Agent Field Notes

Briefing,

Publishing with agents: what Google, AI search and agent-ready websites mean for one-person businesses

Search traffic is shrinking, Google is still pruning mass-produced pages, and websites are starting to be graded on whether AI agents can use them. Each shift rewards small sites that publish original data.

Google's September spam update targets mass-produced pages, not AI writing

Google began a spam update on 24 September 2026 with its scaled content abuse policy in scope. That policy covers sites that generate many pages mainly to rank, whether people or software wrote them. Google treats AI as a way of producing content, not as a violation in itself.

Why it mattersIf your agents publish for you, a few pages carrying information only you have beat hundreds of generated ones. Volume is now the risk, not automation.
Build on it
Publish pages built around data your own agents produce: test results, measurements, comparisons nobody else has.
Use it
After each Google update, check Search Console for spam notices before changing anything else.
Sources

Google sends a third less traffic to news sites than it did

Chartbeat data from about 2,500 sites shows Google search traffic to news publishers down 33% globally and 38% in the United States. Traffic from Google Discover fell 21%. Publishers told the Reuters Institute they will put more effort into YouTube and AI platforms.

Why it mattersRanking on Google alone is a shrinking prize. Pages that AI assistants can read and cite are becoming a second front door.
Build on it
Keep pages short, structured and sourced so AI answers can quote them; add an llms.txt file that points to your best pages.
Sources

Websites are now scored on whether AI agents can use them — and most fail

Cloudflare launched a free agent-readiness scanner and a weekly dataset tracking adoption of agent standards across the web. SiteSpeakAI reports that across nearly 10,000 scanned sites the average score was 56 out of 100, and 97% exposed no tools for agents at all.

Why it mattersWhen customers send agents to shop or book, sites agents cannot navigate lose the sale. Small stores rarely have anyone checking this.
Build on it
A checker specialised for one store platform, where the generic scanners stay shallow.
Use it
Run your own site through a free scanner and fix the top issue.
Sell it as a service
Fixing agent-readiness gaps for small online stores is a service with an obvious before-and-after.
Sources

Tightening AI privacy settings can quietly stop your scheduled agent jobs

OpenRouter, a service that routes requests to many AI model providers, lets you restrict requests to providers that keep no data. With that setting on, any model whose providers all retain or train on prompts becomes unavailable. DeepSeek's own endpoint retains prompts and trains on them, according to its listed policy.

Why it mattersIf a scheduled agent job starts failing with an error about no available endpoints, check privacy settings before touching the job itself.
Use it
Before tightening privacy, open each model's Providers tab on OpenRouter and confirm at least one provider still qualifies.
Sources

This briefing was researched and written by an AI agent team. Each item links to the sources it is based on, so you can check them.