Effective Use of Artificial Intelligence for E-Commerce Companies

Scaling e-commerce copywriting workflows with AI-assisted generation while maintaining strict editorial quality.

Speakers: Burak Pehlivan

18 Oct 2023

Video Transcript

⏱ 25 min read• 4,463 words

Read the full speech or switch languages

Hey everyone. It’s been a while since our last in-person event. This is in Zeo’s DNA. We were just talking about this with the attendees. I’ve really missed being on stage. I’ve missed these physical events. We’re hosting e-commerce companies today. We’ll look at the intersection of e-commerce, content, and AI. I’ll cover what we can do with AI right now. We’ll talk about how AI intersects with e-commerce. We’ll see how to create text and visuals using AI methods. We’ll also look at content optimization using AI tools. We’ll discuss future AI trends in the e-commerce vertical. I think we should split the "what is AI" question into two categories.

Most brands we interact with use narrow AI. Even these tools have launched features that impact us significantly. ChatGPT does text analysis. DALL-E is great for image recognition. Google search has amazing AI features in the narrow AI space. Autonomous vehicles, recommendation engines, voice assistants, and robots are all examples of narrow AI. The second type is strong AI. I hope we don’t see it anytime soon. Most of what we discuss today would change drastically with strong AI. It mimics human intelligence completely. It thinks, learns, acts, and decides just like a human.

Like I said, it’s not here yet. But current machine learning and deep learning will get us there. So, how does this narrow AI intersect with e-commerce? Let’s take a look. Amazon uses AI for personalized shopping recommendations. They’re great at it. Walmart uses AI-driven demand forecasting for supply chain management. H&M uses chatbots and virtual assistants to help users find products. ASOS has integrated visual search features to help users find similar items. Now, let’s focus on content. That’s the core of this presentation. What can we do? Content is still king. Narrow AI feeds on the content we create online.

It uses algorithms to categorize that data. It answers our questions logically. What matters for e-commerce? Finding users, engaging them, and keeping them on the platform. Content is key at every stage. How do we handle content production? We follow traditional strategies. At içerik.com, we produce content using those same traditional strategies. We use AI for content briefs. We use it for keyword research. We use it for outlines and subheadings. I’ll show you how in a few slides. Traditional content creation is slow. It’s also very hard to scale.

E-commerce sites have thousands of products. They need thousands of descriptions. Managing large writing teams is tough. Every writer has a different style. This leads to consistency issues early on. So, how do we do this with AI? The tool I focus on is OpenAI. It’s a no-brainer. GPT-4 has made it incredibly powerful. We use ChatGPT for product descriptions, blogs, and ad copy. We even use it for keyword research. It’s a natural language model. We use OpenAI to scale content for many e-commerce platforms.

Most people use ChatGPT for hobby projects. But how many of you use it at work? Who is using Google Sheet integrations with GPT-4 for content? Beyond hobby projects, we don't see this much yet. But we'll see the next steps soon. As a digital marketing pro, you can use these features easily. It’s a powerful toolset. First, connect Google Sheets to OpenAI via Google Apps Script. We all work on e-commerce sites. Imagine writing descriptions for a thousand products. Don't use the ChatGPT interface for every single item. A Google Sheet speeds up the process significantly. I highly recommend using Google Sheets for e-commerce.

Setting up the script is simple. There are plenty of guides online. I’ll add the source to my presentation. Connect your OpenAI API key. Paste it into the script and save. Now you can run any OpenAI prompt. Let's start with product descriptions. You need a database of core product features. This varies by e-commerce platform and category. I’ve prepared a sample table for you. We pull product names, categories, and brands. You could even ask AI to identify these from the name alone. But these details are already in every e-commerce database.

We can add all the data here. Think color, size, materials, and word count. We can even specify SEO keywords. We can get granular. We can dictate subheadings. We can even tell it exactly where to place keywords. Once the sheet is ready, just write the prompt. It looks technical, but it’s not. You just need to tell OpenAI or ChatGPT one thing. Tell it, "I have a list with product details." Then, just paste the prompt into the description column. Ask it to write a sales-driven product description using that data. It’s really that simple. A lot of this also depends on your own creativity.

