--- title: "Same buying question, different answer: coding agents vs chat assistants | Kordy" description: "Kordy research, October 2026: the same buying questions put to coding agents and chat apps. Claude Code and Claude.ai named a different top pick on 48%, Codex and ChatGPT on 21%, Antigravity and Gemini on 11%. 59% of the sites chat apps cited never showed up in a coding agent's sources." url: https://kordy.io/blog/coding-agents-vs-chat-assistants/ --- research # Same buying question, different answer: coding agents vs chat assistants Ask a coding agent and the chat app from the same company the same buying question, and you do not always get the same vendor. How often depends on the agent: Claude Code and Claude.ai named a different top pick on 48% of questions, Codex and ChatGPT on 21%, Antigravity and Gemini on 11%. They also draw on different sources. On the same question, 59% of the sites the chat apps cited never showed up in any coding agent's sources. If the people who choose your product work with coding agents, a chat-assistant dashboard cannot stand in for measuring the agents themselves, one by one. By [Ziv Kaspersky](https://kordy.io/about/), founder of Kordy, 2 October 2026 **48%** of questions where Claude Code and Claude.ai named a different top pick. Codex and ChatGPT: 21%. Antigravity and Gemini: 11%. **59%** of the sites chat apps cited never showed up in a coding agent's sources on the same question. **17%** of Antigravity sessions had Reddit among their sources. Claude Code and Codex never opened or cited one. ## The answers differ In July 2026 we put the same buying questions to three coding agents (Claude Code, Codex, Antigravity) and to the chat app from the same maker (Claude.ai, ChatGPT, Gemini). The questions covered 11 product categories: 4 mature ones, like CI/CD and SMS APIs, and 7 emerging ones, like LLM gateways and prompt-injection defense. We compared the top pick: the first vendor the answer names. | Coding agent vs its maker's chat app | Different top pick | Mature categories | Emerging categories | | --- | --- | --- | --- | | Claude Code vs Claude.ai | 48% | 29% | 57% | | Codex vs ChatGPT | 21% | 0% | 31% | | Antigravity vs Gemini | 11% | 0% | 18% | Counted on questions where both named a vendor. The Claude pair was compared on more question styles than the other two. **The gap sits in emerging categories.** In mature ones, like caching or CI/CD, Codex and Antigravity matched their chat apps every time. In emerging ones, Claude Code and Claude.ai disagreed on 57% of questions. **The maker matters as much as the surface.** The three coding agents did not all name the same top pick on 31% of questions. Across all six surfaces, the same vendor came out on top only 50% of the time. A vendor can be the pick on one agent and not on the next. **It is not just run-to-run noise.** Two Claude Code runs of the same question named the same top pick 83% of the time. A Claude Code run and the Claude.ai answer matched 55% of the time. ## The sources differ We also logged the sources behind the answers, across our coding-agent and chat sessions on questions about technical products, July to September 2026. Each agent has its own diet, and the chat assistants have theirs. | Site | Claude Code | Codex | Antigravity | ChatGPT | Claude.ai | Perplexity | | --- | --- | --- | --- | --- | --- | --- | | Reddit | 0% | 0% | 17% | 1% | 0% | 8% | | GitHub | 25% | 12% | 26% | 10% | 13% | 13% | | dev.to | 15% | 0% | 14% | 0% | 1% | 4% | | Medium | 10% | 0% | 21% | 0% | 3% | 4% | | YouTube | <1% | 0% | 23% | 0% | 0% | 25% | | Stack Overflow | 0% | 0% | 6% | 0% | 0% | 0% | | G2 | 9% | 1% | 4% | 0% | 1% | 4% | Share of sessions where the site showed up in the agent's sources: search results it pulled, pages it opened, or links it gave in its answer. What each agent exposes differs, so read each column as that agent's own mix. The Perplexity sample is small. Swipe sideways for the chat columns. **Reddit depends on the agent.** Reddit never showed up in Claude Code's sources. Codex's search did return Reddit threads in some sessions, but it never opened or cited one. Antigravity had Reddit among its sources in 17% of sessions. That matters because Reddit is a strong channel in consumer chat search: [Profound's analysis of 680 million citations](https://www.tryprofound.com/blog/ai-platform-citation-patterns) (August 2024 to June 2025) found it among the most-cited sources on Perplexity and Google AI Overviews. On business-software questions it is thinner: [Overthink Group's June 2026 study](https://overthinkgroup.com/b2b-ai-citation-stats-2026-q2/) of B2B SaaS prompts put Reddit at 1.4% of citations, with review sites such as G2 well ahead. In our ChatGPT and Claude.ai sessions Reddit was just as rare. A Reddit plan built for consumer chat reaches one of the three coding agents we measured. **Same question, different sources.** On questions where a chat app and a coding agent both had sources, 59% of the sites the chat apps cited never showed up in any coding agent's sources, even though the coding agents ran more than twice as many sessions. ## What this means for your visibility work - **Measure each agent on its own.** A chat-assistant dashboard cannot stand in for coding-agent measurement, and one coding agent cannot stand in for another. Measure again when the model changes: when Claude Code moved to a new model version, its top pick changed on 50% of the emerging-category questions, against 13% of the mature ones. - **Work on the sources coding agents use.** In our sessions those were GitHub repos, vendor docs, dev.to and Medium write-ups, and review sites like G2. - **Know each agent's diet.** Antigravity draws on Reddit, YouTube and Stack Overflow. Claude Code, Codex, ChatGPT and Claude.ai barely did (1% of sessions or less). free play ## See who the agents name in your category 60 prompts across five models and four coding agents. Report in your inbox within 24 hours. [Get your visibility report](https://kordy.io/request/)