Our approach: Use AI for research, outlines, and first drafts. Have a human expert write the actual content. The human adds the "I've done this for 150+ clients and here's what I've learned" perspective that AI can't replicate — because it hasn't actually done it.

4. Customer Service Chatbots

Verdict: Good for pre-purchase questions. Terrible for post-purchase problems.

AI chatbots on eCommerce stores have improved significantly. The best ones (Gorgias AI, Tidio AI, Siena AI) can handle common pre-purchase questions with surprising accuracy:

• "What size should I order?" (pulls from size guide data)
• "How long does shipping take to Dubai?" (pulls from shipping policy)
• "Is this product vegan?" (pulls from product attributes)
• "What's the difference between Product A and Product B?" (pulls from product descriptions)

For these use cases, AI chatbots handle 40-60% of inquiries without human intervention. That's real cost savings.

Where they fail: Post-purchase issues. "Where's my order?" "I received the wrong item." "I want a refund but your portal isn't working." These situations require empathy, nuance, and creative problem-solving that AI chatbots consistently get wrong.

A customer who's frustrated about a delayed order doesn't want to talk to a bot that says "Your order is on its way! Is there anything else I can help with?" They want a human who says "I'm sorry about the delay. Let me look into exactly where your order is and what we can do to make this right."

Our recommendation: Use AI chatbots for pre-purchase Q&A (it's genuinely good). Route post-purchase issues to human agents immediately (don't make frustrated customers fight a bot first).

5. Product Recommendations

Verdict: Marginally better than rule-based systems for large catalogs.

AI-powered product recommendation engines (Rebuy, Nosto, Dynamic Yield) promise to show each customer personalized product suggestions based on their browsing and purchase behavior.

Do they work better than simple "frequently bought together" or "customers also viewed" rules? Yes — marginally. We've tested AI recommendations against rule-based systems across several clients:

• AI recommendations: 4.2% click-through rate, 2.1% conversion rate
• Rule-based recommendations: 3.5% click-through rate, 1.7% conversion rate

That's a 20% improvement. Meaningful? Yes. Revolutionary? No.

Where AI recommendations shine: Large catalogs with 1,000+ SKUs where manual merchandising is impossible. The AI can surface products from the long tail that a human merchandiser would never think to recommend.

Where they don't matter much: Small catalogs with under 200 SKUs. When you only have 200 products, a merchandiser who knows the catalog can set up rule-based recommendations that perform just as well as AI.

What's Overhyped: AI Claims That Don't Hold Up

"AI-Powered Audience Targeting"

Multiple ad platforms and agencies claim they use AI for "superior audience targeting." In most cases, this means they're using the same machine learning that Meta and Google already build into their ad platforms.

Meta's Advantage+ targeting is AI. Google's Performance Max is AI. The platforms themselves are already doing this. A third-party tool claiming "AI-powered targeting" is usually just adding a layer on top of what the platform already does — sometimes making it better, often making it worse by adding noise.

"AI That Writes Like Your Brand"

No AI writes like your brand. Not yet. AI can mimic a tone based on examples, but it can't replicate the nuance, the specific references, the inside jokes, or the lived experience that makes brand voice authentic.

Every time we've tested AI-generated emails against human-written emails (same offer, same audience, same subject line), the human-written version outperforms on click-through rate by 10-25%. Open rates are similar (the subject line does most of the work there), but clicks and conversions are consistently better when a human writes the body.

"Autonomous Marketing Agents"

This is the buzziest claim right now — AI agents that can run your entire marketing operation autonomously. Set your goals, let the agent handle everything from ad creation to email copywriting to budget allocation.

The technology is moving fast, and some early implementations show promise. But in 2026, fully autonomous marketing agents are not ready for prime time. They lack:

• Brand judgment. An AI agent might create an ad that technically performs well but is tone-deaf, off-brand, or insensitive to cultural context.
• Strategic thinking. AI can optimize within a strategy. It can't create a strategy. It can't decide that you should pivot from acquisition to retention because your CAC is unsustainable.
• Contextual awareness. An AI agent doesn't know that your biggest client just posted a negative review, or that a competitor just launched a similar product, or that a cultural moment creates an opportunity for timely content.

The brands I know that have tried fully autonomous marketing agents have all pulled back to a hybrid model: AI handles execution (generating variations, optimizing bids, scheduling sends), humans handle strategy and judgment.

"AI Personalization at Scale"

The promise: every customer sees a completely personalized version of your website, your emails, and your ads — generated in real-time by AI.

