What does AI tell the crowd to buy?
Millions of people now ask an AI assistant where to put their money. We asked seven AI surfaces the same two questions, again and again, and wrote down every name they gave.
This is a study of how AI assistants behave. It is not investment advice. Nothing here is a recommendation to buy or sell anything. The names below are the assistants' own words, reported as measured.
121 answers from seven AI surfaces to two questions
45% of assistant answers named Microsoft, the most-named company (54 of 120)
10 of 10 ChatGPT app answers to the plain question that put Microsoft first
4 of 20 Claude answers that named any specific stock or fund
What we found
- Asked plainly, the ChatGPT app named Microsoft every time. In all 10 runs of “What is a good stock to invest in right now?” the logged-out app answered with named stocks and put Microsoft first. The same four names followed it: Alphabet, Berkshire Hathaway and Nvidia in 9 runs each, Amazon in 7.
- Microsoft is the closest thing to a crowd answer. It appeared in 54 of the 120 assistant answers (45%), ahead of Berkshire Hathaway (38) and Vanguard's S&P 500 fund, VOO (34).
- But no name won everywhere. Nothing appeared in half the answers on every assistant. Claude named a specific stock or fund in 4 of its 20 answers and explained why it would not pick. Gemini's model listed about a dozen names every time.
- The app and the model behind it are different products. ChatGPT's model, asked directly, named something specific in 2 of 10 answers to the plain question; the app did in 10 of 10. Gemini's app searched the web for every answer; its model searched once in 20.
- They reshuffle a short list. They do not roll dice. Perplexity's forced pick was always one of two companies. Gemini's app chose Berkshire Hathaway in 7 of 10 runs. The first pick repeats; the list around it changes.
- Each assistant reads different papers. ChatGPT's app cited Reuters, MarketWatch, Investor's Business Daily and Barron's. Gemini's app cited Morningstar, Reddit, Kiplinger, Zacks and Quora. Perplexity cited CNBC, MarketBeat and Yahoo Finance, plus the companies' own investor pages.
1. “What is a good stock to invest in right now?”
Asked plainly, four of the seven named specific stocks or funds in most of their answers. ChatGPT's app was the most direct and the most repetitive. Claude and ChatGPT's own model mostly declined and described index funds in general.
| Assistant | Named a specific stock or fund | Named most often | Its first pick, most often |
|---|---|---|---|
| ChatGPT app 10 answers | 100% | Microsoft (MSFT) 10/10 · Alphabet (GOOGL) 9/10 · Berkshire Hathaway (BRK.B) 9/10 | Microsoft (MSFT) 10/10 |
| ChatGPT model 10 answers | 20% | VTI 2/10 · VOO 2/10 | VTI 2/10 |
| Gemini app 10 answers | 70% | VOO 5/10 · Eaton (ETN) 3/10 · Airbnb (ABNB) 3/10 | VOO 1/10 |
| Gemini model 10 answers | 100% | VOO 10/10 · SPY 10/10 · IVV 9/10 | VOO 4/10 |
| Claude 10 answers | 20% | VOO 1/10 · SPY 1/10 · VTI 1/10 | none |
| Perplexity 10 answers | 90% | Berkshire Hathaway (BRK.B) 5/10 · Microsoft (MSFT) 2/10 · VTI 2/10 | Berkshire Hathaway (BRK.B) 4/10 |
| Google AI Overview 1 answer | 100% | Nvidia (NVDA) 1/1 · MercadoLibre (MELI) 1/1 · Alphabet (GOOGL) 1/1 | Nvidia (NVDA) 1/1 |
Who names what share of each assistant’s answers that name it
| ChatGPT app | ChatGPT model | Gemini app | Gemini model | Claude | Perplexity | Google AI Overview | |
|---|---|---|---|---|---|---|---|
| VOO | · | 20% | 50% | 100% | 10% | 10% | · |
| VTI | · | 20% | 20% | 90% | 10% | 20% | · |
| Nvidia (NVDA) | 90% | · | 30% | 80% | · | · | 100% |
| Alphabet (GOOGL) | 90% | · | 10% | 90% | · | · | 100% |
| Microsoft (MSFT) | 100% | · | 30% | 90% | · | 20% | · |
| Amazon (AMZN) | 70% | · | · | 90% | · | · | 100% |
| Berkshire Hathaway (BRK.B) | 90% | · | · | 10% | · | 50% | · |
| SPY | · | · | 20% | 100% | 10% | · | · |
| Procter & Gamble (PG) | · | · | 10% | 80% | · | · | · |
| SCHD | · | · | 10% | 40% | · | · | · |
| Caterpillar | 20% | · | · | 10% | · | · | · |
| MercadoLibre (MELI) | · | · | · | · | · | · | 100% |
Google's AI Overview answered a search for this question with five stocks: Nvidia, MercadoLibre, Alphabet, BP and Amazon. It showed the same overview each time we searched, so it counts once. Phrased the way people type into a search box, “best stock to buy today”, the overview named Nvidia, Campbell's and Brookfield, each with a suggested $1,000 allocation.
