ChatGPT, Perplexity, Google AI Mode Cite Websites at Vastly Different Rates, Study Finds

·2 views

A researcher posed 47 identical buying-related questions to ChatGPT, Perplexity, and Google AI Mode on the same day, analyzing citations across all 141 responses. Perplexity averaged 16.7 citations per answer, Google AI Mode 3.7, and ChatGPT just 1.6, suggesting each platform behaves very differently as a traffic source for websites. The study also flagged a technical pitfall: JSON-LD structured data rendered via Next.js next/script is injected after hydration and invisible to non-JavaScript fetchers used by most AI assistants, meaning a plain script tag is necessary for reliable structured data delivery. Additionally, the researcher noted that ChatGPT's frequent Google Maps links were auto-constructed query URLs rather than verified business listings, warning against misreading hostname patterns as proof of citation. The findings suggest that optimizing content for traditional search results does not translate directly to visibility in AI assistant answers, as the two channels draw from different source types.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Voice AI Delivers ROI But Measuring It Remains a Major Challenge, Report Finds

A G2 report surveying 17 vendors and analyzing 1,419 verified reviews found that while 76% of voice AI users report significant operational ROI, measuring that value ranks as the third-biggest challenge in the category. Adoption has surged, with 88% of platforms reporting rapid growth over the past 12 months, yet many teams deploy these tools without analytics frameworks in place. Small businesses with fewer than 10 employees form the largest user segment, suggesting voice AI spread bottom-up rather than through enterprise procurement processes with formal measurement plans. The software itself earns strong satisfaction scores — a 9.26 out of 10 recommendation rating and 93.1% of reviewers giving four stars or higher — indicating users are pleased but struggle to quantify results internally. Experts and practitioners alike recommend establishing ROI measurement criteria before deployment, not after, to avoid spending months trying to prove value that was never tracked from the start.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Guide Warns AI-Assisted CRM Research Can Hide Uncertainty Behind Plausible Data

A workflow guide published on DEV Community cautions that AI-assisted company research can appear thorough while quietly becoming unreliable when unverified fields are filled with plausible but unsourced values. The guide recommends tracking four elements for every data field: the candidate value, its source, when it was observed, and a status of VERIFIED, NEEDS_CHECK, or NOT_FOUND. It emphasizes that NOT_FOUND should be treated as a legitimate outcome rather than a prompt for the model to guess, and that conflicting values should be held in a review record rather than silently overwritten. The proposed workflow separates retrieval, evaluation, and write-back steps, requiring human approval before any uncertain data is committed to a CRM. The guide also stresses distinguishing between what a system currently returns and what it could theoretically be designed to return, urging researchers to label unverified capabilities rather than treat them as confirmed facts.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Technical Writers Now Own AI Chatbot Answer Quality in Docs-as-Evals Role

The role of technical writers is evolving beyond traditional documentation as AI agents now consume product docs more frequently than human users do. A technical writer at a multi-product developer platform observed that their documentation chatbot was returning incorrect links and mixing product contexts when users switched topics mid-conversation. The root cause was a long chat history being carried across product switches, causing the AI to blend unrelated content from earlier turns into new answers. This has given rise to a practice called docs-as-evals, where technical writers test documentation quality by running real user queries against the AI and evaluating whether responses stay on-topic, cite correct sources, and handle context shifts accurately. Technical writers are seen as well-suited for this role because they already track user language patterns, source-of-truth pages, and product boundaries through their existing documentation work.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Independent Developer Builds Hardened Arch Linux OS, Drops Cryptographic Rootfs Protection

An independent Linux developer building a security app called RoamSwitch has begun creating a companion hardened operating system, RoamSwitch OS, to address security gaps that userland applications cannot reach. The project is based on Arch Linux and packaged as a bootable ISO, layering proven tools such as AppArmor, fapolicyd, auditd, Falco, and TPM2 rather than building components from scratch. The system is implemented as a Rust workspace with around 39 crates, each handling a specific defensive feature and running as systemd services or timer units, with modules sharing state through a unified incident timeline. The developer ruled out an immutable image-based approach similar to Fedora Silverblue or ChromeOS, citing the significantly greater engineering complexity involved. The project remains in a research and showcase phase, verified on real hardware and in QEMU but not yet recommended for production use due to the absence of third-party security audits or legal review.

0 comments Read more at DEV Community