Ten publications, one standard: a network-wide search and AI-answer optimisation pass ahead of the next core update
Google shipped two core updates this year (March and May) and three spam updates (March, June and August). Each one rewards the same things: pages that load fast, say plainly what they are, carry no migration debris, and give search engines and AI answer engines something concrete to extract. Over 4 and 5 September we brought all ten publications on the network, 4,456 articles, up to one shared standard rather than waiting to see which site the next update would single out.
Improvement: Every site now publishes the same crawler policy, and it matches what our ai.txt files already promised. Search and answer agents (Google, Bing, OpenAI search, Perplexity, Claude search, Apple, Amazon, DuckDuckGo) are welcome. Model-training crawlers are not. Eight of the ten sites had a robots.txt that quietly contradicted this until now.
Fix: Images were the biggest technical debt. Hundreds of feature images were still loading from a third-party photo service, thousands were oversized PNGs, and none were cached at the edge. Every hero on every site is now stored locally as WebP with explicit dimensions and a one-year cache. Two of the smaller sites went from roughly 180 MB of images to under 35 MB.
Improvement: Search titles and descriptions: headlines are written for readers, and most ran well past the 70 characters a result snippet shows. Every long headline now has a dedicated search title, and every article has a real description instead of a body-text fallback. Across the network that was several thousand titles and descriptions.
Improvement: Tags were a mess left over from the Ghost era, with thousands of one-off terms, spelling variants and placeholders. Each site now uses a controlled vocabulary, every article was re-tagged against it, and a topic page exists only where there is real depth. One site went from 1,034 tags to about 210.
Fix: Migration residue in the articles themselves: link cards that had turned into raw blobs, bulleted lists glued onto a single line (over 17,000 items across four sites), duplicate headings and stale internal links. All repaired by script and checked.
Improvement: Key facts on news articles. Incident, enforcement and vulnerability stories now open with a short, extract-only facts list (who, what, when, how much, status) taken strictly from the article text. The same facts feed NewsArticle structured data so AI Overviews and answer engines can cite the number instead of paraphrasing the paragraph.
Improvement: Entity hubs and running trackers. Each site now has pages for the things it covers repeatedly: threat actors and breached organisations on breached.company, laws on compliancehub.wiki, companies and agencies on myprivacy.blog and scamwatchhq.com, states and vendors on cannasecure.tech, roles and certifications on securitycareers.help, tools and learning paths on hackernoob.tips, exchanges and regulators on cryptoimpacthub.com, devices, brands and building systems on the two SecureIoT sites. Trackers show where each story stands, grouped by jurisdiction, vendor or sector.
Improvement: A machine-readable index on every site: llms.txt with the hub structure and llms-full.txt with every article, regenerated on each build. Sitemaps carry a real last-modified date on every article, and three sites that had hand-made sitemaps now generate them properly.
Notice: What this is not: no content was rewritten for search engines, no facts were added that the articles did not already state, and no site was interlinked to another for ranking purposes. The work was structural. We will read the effect per site in Search Console after the next core update and report it here.