Unprompted

Operated by Skald Studio, which sells AI visibility work. No placement on this chart is for sale.

AI Coding Assistants · measured 2026-11-16

AI names Cursor first in 32% of answers.

When someone asks an AI assistant “what should I buy?”, which brands does it recommend? Every week we ask ChatGPT, Claude, Claude Code (local), Codex (local), Gemini, Perplexity the same 15 buying questions, 5 times each, and count who they name. Nobody can pay to appear, and every answer is public.

Questions

15

Real buying questions, worded the way people ask them.

Answers read

449

Each question asked 5 times per assistant.

Brands named

26

Every product any assistant recommended.

AI assistants

6

ChatGPT · Claude · Claude Code (local) · Codex (local) · Gemini · Perplexity

The board

Who the assistants recommend

Open full board
Published 2026-11-16Method v5Each question asked 5× per assistant6 assistants: ChatGPT · Claude · Claude Code (local) · Codex (local) · Gemini · Perplexity
Measured 2026-11-16No newer measurement has publishedUpdates every MondayCheck the raw data →

The methodology version changed; movement is not comparable.

What stands out

  • Cursor leads Claude Code by 2 points on first picks (32% to 30%). At this sample size that is too close to call: treat them as level.
  • GitHub Copilot comes up in 82% of answers but is named first in 22%: the widest gap in the top eight between being mentioned and being the first pick.

SquaresOne per question. Darker means more of its answers named the brand (5 per assistant, all assistants together).

NamedAnswers that mentioned the brand at all, out of every answer.

FirstHow often it was recommended first. The ranking uses this.

Δ changeUp or down since the last comparable reading. Small moves get no arrow: the sample cannot back them.

Every term explained →

15 questions · hover or tap a square to see that question and who leads it
#BrandBy question
01Cursor351/44932%
02Claude Code327/44930%
03GitHub Copilot366/44922%
04CodeRabbit30/4496%
05Replit Agent35/4492%
06Devin20/4491%
07Windsurf104/4491%
08Codex181/4491%
11Greptile20/4491%
12Aider49/4490%
13Augment19/4490%
15OpenCode21/4490%
16Qodo13/4490%
17Cline53/4490%
18JetBrains AI32/4490%
19Amazon Q29/4490%
20Continue22/4490%
21Tabnine14/4490%
22Zed14/4490%
23Bolt.new10/4490%
24Lovable9/4490%
25OpenHands9/4490%
26Roo Code7/4490%

At a glance

Mentioned often is not the same as recommended first

The pale bar is how often a brand came up at all. The solid bar is how often it was the first pick. When the two differ, the assistants know the brand but reach for something else first.

01Cursor

Assistant by assistant

Each assistant’s favourite

Taken one at a time, the 6 assistants split between 3 favourites: Claude Code, Cursor and GitHub Copilot.

ChatGPTClaude Codefirst in 31 of 75 answers
ClaudeCursorfirst in 26 of 75 answers
Claude Code (local)Claude Codefirst in 31 of 75 answers
Codex (local)Cursorfirst in 29 of 75 answers
GeminiCursorfirst in 43 of 75 answers
PerplexityGitHub Copilotfirst in 26 of 74 answers

How it works

Three steps, every Monday

  1. 01

    Ask

    We put the same 15 buying questions to each assistant, worded like a real person would. No brand names in the question.

  2. 02

    Repeat

    AI rarely gives the same answer twice, so each question is asked 5 times. One answer is a coin flip; 5 is a pattern.

  3. 03

    Count

    We record every brand named and which came first, then publish the totals with the raw answers so anyone can check them.

Other boards

Latest published reading in each category

All categories

Common questions

Before you trust a number here

Can a brand pay to appear or rank higher?

No. Unprompted is run by Skald Studio, which sells AI visibility work, and says so on every page. No company has any say in the questions, the method or the results, no placement is for sale, and no charted company is a Skald Studio client.

Does being named first mean a product is the best?

No. It records what the assistants said, not what is good. A brand named first in 30% of answers is simply the one these assistants put first in about three of every ten answers.

Why ask every question 5 times?

AI assistants rarely give the same answer twice. One answer is a coin flip, so each question is asked 5 times per assistant and the page reports how often a brand came up.

Is this what I would see in the ChatGPT or Gemini app?

Not exactly. We ask each provider's API, with its own web search where it has one. The consumer apps add their own instructions, personalisation and shopping features, which an API does not reproduce.

Can I check the numbers myself?

Yes. Every question, every recorded answer, the code and the full history are public. Each board links to the answers behind it.