Will Shopping Be Transformed with ChatGPT and Generative AI?

Insight: A visual story of how online shopping can be made simpler.

---

Note: This article was first published on March 13, 2023 and still holds!

---

Imagine that instead of going on an e-commerce site, selecting a category, searching for keywords, trawling through the results, and refining the results based on their filters, you are on a stark white landing page where you can simply ask for what you want. Imagine, for instance, that you are on a grocery site writing:

Visual created by the author.

How about a prompt for a food site?

Visual created by the author.

How about shopping in general?

Visual created by the author.

Fun Fact: The reference site jayshreegururaj.carrd.co is a professional bio site and not a shopping site. It’s used here for illustrative purposes only.

So, what would a response look like? The best approximation with language modeling today is chatGPT, so let’s stick with it. Consider a shopping session that is completed in 5 minutes versus the average of 20–30 minutes it may take you to shop online now.

For the search for a solid black cotton T-shirt in small, I created a visual page with the results that chatGPT listed. A more customized model may query you further on specifics (men/women) and the more specific you can make the query up-front, the faster and more accurate the response is to your desired result.

Visual created by the author. Images were taken from the respective websites as indicated.

The same search on Zara today yields nothing — even after removing the brand name Zara and the size specification. This also shows you the limited power of keyword searches on websites and why the power of generative AI can make frequent online interaction for businesses and consumers that much easier.

Visual created by the author. Zara's site shows a lack of comprehension for even simple requests.

Mindless browsing and fun shopping? If that’s your thing then you should be able to simply ask:

I am bored. Show me fun, trendy, or bestselling bags.

“Sure, here are some fun, trendy, or bestselling bags that you might be interested in:”

I converted the list chatGPT returned to mimic a website page using each brand’s site images but with the text descriptor for each written by chatGPT. What I created is a visual that can resemble a style magazine except for an e-commerce site. Now wouldn’t this make shopping more fun?

Visual created by the author. Images were taken from the respective websites with text by chatGPT.

Here’s a test — how did chatGPT decide on these 5 results? I wanted to know:

Were any of them used by a celebrity

“Yes, some of the bags I mentioned have been spotted on celebrities. Here are a few examples:

1. Fjallraven Kanken Backpack: This backpack has been spotted on celebrities such as Millie Bobby Brown and Emma Watson.

2. Longchamp Le Pliage Tote Bag: Celebrities such as Meghan Markle, Kate Middleton, and Kendall Jenner have been spotted carrying this tote bag.

3. Gucci GG Marmont Shoulder Bag: This shoulder bag has been spotted on celebrities such as Jennifer Lopez, Kim Kardashian, and Beyonce.

4. Michael Kors Jet Set Tote Bag: This tote bag has been spotted on celebrities such as Miranda Kerr, Heidi Klum, and Kate Hudson.

Note: Celebrity sightings may vary depending on location and time.”

Indeed they may!

Imagine if the model understands a longer and more complicated request and provides options.

Dress ensemble in black for work that can also be worn for a happy hour social after work under $200:

“H&M Blazer ($49.99)

Zara Straight Leg Pants ($39.90)

Banana Republic Modal Crew-Neck Tee ($19.99)

DSW Lulu Townsend Jacinta Sandal ($39.99)

Forever 21 Faux Leather Crossbody Bag ($24.99)

Note: Prices may vary depending on the retailer and availability.”

Ok, some of these results are quite generic but a customized language model and program would use a company’s product catalog, images, and modified descriptors. This request in any case didn’t land that well — let’s try again with a clarification.

Find ones that are closer to $200

Certainly, here are some dress ensembles in black for work that can also be worn for a happy hour social after work, that are under $200 but closer to the $200 mark:

Dress: Calvin Klein Scuba Crepe Sheath Dress ($99.00) Blazer: H&M Fitted Jacket ($49.99) Shoes: DSW Lulu Townsend Jacinta Sandal ($34.98)

Total: $183.97

Dress: Zara Belted Wrap Dress ($69.90) Blazer: H&M Fitted Jacket ($49.99) Shoes: DSW Steve Madden Irenee Sandal ($79.99)

Total: $199.88

Dress: ASOS Design Minimal Crepe Shift Dress ($40.00) Blazer: H&M Fitted Jacket ($49.99) Shoes: DSW Lulu Townsend Jacinta Sandal ($34.98)

Total: $124.97

Note: Prices may vary depending on location and availability.