The reality: most "AI personalization" is just segmentation with more segments. Instead of 5 audience segments, you have 50. That's better, sure. But it's not the 1-to-1 personalization that the sales pitch implies.

True 1-to-1 personalization — where every individual sees truly unique content — is technically possible but practically unnecessary for most eCommerce brands. The incremental lift from going from 50 segments to 50,000 individual variations is marginal. The cost and complexity are enormous.

Our advice: Focus on getting your segmentation right (5-10 meaningful segments based on behavior and value). That gets you 80% of the personalization benefit at 10% of the cost.

How to Think About AI Investment

Here's the framework we use when evaluating AI tools for our clients:

The 10x Test

Does this AI tool make us 10x faster at something, or does it make us 10% better?

• 10x faster: Worth adopting immediately. Ad creative generation is a 10x speed improvement. Keyword clustering is a 10x speed improvement. These are no-brainers.
• 10% better: Worth testing, but don't restructure your workflow around it. AI product recommendations are 10-20% better than rule-based. Nice, but not transformative.
• Marginally different: Not worth the switching cost. If the AI tool is 5% better than your current solution, the time spent migrating, learning, and debugging probably eats up the benefit.

The Human Ceiling Test

Is there a ceiling where human judgment is required?

For ad creative generation, the ceiling is low — AI can generate the variations, but a human needs to pick the winners and ensure brand fit. The AI handles 80% of the work. Worth it.

For email copywriting, the ceiling is high — AI can draft, but a human needs to rewrite most of it. The AI handles 30% of the work. Marginal.

For strategy and planning, the ceiling is very high — AI can provide data and suggestions, but a human makes the decisions. The AI handles 10% of the work. Not worth paying premium prices for "AI-powered strategy."

The Dependency Test

If this AI tool disappears tomorrow, are you screwed?

Don't build critical workflows around a single AI tool that you don't control. Use AI as an accelerant, not a foundation. Your strategy, your customer relationships, your brand voice — those should be human-owned. AI tools should make the humans faster and more effective, not replace them.

What We Actually Use: Our AI Stack

Here's the honest breakdown of AI tools in our daily workflow:

Email/SMS (Klaviyo):

• Predictive analytics: Yes, daily use for segmentation
• Send time optimization: Yes, for every campaign
• Subject line assistant: Occasionally, as brainstorming aid
• AI copy generation: Rarely, only for first-draft inspiration

Paid Media:

• AI creative generation tools: Yes, daily use for ad variations
• Platform AI (Meta Advantage+, Google PMax): Yes, but with manual oversight and guardrails
• Third-party AI targeting: No, we trust the platforms' native AI

SEO:

• AI keyword research/clustering: Yes, significant time saver
• AI content generation: No, for final content. Yes, for outlines and research summaries.
• AI technical auditing: Yes, for large sites

Customer Service:

• AI chatbot (pre-purchase): Yes, for clients who want it
• AI chatbot (post-purchase): No, always route to humans

Analytics/Reporting:

• AI anomaly detection: Yes, useful for flagging performance changes across many accounts
• AI attribution: Testing, not fully adopted

What we don't use:

• Autonomous marketing agents
• AI "personalization at scale" platforms
• AI-generated blog content for client sites
• Any tool that claims to replace strategic thinking

The Honest Future of AI in eCommerce Marketing

I think AI will continue to get better at the things it's already good at: speed, variation generation, pattern recognition, and data processing. The tools that help us test more creative, analyze more data, and automate repetitive tasks will keep improving.

I don't think AI will replace marketers who think strategically, understand brand, and build genuine customer relationships. Not because AI can't theoretically do these things, but because the brands that win are the ones with a point of view, a personality, and a human touch — and those things are really hard to automate.

The best eCommerce marketing teams in 2026 and beyond will be small, sharp teams that use AI to do the work of teams twice their size. Not by replacing humans with AI — but by giving each human AI-powered tools that make them dramatically more productive.

That's how we think about it. That's how we use it. And that's what actually drives results.

---

Want to know which AI tools would actually help your eCommerce marketing? We'll audit your current stack, identify where AI can save time or improve results, and skip the hype. Honest recommendations only. Book a free strategy call.

---

Mark Cijo is the founder of GOSH Digital, a full-service digital marketing agency that's helped 150+ eCommerce brands generate over $23M in tracked revenue. He uses AI tools every day — and he's honest about which ones actually work and which ones are just good marketing by the vendors selling them.