# Perplexity vs ChatGPT Search for Real Work

> Perplexity vs ChatGPT Search on citation accuracy, using the Columbia Journalism Review Tow Center study, and how to use either safely for work research.

- Canonical URL: https://webofmike.com/ai-life/work-study/research-reading/perplexity-vs-chatgpt-search/
- Author: Mike Moore (https://webofmike.com/about/)
- Published: 2026-11-03
- Last modified: 2026-11-03
- Tags: AI Life
- Cite as: Mike Moore, "Perplexity vs ChatGPT Search for Real Work", Web of Mike (webofmike.com), 2026-11-03. https://webofmike.com/ai-life/work-study/research-reading/perplexity-vs-chatgpt-search/


[AI Life](/ai-life/) / [Work & Study](/ai-life/work-study/) / [Research & Reading](/ai-life/work-study/research-reading/)

Both answer questions with links to sources. The best independent test of how accurate those answers are comes from the Columbia Journalism Review's Tow Center. More on the [research and reading hub](/ai-life/work-study/research-reading/).

## The study

The Tow Center tested eight AI search engines on finding and citing news articles. Perplexity answered incorrectly 37% of the time. ChatGPT Search was wrong 67% of the time ([CJR](https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php)). The headline: they're all bad at citing news, and Perplexity was less bad.

## What that means for work

- **Use them to find sources, not to be the source.** Click through and read.
- **Ask for links.** Then check that the link says what the answer claims.
- **Be careful with numbers.** Prices, dates and stats are where errors hide.

## Which one

- Research with sources: Perplexity did better in the test.
- Drafting and thinking through a problem: ChatGPT.

For students, see [best AI study tools](/ai-life/work-study/study-tools/best-ai-study-tools-college/). For research from your own documents instead of the web, [Gemini Notebook](/ai-life/work-study/study-tools/notebooklm-gemini-notebook-study-workflow/) sticks to your sources.

## What my agent can do with it

I let agents search the web, but I don't let them trust it. I have them return the link with every claim so I can check. Agents fetching web pages can also pick up prompt injection; I covered one version in [MCP discovery prompt injection](/mcp-discovery-prompt-injection/).