I translated this into a visual below to mimic a website but mixed up the options as I couldn’t find the older Jacinta sandal on the DSW site. In a business context, data inputs would be updated based on stock and in-season styles so this is less of a concern to us.

You can see how a simple query — asking for outfits for specific occasions can translate into a visually appealing shopping experience with greater freedom to add in accessories, virtual avatars, and custom photos (for the brave ones!) You can also see a need to do a mix-and-match of choices — pair the shoes of one with the other options.

Visual created by the author. Images were taken from their respective websites as indicated.

Consider this: Stylists, fashionistas and personal shoppers will be in-demand for online shopping to create, curate and recommend styles which is a regular feature in magazines.

Where can I buy this?

“I provided several options for each of your requests, so it depends on which specific item(s) you are interested in purchasing. You can find the items I listed at various retailers both online and in-store. Here are some places you can check.”

Despite being unsure and without needing another prompt, it provides a short list for all of them.

You won’t need to ask such a question if you are on a Direct-to-Consumer (DTC) site but you may well need to ask for shipping dates or other shopping cart-related inputs. Or perhaps, “is this available in a petite medium?” or “What are petite medium measurements?” Or imagine, you can suggest to it, “What size would a 32” waist for men translate to in pants?” Remember that each of these single commands requires several clicks, browses, and parsing of information presented when you shop now.

Returning to our query on where to buy these items, you get a list of sites (as of 2021 of course!). Perhaps, a site can add to this by sharing links, and price comparisons. Let’s consider one of the query results which shows available sites even though the data was frozen in chatGPT in 2021.

Cross-training shoes from Nike and Adidas under $150:

“Nike Flex Control 4 Training Shoes — Black/White: available on Nike’s website, Foot Locker, and Dick’s Sporting Goods.

Nike Air Max Bella TR 3 Training Shoes — Black/White: available on Nike’s website, Foot Locker, and Dick’s Sporting Goods.

Adidas Edge Gameday Training Shoes — Black/White: available on Adidas’ website, Foot Locker, and Dick’s Sporting Goods.

Adidas Duramo SL Training Shoes — Black/White: available on Adidas’ website, Foot Locker, and Dick’s Sporting Goods.”

Moving on — since the need is to just get a dress, we need to fix our earlier query because the problem was the word “Ensemble.”

Pro-tip: Prompting an AI model to respond appropriately is a skill and will be one of the new skills for product managers to cultivate. New non-technical roles will be created for various other functions.

Though the issue is in our prompt, chatGPT politely suggests it made a mistake!

I apologize for misunderstanding your request earlier. Here are some black dresses under $200 that can be worn for work and a happy hour social:

Lark & Ro Women’s Cap Sleeve Faux Wrap Dress ($49.90)

Calvin Klein Women’s Sleeveless Scuba Crepe Sheath Dress ($99.00)

Everlane The Japanese GoWeave Essential Tank Midi Dress ($98.00)

Anne Klein Women’s Solid Crepe A-Line Dress ($116.10)

Amazon Brand — Daily Ritual Women’s Supersoft Terry Sleeveless V-Neck Dress ($39.00)

Note: Prices may vary depending on location and availability.

But we want a higher-priced dress. So we have to fix the prompt again and this time, remember to ask where to buy it.

“Here are the retailers where you can buy the dresses I suggested:”

What’s this though? It adds:

“I suggested!”

So I am compelled to ask, “Who is I?” ignoring Grammarly’s desperate prompts to change this to “Who am I?” [Aside: Grammarly needs to get with the bot prompts.]

“I” refers to me, ChatGPT, the AI language model you are currently interacting with. I am an artificial intelligence designed to provide answers to your questions and assist with tasks to the best of my ability.”