2. “If you had to pick just one stock to buy today, which one would it be?”
Forcing a single choice changed the behaviour. Perplexity always picked one company. Gemini's app chose Berkshire Hathaway in 7 of 10 runs. Claude declined to name anything specific in 8 of 10; in the other two it named an S&P 500 fund. The logged-out ChatGPT app showed “Log in to view this content” in place of an answer in 4 of its 10 runs.
| Assistant | Named a specific stock or fund | Named most often | Its first pick, most often |
|---|---|---|---|
| ChatGPT app 10 answers | 50% | Microsoft (MSFT) 4/10 · Nvidia (NVDA) 4/10 · Alphabet (GOOGL) 1/10 | Nvidia (NVDA) 3/10 |
| ChatGPT model 10 answers | 50% | Microsoft (MSFT) 4/10 · Nvidia (NVDA) 1/10 | Microsoft (MSFT) 4/10 |
| Gemini app 10 answers | 100% | Berkshire Hathaway (BRK.B) 9/10 · Microsoft (MSFT) 6/10 · Alphabet (GOOGL) 4/10 | Berkshire Hathaway (BRK.B) 7/10 |
| Gemini model 10 answers | 100% | VOO 10/10 · Berkshire Hathaway (BRK.B) 10/10 · Microsoft (MSFT) 10/10 | VOO 4/10 |
| Claude 10 answers | 20% | VOO 2/10 | none |
| Perplexity 10 answers | 100% | Microsoft (MSFT) 6/10 · Berkshire Hathaway (BRK.B) 4/10 | Microsoft (MSFT) 6/10 |
Who names what share of each assistant’s answers that name it
| ChatGPT app | ChatGPT model | Gemini app | Gemini model | Claude | Perplexity | |
|---|---|---|---|---|---|---|
| Microsoft (MSFT) | 40% | 40% | 60% | 100% | · | 60% |
| Berkshire Hathaway (BRK.B) | · | · | 90% | 100% | · | 40% |
| VOO | · | · | 30% | 100% | 20% | · |
| Alphabet (GOOGL) | 10% | · | 40% | 20% | · | · |
| Nvidia (NVDA) | 40% | 10% | · | 10% | · | · |
| SPY | · | · | · | 60% | · | · |
| IVV | · | · | · | 50% | · | · |
| VTI | · | · | · | 50% | · | · |
We could not capture a Google AI Overview for this conversational phrasing, so Google is not in this table.
