Chunk-Aware Writing for 800-Token Retrieval
Retrieval systems split pages into chunks: OpenAI’s vector stores default to 800 tokens with 400-token overlap. Write sections that stand alone when quoted.
Retrieval systems split pages into chunks: OpenAI’s vector stores default to 800 tokens with 400-token overlap. Write sections that stand alone when quoted.
An AI sales call is a telemarketing call under the Do Not Call Register Act 2006. Wash your lists, call in permitted hours, identify yourself, stop on request.
No major AI engine documents reading llms.txt from third-party sites; crawler docs name only robots.txt. Serving one is cheap; citation claims are unproven.
Australia’s Privacy Act reform arrives in stages: a serious-invasion-of-privacy tort live since June 2025, and ADM transparency due by 10 December 2026.
AI engines quote page positions, not whole pages. A CXL analysis of 100 Google AI Overview citations found 55% came from the first 30% of the source page.
Local pages can win AI citations because engines localise answers, but Google’s doorway spam policy targets region pages that funnel users to one destination.
Without a review-site profile you can still earn AI citations. Pre-launch and beta products have citable surfaces left, and this is the playbook we run.
AI crawlers split into two families: training crawlers such as GPTBot and ClaudeBot that feed future models, and on-demand fetchers serving a buyer reading now.
AI crawler allowlists need nuance: blocking GPTBot keeps content out of training, but blocking OAI-SearchBot or ChatGPT-User removes you from ChatGPT’s answers.
For citations, AI engines remain firmly text-first, so the citable layer of any video or podcast is its transcript and the page around it, not the media file.