Ok, but I still think writing “I suggested!” at the end of that sentence is eerie. Something to get used to while noting that ChatGPT is not a person!

Shopping for Gadgets

Now, imagine a search that can be expedited while on a gadget hunt:

Compare and show me the differences between the latest MacBook Pro and MacBook Air.

Though Apple has a nice comparative menu on its site, it needs a) clicks b) user navigation, and c) a user still needs to understand all the technical jargon. Instead, how about getting a concise and categorized summary?

“Sure, here are some key differences between the latest MacBook Pro and MacBook Air:

Visual created by the author. Features and note text is from chatGPT.

Overall, if you need a more powerful machine for demanding tasks, the MacBook Pro is the better option. However, if you need a lightweight and more affordable machine for basic productivity tasks, the MacBook Air is a good choice.”

Ok. Is there a lower-priced item that is similar?

This brings up a more definitive answer than I expected and proceeds to show similarities between the two similar items: the new MacBook Air and the older MacBook Air (2020). I did not ask for such a comparison, but chatGPT assumed that it was a logical question to answer as well.

This shows you the power of language models when combined with a program like chatGPT.

We continue.

Yes, if you are looking for a lower-priced option that is similar to the MacBook Air, Apple offers the MacBook Air (2020) model that is still available for purchase. The main difference between the two is that the MacBook Air (2020) uses an Intel processor instead of the M1 chip found in the latest MacBook Air.

Here are some similarities between the latest MacBook Air and the MacBook Air (2020):

Visual created by the author. Features and note text is from chatGPT.

But what if I want something other than Macs?

“If you are looking for a similar laptop to the MacBook Air that is not made by Apple, here are some options:

Visual created by the author with text from chatGPT.

Note that the contextual conversation continues just like one with a salesperson at the store. To get these results on any site today, you have to spend at least 30 minutes if you are an expert, and more if you are a novice learning the ins and outs of gadget buys. You would have to do all of the searches, understand the results, refine them — multiple times and repeat it on multiple sites.

How about if we switch gears?

Never mind. Show me 4-star and above-rated smartphones.

“Sure, here are some highly rated smartphones with a rating of 4-stars or higher:

Samsung Galaxy S21 Ultra: This is a high-end smartphone with a 6.8-inch AMOLED display, 5G support, a 108MP primary camera, up to 16GB of RAM, and up to 512GB of storage. ($1,199)

iPhone 13 Pro Max: This is Apple’s latest flagship smartphone with a 6.7-inch Super Retina XDR display, 5G support, a triple-camera system with LiDAR scanner, up to 1TB of storage, and up to 6GB of RAM. ($1,099)

Google Pixel 6 Pro: This is Google’s latest high-end smartphone with a 6.7-inch AMOLED display, 5G support, a 50MP primary camera, up to 12GB of RAM, and up to 512GB of storage. ($899)

OnePlus 9 Pro: This is a high-end smartphone with a 6.7-inch AMOLED display, 5G support, a quad-camera system, up to 12GB of RAM, and up to 256GB of storage. ($969)

Note: Prices may vary depending on location and availability.”

This type of interactive shopping online means you don’t have to reset the searches, restart, or endlessly filter by ratings, price, color, size, and multiple categories. You can even imagine a query — “Summarize all the negative ratings,” and “suggest if it is still a good buy” versus trawling through dozens of reviews and using grouped keywords to make sense of them all to buy. A whole host of queries around reviews can provide a more incisive view into audience sentiment.

In addition, most of the catalog entries can also be generated with chatGPT-like interfaces, reducing the need for manual creation (AI tools for copywriting can help too!).

Limitations

Sources used by chatGPT for its results are unclear. Who rated these 4-star products? While these results may or may not be valid anymore, the point of the exercise is to show how e-commerce can be refashioned to model a generative AI workflow. A custom site will resolve these issues. There may be other considerations: processing power, data wrangling, information updates to keep this accurate (true even of current sites), business models, user experience, and a whole host of business considerations including tradeoffs and cost of maintaining the status quo.