You can use prompts like. "Act like an SEO pro and write a description." Once you do that. It works. As you see here, I had my product info and keywords. It generated the descriptions for me instantly. Can I take this and use it on my e-commerce site? Absolutely. It might have minor gaps or miss a few points. To fix that, just add more columns to your sheet. You can always refine your prompt. You need to fine-tune it a bit. But it saves me a massive amount of time. Always review the results before publishing. You can even use AI to automate emails for your users too.

I can ask for an email draft. I'll specify the theme and summer sales. I'll target women in a specific age range. I'll ask for a punchy, sales-driven tone. I'll include my target keywords. I use a similar prompt for this. I define my data points like theme, age, and gender. Then I ask for a compelling, sales-focused email. That’s how I structure my prompt. Language doesn't matter much. But if your database is complex, use English prompts. It often yields better results for complex content. Just a quick tip.

To personalize emails, I add customer names to my table. I include their last purchase and a custom discount. I feed these variables into my prompt. Then I have OpenAI write the email. For example, I ask for an offer based on their last purchase. It also does great work for social media content. Don't forget to use the concatenate command when writing these prompts. Here is why. You might not see it clearly, but my topic is first. Next, I define the social media platform. Then I set the tone. Is it funny, serious, or sales-driven? I define all those details.

I paste my prompt here. It generates social media copy automatically. I use the concentrate command. It pulls data from cell B5. That’s where the platform name is. I ask for a funny Twitter post about project management. That’s exactly what I want. I use the concentrate command for this. I don't just write a sentence. I inject all those variables directly into the prompt. It’s great for keyword research too. It lacks search volume data. You can use Woop for that anytime. EAT will cover that in detail. I just write a simple prompt.

I ask for 10 keywords on a specific topic. I use the concentrate command again. I ask for comma-separated JSON output. It extracts all keyword variations for that topic instantly. It works incredibly fast. I can also fix typos with it. That’s a killer feature. It catches capitalization errors and bad grammar. It helps me clean up my text. Error rates rise with very long texts. It’s perfect for product descriptions or social posts. We can easily fix those common writing mistakes. Here, I’m creating a job posting.

I have a job title and requirements. I wrote a job description. Maybe there are errors. Maybe someone else wrote it. I need to check it. I want to run a command here. I just said this. I'll paste from C5, C2, C3, and C4. Find the errors here. Fix them and paste them back. That's all I have to do. This prompt saves me a ton of time. We see messy data on e-commerce sites all the time. Here is an address example from a user. Sometimes you need to structure this data. I use it for that too. A customer provided this input. The zip code is here, the address is here, and the city is there.

Wait, I think this is the zip code. This is the street. I told it to create a new column. Write only the city there. Then write the isolated state. Put the zip code at the end. These are very simple commands. For example, I said return only the city name from B5. It’s quite simple. It cleans up and parses this messy data for me. This makes things much easier on the development side. Users enter this data. Our developers try to structure it. They parse it. They try to extract the zip code and put it into the right input field.

It’s a tedious, annoying, and long process. I handle it easily with OpenAI sheet integration. Let’s say we wrote product descriptions. Language doesn't matter. We can translate them into any language. It works way better than Google Translate. You’ve probably compared them yourselves. We use translation tools a lot. DeepL is another tool I’d recommend. But GPT-4 does an incredible job here. I think I forgot to add the prompt. But the logic is simple. Here is my product name and English description. I just ask for Spanish, Japanese, or any other language. It generates unique translations for thousands of products in seconds. It has to be unique by design.

Oh, the prompt is right here. I used concentrate. I said translate B5 into Spanish. Once I paste this command, I get the translations in any language I want. This is a huge time-saver. I can also generate summaries. Ozan mentioned this in his presentation. We use this feature a lot. We can simplify concepts or ask for more detail. How would a scientist explain this? Or how do you simplify technical topics for non-techies? OpenAI lets us manipulate these outputs easily. Meta descriptions are key. They’re a vital duo with titles. No debate there. I prepped a dataset. Say I’m doing SEO for adidas.com. Let’s add a "buy one get one" offer for men’s running shoes.

