Ask an AI assistant which electric scooter to buy and you’ll get a calm, confident answer in about four seconds. No ads, no ten blue links, no sense that anyone had a stake in the outcome. But plenty of brands did, and many of them spent months shaping the answer before you ever asked the question.
The shift: from ranking to being the answer
For two decades, brands competed for position on a results page. Even a mediocre ranking earned some clicks, and the shopper did the final comparison.
That model is eroding. A growing share of shoppers now ask an AI tool and accept what it says. The unit of competition has changed from the ranked link to the recommended name. You’re either in the paragraph or you don’t exist.
This created a new discipline: generative engine optimization, or GEO. Where SEO tried to rank in a list, GEO tries to shape product content so AI tools mention and favor a brand in their written answers. The common tactics are adding statistics, citing experts, and including quotations. The researchers who introduced the term reported that tactics like these could raise a source’s visibility in generated answers by a sizable margin, in some tests as much as 40%.
Why it works
Language models don’t “know” a product is good. They assemble answers from text that looks credible and is easy to extract. A page packed with specific numbers, named experts, and clean quotable sentences is easier to lift into an answer than a vague brand pitch. That has three consequences worth understanding:
- Format can beat substance. A product with mediocre performance but beautifully structured evidence can outperform a better product with sloppy marketing.
- Small brands can compete, and so can bad actors. GEO lowers the barrier for newcomers, but it also lets anyone fabricate authority with an invented statistic or a decorative “expert.”
- The shopper can’t see the contest. In search, you could at least see the competing pages. In an AI answer, the losers vanish silently.
The real risk isn’t bias. It’s invisibility.
Bias in recommendations isn’t new. Magazines had advertisers and retailers had endcaps. What’s new is the packaging: a single fluent voice with no visible sponsorship, no alternatives on screen, and no obvious reason to doubt it.
What shoppers can do today
You don’t need to avoid AI tools. You need a short checklist:
- Ask what was rejected. Prompt: “Which two alternatives did you rule out, and why?” A weak recommendation tends to fall apart here.
- Demand the source, then open it. If the “independent testing” traces back to the brand’s own blog, discount it.
- Check for a trail of numbers. A statistic without a named study, date, or method is decoration.
- Cross-check with one source that has no sales interest, such as long-form reviews or owner forums.
- Rerun the question differently. Ask the same thing with a different budget or priority. If the same brand wins every version, that’s worth a closer look.
What brands should do (without crossing the line)
Not all GEO is manipulation. The defensible version is simply publishing better information:
- Share real benchmarks with the method stated.
- Quote actual customers and experts who agreed to be named.
- Keep specs, pricing, and comparisons clear, current, and easy to parse.
The takeaway
The shift from search to AI answers didn’t just change where we look. It changed who gets a say before we decide. Brands are adapting faster than shoppers are, and that gap is the opportunity and the danger.
So the next time an assistant gives you a perfect answer, treat it as the start of your research, not the end of it. The smoothest recommendation isn’t always the best one. Sometimes it’s just the best prepared.