Only 14.4% of ChatGPT responses involve a web search, Ahrefs analysis finds

·1 views

Ahrefs examined 66.6 million free, logged-out ChatGPT responses from August and September 2026, published October 12, 2026. Just 14.4% of these triggered a web search, meaning the vast majority were generated from training data alone. When searches did occur, ChatGPT ran about 1.6 targeted queries per response rather than broad crawls. Site-specific lookups mostly targeted official brand websites, and the wording of these queries shifted notably toward terms like "official" and "price" over a single month. Ahrefs also reported that ChatGPT's Reddit citations fell 88% over six months, while Perplexity still draws roughly a fifth of its citations from Reddit.

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)

Terraform's sensitive marking hid a database change, causing nine minutes of failed writes

A pull request merged in March pointed an order service at a reporting replica, causing every write to fail for nine minutes with a read-only transaction error. The change was hidden in a Terraform plan because a secret in the ECS task definition marked the entire JSON-encoded container definitions as sensitive. Reviewers approved the plan line they usually approve, not realizing the database host change was masked. The team reverted and reapplied, then moved the secret into AWS Secrets Manager so the plan shows full diffs again. They also added a policy check that fails pipelines on fully masked production attributes and a startup check that refuses to report ready if the database is not a primary.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Meeting Transcription Tools in 2025: What Works and What to Look For

AI meeting transcription, which converts spoken words into text using speech recognition and language processing, has matured considerably and now offers real-time transcription and speaker identification. Advances in transformer-based models such as Whisper have improved accuracy even in noisy settings, while hybrid on-device and cloud processing reduces latency and supports privacy. Modern integrations allow transcripts to start automatically and route results to CRMs or project tools. The article stresses that accuracy, measured by word error rate, depends on factors like microphone quality, accent handling, domain vocabulary, and noise robustness. It also notes that open-source options such as Whisper are free but may not suit every team's needs.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Idempotency Keys Help Prevent Duplicate Actions in AI Agent Tool Calls

As AI agents increasingly carry out business tasks such as refunds and order creation, retries after timeouts can cause the same action to run twice. The article explains that idempotency keys solve this by marking repeated attempts at one logical operation, so the system recognizes duplicates. It argues that a reliable setup needs stable operation identifiers, request validation, durable execution records, concurrency control, and downstream API support. The author stresses that this protection belongs in the execution architecture rather than in the language model's reasoning. The problem is framed as a distributed-systems issue, since a failed response does not always mean the underlying operation failed.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

CSS Flexbox Explained with Simple Examples

Flexbox is a CSS layout tool for arranging items in a row or column, addressing spacing, centring and navigation-bar problems that web designers long struggled with. Applying display: flex to a parent turns its direct children into flex items that can be aligned, spread, stretched and wrapped using a handful of properties. The article highlights six core properties, namely flex-direction, justify-content, align-items, flex-wrap, gap and flex, which together cover most one-dimensional layouts. Its central concept is the pair of axes: the main axis follows the flow of items, while the cross axis runs perpendicular to it, with justify-content and align-items each controlling one. The piece notes that its guidance draws on MDN's reference material and that confusion typically arises from applying properties to the wrong element or the wrong axis.

0 comments Read more at DEV Community