Five AI Assistants, Five Different Hot Sauces

Every assistant is its own shelf

We typed the same question into five AI assistants on the same day: "What is the best small-batch hot sauce I can buy online?" We got five different answers at the top.

Think of each assistant as a separate store shelf, stocked by a different buyer. A shopper who asks ChatGPT walks away with Karma Sauce from Rochester, New York. The same shopper asking Gemini hears about Barbaro Mojo, a Cuban-style sauce from Miami. Neither shopper sees the other's shelf.

That matters if you sell something. Being the first name in one assistant tells you almost nothing about the other four. And the order can move on a different day, which is why AI answers change from one check to the next even when nothing about your store has changed.

The assistants did not always disagree. On hiking socks, all four that answered put Darn Tough first. On beef jerky, Country Archer led in four answers. Disagreement showed up most in crowded categories with no clear leader, like hot sauce, dog treats and kosher snacks.

Big stores win "which store." Small brands win "which product."

The clearest pattern in the 20 questions was the shape of the question itself. When a shopper asked where to buy, the big marketplaces often won. When a shopper asked what to buy, small and independent brands took the answer.

Even the "which store" answers were not a clean sweep for the giants. For board games, Perplexity and Google AI Mode both sent shoppers to GameNerdz, a specialist shop. For a chef knife, Google AI Mode led with Chef Knives To Go and Perplexity with Knivesandtools. Amazon came first only in Claude's board game and chef knife answers, and Claude gave both of those without searching the web.

This lines up with what we saw in a separate test on local businesses, where AI named local shops over national chains for coffee, pizza and repair work. Online, the same thing happens one product at a time.

The bearded dragon split

One question showed how differently two assistants can read the same market. We asked where to buy live feeder insects online for a bearded dragon.

Claude, Gemini and Perplexity all listed Chewy first, the big pet retailer. Google AI Mode went the other way. It sent bearded dragon owners to specialist breeders first: Dubia.com, Rainbow Mealworms and Fluker Farms. Chewy was not its lead.

Same question, same day, and one assistant handed the sale to a national retailer while another handed it to the people who actually raise the insects. If you sell in a niche like this, you cannot assume the answer you saw in one assistant is the answer your customers see. Here is how to check Google AI Mode for your own business, step by step.

Where the answers came from

Most of these assistants show their sources. Reading them was the most useful part of the study. The answers leaned on third-party writing: taste tests, gear reviews, buying guides and niche blogs. Brands describing their own products came up far less.

• Beef jerky: Gemini and ChatGPT both drew on a taste test from The Kitchn. Google AI Mode's top pick cited The Kitchn and Serious Eats.
• Olive oil: Claude's pick, Graza, came with a line about a New York Times ranking. Perplexity's pick, La Tourangelle, came from Food Network's list.
• Hiking socks: ChatGPT credited GearJunkie's top overall pick. Perplexity's sources included Wirecutter and Treeline Review.
• Niche blogs: jerky specialists like JerkyGent, a biltong guide on btbiltong.com, and an independent hot sauce roundup on madelines.co all fed answers.

One exception is worth knowing about. Gemini's hot sauce sources included a "best small-batch hot sauce" roundup published on Barbaro Mojo's own blog, and Barbaro Mojo was Gemini's #1. That page was a guide answering the exact question, not a product page. It is a small, single example, but it points the same way as everything else: the assistants look for pages that answer the shopper's question.

We have measured this at a larger scale before. In a study of more than 2,000 citations, independent blogs made up most of what AI engines cited for buying questions. And reviews feed AI recommendations in ways most store owners never see.

What this means for a store owner

That is good news for small brands, because it is a game you can play without a national ad budget. In practice it means four things:

• Get reviewed for the exact product. A mention in a general gift guide is weaker than a real test of your hot sauce, your candle or your socks by someone who covers that category.
• Get onto the best-of lists the assistants read. Look at the sources each assistant shows for your category. Those are the pages to be on.
• Answer the shopper's question on your own site. A clear guide that answers "what is the best X for Y" gives an assistant something to quote.
• Check every assistant, not just one. Each one is its own shelf. Run the same buying questions in all five. This question list for an online store is a good place to start.

The thinking behind all of this is simple, and we wrote it up in full: AI recommends the business other people vouch for.

Every question, every first pick

Here is the first product or store each assistant named, for all 20 questions. The hot sauce row is highlighted.

"No pick named" means the assistant answered with general advice and did not name a specific product or store. "No answer" means we did not collect an answer for that pair. On a phone, assistants with no answer are left out of each card.

How we ran it

We wrote 20 questions the way a shopper would type them, covering food, home, pets, gifts, apparel and personal care. Each question was asked once, on October 8, 2026, with no follow-up questions.

Google AI Mode was asked on its public website, the same way a shopper would use it. ChatGPT, Claude, Gemini and Perplexity were asked through each company's official developer access, with web search turned on.

Not every assistant answered every question. We report only the answers we got. A blank cell in the table means we have no answer for that pair, not that the assistant refused.

For each answer we recorded the first product, brand or store named as a recommendation, and read the sources the assistant showed. Answers collected per assistant: ChatGPT 8, Google AI Mode 11, Claude 20, Gemini 16, Perplexity 20, out of 20.

Some limits to keep in mind. This is one day and one asking per question, so it is a snapshot. Answers through developer access can differ from what the same assistant says in its consumer app. On 7 of the 20 questions, Claude answered without searching the web, and those answers leaned on general advice and familiar names. We list those picks as they came, and say so where it matters.

RunOctopus helps independent brands get named in AI answers like these.