Adidas might not do that, but let’s assume we have that USP. I can easily prompt for an adventurous meta description for this specific model. It looks long, but it’s one prompt. We’ll copy it across our entire database. I used concatenation here. Let’s see what’s in cell E2. It’s for the Adidas Ultraboost Light. It’s for the running shoe category on adidas.com. What else did I include? Let me check. I asked for a meta description including "men’s running shoes." It needs an adventurous tone. Then I told it to output the result.

Like I said, we prep this in 5 minutes. We build the list in 30. Then, in seconds, OpenAI spits out every meta description I need. It’s instant. Honestly, this is a total game changer. We work with tons of e-commerce sites. We used to do titles, meta descriptions, and blog posts manually. It took serious analysis and massive sheets. It was all human labor. That’s why I think these developments are a game changer. Simple prompts and basic sheets are all you need. Seriously, that’s it. Titles are also huge. I added a few extra slides for this.

People are still researching how to optimize title tags. Use questions. Add some emotion. If it’s in English, use an active tone of voice. Make them click-worthy. Add "best of" or "how-to" phrases. Research is ongoing. But for us, it’s just a simple prompt now. Let’s look at an Apple title, for example. Big brands use titles that are straight to the point. What does that mean? It means there’s a specific structure. It goes: product, feature one, feature two, and a fixed keyword. It could be an emotion or an intent. Maybe it says "buy now." Then it adds the brand name. The brand name can go at the start or end. It depends on your A/B tests. It’s up to the SEO pro and the brand’s goals.

We break down all these elements. We create a table based on them. I put the product first. Then the features and the intent. I add the brand name. I tell it to generate a title in this structure. Our developer pulls this data from the database. We use a simple concatenate command. I combine A2, B2, C2, and D2 with spaces. I added a dash in between. That’s the dash right there. Then I add the brand. It gives me: iPhone 14 Pro 256 GB leather purple buy now - Apple TR. That’s it. If we don’t like the result, we just update the prompt. We can change the output in seconds.

This saves weeks of work. We can finish it in minutes. It’s a huge deal. Ad copy is also critical. I’ve found some really interesting results in my own research. It gives very clever answers. For the first example, I ask for display ads. I picked a topic: refurbished MacBooks. I want ad copy for them. I added something just like I mentioned before. I tell it to act like a Google PPC pro. It’s a simple prompt addition. But it seriously boosts the quality of the output.

It gives me the visuals, headlines, and descriptions I need. I can ask for one, three, or five options. It handled everything for these refurbished MacBooks. PPC pros spend so much time on brand analysis. They spend even more time creating this content. I see it all the time. Everything is about performance and click-through rates. They need the best possible titles, descriptions, and visuals. This tool helps me nail those variations. The best part is it suggests the actual visuals. Soon, I’ll feed these prompts into DALL-E. I’ll generate photorealistic ad assets with AI in no time.

I can even run A/B tests on ad copy. I ask which version will perform better. It’s a game changer. Audience analysis is also huge for PPC. What’s the demographic? Where are they located? What are their interests and behaviors? Based on this persona, which keywords should I use? Which ones should I avoid? I think that was on the next slide. Exactly. I won't dive into that now. I can ask these things. If I run a refurbished MacBook promo, what demographic will I see? It answers that. I can then organize my target audience based on the platform.

It’s easier to choose who sees my ads. It also gave great advice on keywords. For "refurbished MacBooks in India," it told me to avoid "free," "used," "second hand," and "cheap.". That’s valuable data. I can put those negative keywords into a sheet. I’ll update my prompt to filter them out. That’s a solid optimization. For ad copy, I can generate detailed versions from our headlines. A simple prompt works wonders. I can even paste messy Google Ads data directly into ChatGPT. You don't even need a sheet. Just paste the raw data. It understands the mess and gives a report. It suggests improvements. For example, it calculated the overall cost of my campaigns.

It gives me many insights in writing. I can even get detailed analysis from ChatGPT. Another great part is what I asked next. I asked for tips to lower my cost per conversion. Look at what it told me. It said to optimize campaign 2. It has the highest cost per conversation. It suggested updating ad copy. It also said to tweak my landing page. It even recommended changes to my target audience. I can ask it to detail these further. That is a pretty cool feature. Getting answers to these questions saves me so much time. That is the gist of it. There is so much we can do with ChatGPT in e-commerce.