Market Snapshot

Statista reports that “in 2021, retail e-commerce sales amounted to approximately 5.2 trillion U.S. dollars worldwide and are forecasted to grow by 56 percent over the next years, reaching about 8.1 trillion dollars by 2026.”

The market size it projects is shared by Oberlo here.

Oberlo image of Ecommerce sales in 2022.

Approximately 2.4B global consumers are shopping online today and according to this study on the state of e-commerce:

“Cross-platform shopping is the hallmark of today’s online purchaser, cementing the importance of integrating retail, search, and social strategies.”

It recommends retailers to:

“Prioritize discovery and convenience.”

A new study from Marketing Dive shows:

“47% of consumers are open to using a chatbot to make a purchase,”

and another study by Juniper Research

“estimates that chatbots will account for $112 billion in retail sales by 2023.”

---

Conclusion

According to McKinsey, “e-commerce has grown 2–5 times faster than before the pandemic.”

The pandemic forced a reckoning for businesses and created the need to build a direct-to-consumer strategy and sales motion across channels. Consumer behavior also shifted with a focus on personalization, multiple platforms, time for discovery, and delivery became critical factors. Specifically:

“63% of consumers on specific sites visit to do research before they buy while 66% choose a retailer based on convenience.

57% purchase on sites based on “shoppability.”

Those grappling with how to make this an effective part of their digital transformation strategy will now need to pause and leapfrog to generative AI just like mobile technology enabled users to leapfrog the need for wireline connections. Mobile technology and access to it were critical factors in the explosive growth of digital connectivity and e-commerce.

Consider this: According to a Deloitte study, “Smartphones are the “universal remote” of our digital lives, with six in 10 respondents controlling their smart home devices and seven in 10 shopping and paying with their phones.”

Leaders will need to rethink what it means to create an interactive experience online.

For example, you can well imagine a site offering a “Classic View,” and an “Interactive View” (just like your email provider does when switching you to a new user interface) to stay safe with the majority of their users in current mode while allowing some users to experience the future (a more disruptive A/B test). This may be expensive, in the short term, but it allows companies to experiment, get user feedback, and revert to the status quo, if necessary, with minimal impact.

The possibilities for re-envisioning the shopping experience are endless significantly reducing the time it takes to make a purchase today. This can add up fast across all online transactions.

You see how this also translates to B2B for example, for SAAS products, don’t you?

This redefined operational mode will also challenge existing business models. For example, if you are a site reliant on ads, this will change how users shop, spend time, seek and click on information. Your business model will need to evolve to a new way of buying.

Is this on the horizon? Yes, in whatever form or shape, whether one, three or five years hence, the world of shopping will be revitalized. Add to this Web3, blockchain, and possibly crypto, then you have a landscape that will be unrecognizable. There are, of course, several dependencies to sort out first — business models, use of big data, analytics, sales, business logic, user workflows, product catalog definitions, website design, and the whole nine yards.

Consider this: In the space of 18 months, and accelerated by the pandemic, digital payments in countries like India have had a significant impact on everyday buying. They have replaced the use of physical currency and ended the need for small change (a perennial friction point) with even vegetable vendors accepting monies of less than $1 via Google Pay or its equivalent.

Voice-based shopping never really took off for various reasons, not least, the un-intuitive way of shopping, and security concerns but also that it needed users to make a behavioral shift from how they had learned to shop. It also eliminated the visual aspect of shopping.

But with a generative AI-based shopping experience, we are looking at a far more intuitive experience, retaining current modes of shopping — online, mobile — with a far better customer experience.

Rather than replacing people, AI will change the nature of work and its related roles, similar to the changes that occurred during other key inflection points in history (think: horses to cars, the industrial revolution, women entering the workforce, PCs, Internet, mobile phones).

Assume that your competitors are already working on using language models to change how users interact with their products. Can you afford to be left behind?

---

Like the article? Feeling generous? Please consider buying me a ☕! Mucho Gracias!