More than half of Canadians have used AI for a shopping-related task, such as researching and comparing products, according to a survey released this year by marketing technology company Omnisend.Illustration by The Globe and Mail. Sources: iStockPhoto/Getty Images
More than two decades ago, as search engines opened up the internet, brands took some of their first, toddling steps to get noticed online.
“You would add all kinds of fake keywords at the end in white letters on white background,” said Matthieu Houle, chief information officer at the Montreal-based Aldo Group shoes and accessories chain. Retailers crammed their web pages with terms invisible to human visitors, but designed to catch the attention of the crawlers that power the likes of Ask Jeeves, Yahoo and Google.
Today, executives such as Mr. Houle are grappling with another major shift, as artificial-intelligence-powered chatbots upend the way consumers find information online. Shoppers looking for a wedding outfit or the best running shoes for wide feet are not asking Jeeves; they’re asking ChatGPT.
More than half of Canadians have used AI for a shopping-related task, such as researching and comparing products, according to a survey released this year by marketing technology company Omnisend. Another 48 per cent said that ChatGPT offered better recommendations than search engines.
“There’s a new customer on the web, and that’s an AI agent,” said Hamish Gunasekara, co-founder of San Francisco-based startup Catalog, which works with retailers to tailor their product information for chatbots – taking websites designed for humans and making them easier to process for AI applications, including information invisible to a human consumer.
Marketers who have spent years becoming experts in SEO, or search engine optimization, now have to become versed in generative engine optimization (GEO), a term that didn’t exist until two years ago. The goal is to get recommended by ChatGPT, Gemini or Claude. In effect, companies are trying to influence the chatbots to influence you, the consumer.
For retailers, the shift brings uncertainty. Companies are revamping their content to boost visibility with chatbots while startups are selling their own services to help brands — even as it’s not precisely clear why chatbots say the things they do. These applications can give different answers each time a question is asked, too, meaning that brands hoping for mentions are playing a game of odds to some extent.
But it’s not something companies can ignore.
“It won’t take very long for most of the traffic on the web to be driven by chat,” said Gil Luria, a veteran technology industry analyst with D.A. Davidson & Co. “...This is going to be very, very transformative to online commerce.”
For retailers looking to be found online, a paradox of this new world is that AI assistants – which are not human – seem to prioritize content that sounds more human. Chatbots are trained to divine the meaning and intent behind questions, and to return relevant answers. That means retailers have to write about their products in ways that match the sorts of questions consumers are likely to ask.
“Instead of saying ‘red dress,’ it’s ‘red dress for a party,’” said Dave Stevens, chief technology officer at Montreal-based fashion retailer Groupe Dynamite Inc. His team has found that many AI agents respond better to information about the customers’ needs than to a hard sales pitch. “It’s more contextual.”
We tried to buy sweaters with AI. Here’s how it went
Mr. Houle’s team at Aldo is moving from including tags such as “brown” and “leather” to providing more information about how to wear shoes with certain outfits and other how-to guides. Aldo has also realized it may need to publish customer reviews, which its website currently lacks. “It’s really thirsty for user-generated content,” Mr. Houle said.
But because AI models rely on information pulled from across the internet, every critical blog post and scathing Reddit thread can help to shape their answers.
So, to influence what chatbots say, companies have to produce a lot of written content – some of it for their own websites, but also sponsored content in other outlets and media coverage from established publications. Some marketers say this is adding more weight to their public relations strategy.
At Canadian Tire Corp. Ltd., for example, communications executives may not have played as prominent a role in marketing meetings in the past, but they are now seen as “a critical element of that team,” said Ilana Santone, the retailer’s senior vice-president of business and AI transformation.
“You have to be much more broad in terms of your thinking on how you build that brand credibility.”
Some of these techniques are grounded in research. In 2024, academics from Princeton University and the Indian Institute of Technology published a paper that coined the term GEO, and they found that AI pulled from content that included citations, quotations and statistics. Further research has shown that FAQs and bullet points can help, too, as does updating content regularly.
A number of startups are now promising to have mastered these techniques to help brands ensure visibility.