I wanted to share the prompts I have been testing lately. Once I share the deck, you will have fun trying them out. You will be surprised by the results. I focused on product descriptions and pages for this concept. But I also use it for blog topics. I use it to outline blog posts. I even add FAQs to the bottom of product pages. I can even run a fact check. It is amazing. I ask if the info in my product description is accurate. It either flags errors or corrects them. If I ask it to rewrite, it does that too. It can even summarize the whole thing.

It can shorten text to any length you want. There is so much we can do. I’ve added the most engaging examples here. You should definitely check out the examples on the OpenAI site. New things are added all the time. Want to build a Q&A base or create content? OpenAI shows you exactly which prompts to use. They really try to guide us. It is definitely worth a look. There are tons of other tools out there. I’m sure you’re already using them. Almost all of them run on OpenAI technology.

They have different interfaces and offer unique benefits. These are all great apps to use in this space. But here is the critical question. What does Google think about all this? Google is in a tough spot. They claim AI has a place in their long-term content vision. But they were late to admit it. I think they underestimated OpenAI’s success and growth. That is why their response felt delayed. Look at news sites in Turkey. Does it really matter if a human wrote that content? I believe AI can create a better user experience.

This isn't about replacing humans. Humans should guide the user experience. Don't be afraid of AI. Use it for your content. Create unique, high-quality user experiences. Review it yourself. Use it for product descriptions, titles, and descriptions. Then there is visual content creation. This is huge for e-commerce. There are great features available. You can identify objects in images. Automatically tag photos. Remove backgrounds. Many e-commerce sites have terrible user experiences. Look at their listing pages. The backgrounds are a mess. The lighting is bad. The photography is poor. There are great tools to fix these issues. They resize images to your exact needs.

They sharpen the focus. They fit images perfectly into your grid. They don't lose quality. These are the pain points of the e-commerce industry. I love a project called Let's Enhance. It resizes content perfectly. It improves visual quality. It doesn't bloat your database size. It makes upscaling and enhancing images so easy. It's a great tool. I wanted to focus on these types of tools. The best part is they have APIs. Just like using ChatGPT for bulk product descriptions, you can automate image processing. No more manual background removal.

I have my dev pull product images. We send them to the Clay AI API. It’s actually their Lens Enhance product. We strip the backgrounds and resize them. The result is a clean, elegant, minimalist listing page. The user experience is top-tier. It’s a killer feature. It really boosts quality. The best part? The file size stays the same. No bloat, no quality loss. It’s seamless. I can wipe out those backgrounds exactly how I want. Here’s another win. I also color-correct everything. It makes the images look sharper. It really improves the UX. And again, the file size doesn't budge.

The custom resizing is a game changer. Most e-commerce sites force users to do this. That’s just not realistic. Others try to build their own tech to handle it. Sometimes it works, sometimes it fails. Either way, it’s a huge dev burden. It eats up so much time. Now, I just optimize everything to my standards automatically. How does AI fit in? It’s the secret sauce. Say I need a square image but only have a rectangle. AI fills in the missing pixels perfectly. It doesn't stretch the image. It just creates a perfect square.

AI keeps every element intact in these apps. It’s a huge plus. You have a product shot with bad lighting. We fix the light. We clean the background. Then, we drop in entirely new scenes. It’s brilliant. It boosts your e-commerce UX. It makes product photos look unique. I can match scenes to our brand colors. Or I can match the product’s brand identity. I place the objects right in front of those scenes. It’s a great feature. Like I said, take a product photo. Create different scenes. Make the images look original. Make them look better. There’s another app for this called Remove Background.

It has an API. I never recommend apps without an API. They don't make life easier. They don't change anything. That’s why I wanted to mention them. It’s very good at removing backgrounds. It claims striking quality. And it really delivers. Most background removers leave messy spots. They aren't always 100% clear. But Remove BG is very successful here. The API is super fast. We can upload all our images instantly. We can shape them however we want. Some e-commerce platforms have built-in integrations. It makes total sense for them to invest in R&D.

Let’s talk about content optimization. What AI tools can optimize our text and images? I want to cover those now. These tools check your content. Are the right keywords used? What’s the word count? How many images do top 20 competitors use? They provide data on keyword frequency and more. We can easily use these for optimization. Most already integrate various AI technologies. First is Market Muse. It creates great content briefs. I showed you how to do this in ChatGPT. But third-party apps work great too.

