Coke or Pepsi? Two AIs pick. Two won't.

We asked four AI assistants simple either/or questions about five famous brand rivalries, worded the way people really ask. Then we sorted every answer: did it pick a side, or did it dodge?

Answers: 240
AI assistants: 4
Rivalries: 5
Asked: 21 Oct 2026

What we asked

Five rivalries, four ways of asking

Each rivalry got four everyday questions. Some were plain ("Coke or Pepsi, which one should I buy?"), some gave a situation ("I'm stocking the fridge for a party. Pepsi or Coke?"). We switched which brand came first half the time, and never told the AI it had to choose. Every question went to ChatGPT, Perplexity, Gemini and Claude three times.

Coke vs Pepsi Ford F-150 vs Chevy Silverado McDonald's vs Wendy's Starbucks vs Dunkin' Costco vs Sam's Club

How we sorted the answers

We decided the rule first, then applied it to every answer

Picked

It names a winner. Or it says "it depends", but still tells you what to choose if you know nothing else.

Didn't pick

"It depends", with each brand winning a different situation and no overall choice.

Didn't pick, either

Reporting what critics or taste tests say while declining to choose. If an AI goes out of its way not to pick, we don't pick for it.

A second person checked a random sample of the sorting. One mistake turned up; we fixed it and every answer like it.

What we found

Three things the answers show

Whether you get an answer depends on which AI you ask.

ChatGPT and Perplexity usually pick a side. Gemini and Claude almost never do. The brands didn't change this much. The AI did.

Answers that picked a side, out of 60 per AI

ChatGPT49 of 60

Perplexity44 of 60

Gemini3 of 60

Claude1 of 60

Same question, four AIs: "Coke or Pepsi, which one should I buy?"

ChatGPT · picked

"I'd go with Coke! It's crisp, refreshing, and has a classic taste. … My pick: Coca-Cola."

Perplexity · picked

"Default pick: Coke, especially with burgers, pizza, or ice."

Gemini · didn't pick

"Buy Pepsi if: You want a sweet, citrus-forward soda… Buy Coke if: You want a crisp, fizzy soda…"

Claude · didn't pick

"Honestly, this comes down to personal taste rather than one being objectively 'better'."

Short extracts from real answers, emoji removed. Each answer went on to explain its reasoning.

When an AI picks, it picks the same side almost every time.

There's very little back and forth. Across every AI and every wording, Pepsi and McDonald's were never picked. The real question isn't which brand wins. It's whether the AI picks at all. Costco vs Sam's Club was the one most AIs wouldn't call.

48 answers per rivalry Picked the brand on the left Picked the brand on the right Didn't pick

Coke 250 Pepsi

Wendy's 240 McDonald's

Ford F-150 231 Chevy Silverado

Dunkin' 124 Starbucks

Costco 62 Sam's Club

How you ask can flip the answer.

Coffee was the one rivalry where the winner changed. Ask ChatGPT "which is better, Dunkin' or Starbucks?" and it picked Starbucks. Tell it you need a coffee on the way to work, or for your morning coffee, and it picked Dunkin' every time. Perplexity did the same with the everyday questions, and wouldn't pick at all when asked which is better.

4 of 6

ChatGPT picked Starbucks on "which is better?" (never Dunkin')

12 of 12

ChatGPT and Perplexity picked Dunkin' on the everyday questions

Why it matters

For famous brands it's a curiosity. For yours, it decides whether you're named.

The same reluctance shows up in our commercial studies. When we asked "who is the best dentist in" a city, Claude said there was no single best in 91–94% of answers. When we asked for "top dentists", it said that in 0–2% and gave a list. Which AI your customers use, and how they word the question, changes whether your name comes up at all.

See the dentist study →

Which AIs pick you over your rival?

Your own study measures it properly: the questions your customers really ask, asked many times, across ChatGPT, Gemini, Claude, Perplexity and Google.

Get your own study

A note on this find

This was a quick experiment: 240 answers collected on 21 October 2026 from the versions of ChatGPT, Perplexity, Gemini and Claude being served that day. It shows how the AIs behaved on these questions at that time, not a lasting verdict on any brand, and it's small next to our studies. Claude's answers were collected with an earlier setting that cut off some longer answers. A later re-run with full-length answers showed the same reluctance to pick. No brand was told about this test or paid to be in it.

How we measure →