New York-based Evertune Inc. is just one firm that has popped up to advise clients on structuring content in an AI-friendly way. Evertune now has more than 100 customers, including advertising agency WPP PLC and Canada Goose Holdings Inc. The firm has built an automated process to run about 150 million prompts each month, across multiple AI models, to gather data on how chatbots surface Evertune’s clients’ brands.
But Evertune’s chief product officer, Ed Chater, acknowledges that some of this is outside retailers’ control, because AI agents are learning from third-party sources “at a faster rate” than on brand websites.
Vancouver-based luggage brand Monos runs its own test prompts, such as “show me the best carry-on” or “I’m moving my daughter into her university dorm, what do I need?” to see how the brand appears in chatbot responses. The results can vary widely depending on which AI platform is queried.
“Between the four or five phrases that we like to test on, we would show up, like, 33 to 50 per cent of the time in the list,” said chief executive officer Victor Tam. “...There’s definitely no winning at all times.”
To see how much chatbot responses can differ, The Globe and Mail ran a small experiment by writing an automated script to ask ChatGPT and Claude for product recommendations across five categories. The script asked each question 10 times to each bot.
How AI is transforming work at Canada’s biggest companies
Claude recommended a wider variety than ChatGPT. When asked about backpacks suitable for work and travel, for example, Claude surfaced 18 brands and 38 retailers. ChatGPT recommended only four brands and a single retailer: Best Buy.
The platforms also favoured different products. Asked for a beginner skin-care routine for a man in his 40s, Claude recommended CeraVe in all 10 of its responses; ChatGPT mentioned it in only two. It also suggested buying products from Best Buy again, which isn’t exactly known for skin care.
Even asking the same question to the same chatbot will yield different responses. ChatGPT recommended Saucony running shoes in four of its answers, but left them out the other six times.
As Monos looks to surface in AI answers, it has seen some early results from changes to its website it made a few months ago. It now publishes more how-to guides, on subjects such as choosing carry-on luggage, advice on “scuffs and scratches,” and a list of carry-on size limits by airline.
“When we put stuff like that out, we can see in AI results, it’s pulling that data,” Mr. Tam said.
Vancouver-based outdoor retailer Mountain Equipment Co. has been doing some similar work, taking blog content that it has produced for years – but that had been buried on its website – and updating it. Those articles are now featured more prominently, part of an effort to emphasize the expertise of MEC’s (human) staff, but also to appeal to AI search. They feature subjects such as how to pack a backpack and tips for safer bear country travel.
“Typically, if we get asked those things on the store, then Claude or ChatGPT is getting asked those too,” said Michele Guimond, MEC’s vice-president of marketing. “So, we need to make sure that the top questions are in there.”
Canadian Tire has been changing how it describes products on its site, both in the content that is customer-facing and in the script that speaks to AI large language models.
“We’re taking a very open approach to allowing AI systems to access and interpret our content,” Ms. Santone said.
Still, for all of the attention on GEO, some people in the field argue the concept is overblown. Thenuka Karunaratne, co-founder of marketing technology startup Daydream Labs Inc., offers SEO and GEO services. But maximizing visibility for AI involves many of the same fundamentals as traditional search engine optimization, he said.
“Anyone that says they can get you to number one on ChatGPT is definitely a liar,” Mr. Karunaratne said. “You can increase the probability, but you cannot guarantee it.”
Amid all this hype, there is also a potential worry for consumers: Research is emerging that shows that chatbots can be more persuasive than traditional search engines.
Academics at Princeton recently devised an experiment to measure chatbots’ influence on purchasing decisions: Participants shopped for e-books, with some typing in an online search and others conversing with chatbots. Some titles were engineered to appear at the top of the search results, while the chatbots were prompted to persuade participants to choose certain books.
Around 61 per cent of participants chose the book promoted by the chatbots; only 22 per cent picked the top result when using internet search engines. Even when a “sponsored” tag was placed next to a book shilled by a chatbot, 55.5 per cent of participants still opted for it.
“I was quite surprised by the magnitude of the effects that we found,” said Francesco Salvi, a Princeton PhD student and lead author of the study.
The findings highlight the need for regulation, according to Mr. Salvi, especially as companies such as OpenAI experiment with advertising inside ChatGPT. AI applications need oversight to ensure they are not tailored to manipulate consumers, he said.
“We don’t know what’s going on inside of them.”