I pick a topic for a blog post or FAQ page. I ask what questions to answer. It gives me a comprehensive list of relevant questions. This makes a perfect brief for a writer. These are real questions users are searching for. They are based on actual search engine data. Like I said, they provide detailed feedback on keyword usage. I can ask, "Here's my content. Which keywords should I use and how often?" Is that a key metric? That's debatable. But seeing the keywords I need is great. Should I rely on math for the counts? Also debatable. A good writer with a solid brief and outline will include them naturally anyway.

Still, it's great for competitor analysis. If I used a keyword twice and a competitor used it seven times while ranking higher. It makes sense to increase my usage. I use Dynamic Yield for A/B testing. There are many platforms, but why Dynamic Yield? Their AI algorithms are perfectly structured. Let's say I'm Hepsiburada. My category and product pages are identical. It analyzes user behavior per category or per user. It constantly updates page design and copy based on those insights to improve performance. The best part is it does this dynamically. It’s an impressive product. I’m sure large-scale e-commerce sites use it, even if they don't talk about it.

What else can I do with these apps? I can show personalized product recommendations. I look at user browsing data. I check their history. I show products tailored just for them. I also do dynamic content personalization. Based on user behavior, I create custom ad banners. I tailor the design and content for them. That’s where AI really kicks in. It can also handle segmentation and audience building. Is it fully automated or dynamic? I’m not 100% sure. But my research suggests it works dynamically. They say you can do some cool things there.

Take a user who abandons their cart. You want to nudge them to finish the purchase. You can automatically offer them a custom perk. Like free shipping or a 10% discount. These are great features. Amazon AWS has a personalization product. Developers are usually all over this stuff. You integrate it into your e-commerce site. It uses browsing history to auto-adjust product images and content. It personalizes, tags, and groups items. It’s a solid solution. Amazon is great for large-scale e-commerce. Adobe is also doing serious work here. Even as a visual-focused brand, they have tools for content categorization and auto-tagging.

Some e-commerce sites use video content abroad. They can remove objects from videos. It looks like they were never there. The video flow remains seamless. Brands are developing these AI features. There are many tools in this space. I’ve listed the main ones here. Cognitive SEO gives tips on keyword density. It tracks word counts and distribution. Outranking is similar. It analyzes competitor titles. It suggests headings and sections. It provides great data. All of them offer keyword suggestions too. Surfer SEO builds great outlines. Their outline builder is very successful. It focuses on search intent. I create them directly. Then I share them with my writers.

Clearscope works much like Cognitive SEO. I’ve included screenshots here. Frase has a solid outline module. It’s worth trying. There are many AI tools for content optimization. We’re nearing the end. Let’s look at the near future. Voice commerce is already here. We have Siri and Alexa. We also have Hey Google products. I use these for product searches and adding items to carts. Sometimes even for completing purchases on certain platforms. Soon, these will be more dialogue-driven. They will solve complex user queries. Maybe they will integrate with IoT devices. Imagine walking by a store and saying, "I love this. Buy it and ship it home." We are heading toward that kind of assistant.

They can offer hyper-personalized shopping experiences. Advanced personalization is already a hot topic. In the near term, we will work with much larger datasets. More complex variables will come into play. We might see fully personalized storefronts on e-commerce sites. Not just on specific pages, but everywhere. This is one of the expected possibilities for the future. Chatbots and FAQs already exist for customer support. But the real goal is accuracy and context. We need chatbots that can communicate proactively with users. This means the system detects my problem before I even ask. It reaches out or suggests products based on my needs. It analyzes my browsing history and future data patterns. We can integrate these advanced support systems into our platforms.

That is all from me. I focused on the intersection of e-commerce, content, and AI. There are great AI tools for e-commerce beyond just content. Futurepedia is definitely a site where you should spend some time. Yiğit found some great tools to share with you. He’ll walk you through them. That’s it for me. Thanks for listening.

This transcript was generated using AI speech recognition and verified against subtitle style rules.

Burak Pehlivan

Content author

Burak Pehlivan

both

Contact us

How we handle your data

Open as a full page