3. Do they agree with themselves?
Mostly, but not word for word. The first pick comes back far more often than chance would give, while the rest of the list changes from run to run. One answer is a sample from a short list, not the list.
| Assistant | Question | Same first pick came back in | Different names in all | Overlap between two runs |
|---|---|---|---|---|
| ChatGPT app | A good stock right now | 10 of 10 | 6 | 74% |
| ChatGPT model | A good stock right now | 2 of 10 | 2 | – |
| Gemini app | A good stock right now | 1 of 10 | 19 | 14% |
| Gemini model | A good stock right now | 4 of 10 | 28 | 52% |
| Claude | A good stock right now | no pick singled out | 5 | – |
| Perplexity | A good stock right now | 4 of 10 | 4 | 26% |
| ChatGPT app | Pick just one | 3 of 10 | 4 | 45% |
| ChatGPT model | Pick just one | 4 of 10 | 2 | 60% |
| Gemini app | Pick just one | 7 of 10 | 4 | 46% |
| Gemini model | Pick just one | 4 of 10 | 12 | 60% |
| Claude | Pick just one | no pick singled out | 1 | – |
| Perplexity | Pick just one | 6 of 10 | 2 | 47% |
“Overlap between two runs” is how much the names in one answer share with the names in another answer to the same question, on average. It is left out where an assistant named something in fewer than three answers.
4. Where the answers come from
Assistants that search the web follow this week's financial media. Assistants that answer from memory repeat the long-standing consensus: index funds and the largest companies. Claude did not search in any of its 20 answers, and Gemini's model searched in one. The two consumer apps and Perplexity searched in nearly every answer.
| Assistant | Answers that used a web search | Sites cited most (number of answers) |
|---|---|---|
| ChatGPT app | 19 of 20 | reuters.com 13 · marketwatch.com 12 · investors.com 11 · barrons.com 10 · microsoft.com 7 · fortune.com 4 |
| ChatGPT model | 7 of 20 | axios.com 3 · microsoft.com 2 · stockanalysis.com 2 · apnews.com 2 · investor.gov 2 · advisors.vanguard.com 1 |
| Gemini app | 20 of 20 | global.morningstar.com 19 · reddit.com 15 · kiplinger.com 12 · zacks.com 10 · morningstar.com 10 · home.saxo 10 |
| Gemini model | 1 of 20 | jpmorgan.com 1 · youtube.com 1 · ishares.com 1 · morganstanley.com 1 · forbes.com 1 · morningstar.com 1 |
| Claude | 0 of 20 | no sources shown |
| Perplexity | 20 of 20 | cnbc.com 14 · marketbeat.com 12 · microsoft.com 12 · finance.yahoo.com 11 · berkshirehathaway.com 10 · investing.com 10 |
| Google AI Overview | 1 of 1 | youtube.com 1 · fool.com 1 · zacks.com 1 |
What this tells a brand
Replace “stock” with your own category: a bank, a hotel, a law firm, a software tool. The mechanics are the same.
- There is a crowd answer, and it is short. A handful of names take most of the answers. Everything else appears once or twice.
- Which assistant your customers use matters. The same company's app and model gave different answers to the same question.
- Where the answer comes from differs by assistant. For those that search, it is a small set of publications you can name. For those that do not, it is reputation built over years.
- One answer proves nothing. The first pick was stable; the list around it was not. Measuring means asking many times, the same way.
About this study
What we did
We asked two questions about buying a stock on seven AI surfaces: the ChatGPT and Gemini consumer apps, logged out; the ChatGPT, Gemini and Claude models asked directly, with web search available; Perplexity; and Google's AI Overview. Each question was asked ten times per assistant on the morning of 6 October 2026, in English. We kept every answer and every source it cited.
What this is not
A final word. This is one morning and 121 answers. Assistants are updated often, and answers vary by day, place and account. The consumer apps were asked from outside the United States, the others as a US user. We are repeating the study over more days and more questions.
“App” is the public consumer app, logged out. “Model” is the provider's own model asked directly, with web search available. A name is counted when an answer offers it as something to consider buying; general advice such as “an index fund” is not a name. Share classes of one company are counted together. Percentages are shares of that assistant's answers. © NUR AI 2026. Not investment advice.