Developer details challenges in accurately measuring AI search results for a product

·1 views

A developer built a custom script to analyze how AI search assistants recommend their product, DOS AI. The script tests 35 buyer questions across multiple runs, costing about one dollar per execution. Initial measurement errors included counting AI echoes of the product name as valid mentions and issues with script parameters. The developer found that many AI answers rely on pre-existing training data, not current web searches, limiting the impact of new content. Specific platform policies, like automatic 'noindex' tags or moderation queues, also prevented newly published articles from being considered in the AI's responses.

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)

SAP Partners with Anthropic's Claude for Core Business AI Reasoning Tasks

SAP has integrated Anthropic's Claude AI as its primary model for complex reasoning within its Autonomous Enterprise platform. The system, operational as of mid-2026, uses SAP's Business AI Platform to manage and orchestrate various AI models and pre-built agents for finance, HR, and supply chain tasks. Claude is specifically deployed for high-stakes tasks like financial analysis and workflow optimization, accessed securely through SAP's managed endpoints rather than direct customer contracts. The architecture also utilizes the Model Context Protocol (MCP) to standardize and secure interactions between different AI agents and enterprise systems.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI systems require multi-factor confirmation, not simple yes/no, to ensure secure writes.

An AI system's request for user confirmation should not be recorded as a simple yes/no boolean. A simple approval lacks the detail to specify which exact request, arguments, and destination were authorized. To ensure security, a confirmation must bind and verify multiple specific factors, including the exact data, target system, permissible fields, and a time window. This prevents misuse where the intent of an approval could be misinterpreted or redirected after the fact.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

FieldClass AI app turns textbook STEM concepts into outdoor field experiments

A developer has created FieldClass AI, an educational app designed to help students connect theoretical STEM concepts to real-world applications. The app generates short outdoor experiments, such as estimating tree heights or calculating average velocity, that users complete away from their screens. Built for high school and first-year college students, it encourages practical learning by having users take measurements outdoors before returning to input and check their results. The app was developed for the Hacktoberfest 2026 Open-Source AI Challenge and runs locally on users' machines using the Gemma 3 1B open-weight model.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

GhostAction campaign steals thousands of CI secrets via fake GitHub workflows

Since October 7, a campaign called GhostAction has been injecting fake security audit workflows into repositories via compromised maintainer accounts. The attack has impacted over 772 repositories according to GitGuardian, with StepSecurity reporting over 3,000 secrets stolen. The malicious workflows harvest credentials by accessing git history and exfiltrating them via plain HTTP to a single IP address. The campaign exploits existing automation trust rather than a GitHub zero-day vulnerability. Security researchers recommend auditing workflow files, reviewing run histories, rotating all potentially exposed secrets, and scanning full git history.

0 comments Read more at DEV Community