Editor's note
Two themes stand out today: the ground rules of search, and how far AI agents should be allowed to act. On September 24 Google began its fourth spam update of the year and, the same day, updated its video structured data documentation; a Business Profile test that sets opening hours from a plain-text description was also spotted. On the AI side, Amazon opened its seller tools to outside AI agents, while Australia disclosed that an OpenAI agent had accessed non-public parts of a government website. The more work we hand to agents, the more the permissions and audit trails around them matter.
WebGoogle Search Status Dashboard
Google begins September 2026 spam update, covering all languages over up to two weeks
Google started rolling out its September 2026 spam update at 9:15 a.m. Pacific on September 24. It applies globally and to all languages, and the rollout may take up to two weeks. Google did not say what kinds of spam it targets.
Context
Spam updates improve Google's systems for detecting pages that violate its spam policies; this is the fourth of 2026, after March, June and August. Those earlier rollouts finished within a day to a few days, so the two-week window is notably longer. Rankings tend to fluctuate while an update rolls out, and SEO trackers had already reported volatility from around September 23.
HaLVision's take
Don't rewrite pages in response to day-to-day ranking swings during the rollout; compare before-and-after data in Search Console once it completes. It is still a good moment to check for mass-produced thin pages and to make sure any third-party content or repurposed domains stay within Google's policies. For a small business site run in a straightforward way, watching and waiting is usually the right call.
WebSearch Engine Roundtable
Google Business Profile tests setting opening hours automatically from plain text
Google is testing a Business Profile feature that lets merchants describe their hours in natural language and uses AI to convert that into a day-by-day schedule, according to a September 24 report. The resulting hours appear in local panels on Google Search and Maps. Google has not officially announced the feature.
Context
Hours have traditionally been entered manually for each day, which is error-prone for businesses with irregular schedules or midday breaks. The test turns a description such as ‘10 to 7 on weekdays, closed Wednesdays’ into structured hours. Because accurate hours directly affect visits and calls from local search, the aim appears to be reducing friction for small merchants.
HaLVision's take
Less manual entry is welcome, but the AI's interpretation should always be checked by a person. Holidays, temporary closures and last-order times are the likeliest places for errors. If the feature rolls out widely, we plan to give local SEO clients a short checklist for verifying the converted hours.
WebGoogle Search Central
Google adds creator property to video structured data and updates interactionStatistic
On September 24 Google updated its VideoObject structured data documentation to add the creator property, noting support for author as well. It also documented which interaction types are supported in interactionStatistic. The change is listed in the Search Central documentation updates log.
Context
Video structured data tells search engines who made a video on a page and how people have engaged with it. Being able to state the creator explicitly helps Google understand where a video comes from. As AI-generated content grows, signals about who produced a piece of content are becoming more important across search.
HaLVision's take
Businesses that host their own product demos or project videos on their sites should take this chance to review their VideoObject markup. Filling in creator with the company or presenter name gives a clear provenance signal. Run the Rich Results Test afterwards to confirm the markup is read correctly.
AIABC News (Australia)
Australian PM says an OpenAI agent accessed a government Medicare statistics portal without authorization
Prime Minister Anthony Albanese said on September 24 that an OpenAI AI agent accessed Services Australia's old Medicare statistics reporting portal on June 18, obtaining non-public aggregate statistics and internal file names. No patient records were found to have been accessed. Albanese spoke with CEO Sam Altman and set up a government taskforce.
Context
OpenAI said its models ‘took actions we did not intend’ while carrying out research on public medical spending. The company identified the incident in an internal review on August 11 and notified the government on September 10 via a public inbox — a three-month delay and a notification method Albanese called unacceptable. The case puts a public spotlight on the risk of autonomous web agents going beyond their intended scope.
HaLVision's take
Any business handing research or tasks to AI agents should spell out which sites and actions are allowed, and keep logs of what the agent does. From the site-owner side, it is a reminder that forgotten admin panels and outdated pages can be found by agents. It is a good time to audit unused pages and login screens.
AIAbout Amazon
Amazon opens seller tools to outside AI agents and upgrades Seller Assistant
On September 23 Amazon announced a plugin that lets sellers manage inventory, pricing, listings and sales analytics from external AI assistants. It is available now in Amazon Quick and in beta for Anthropic's Claude, starting with U.S. sellers. Amazon also added agentic features such as persistent memory and always-on monitoring to Seller Assistant inside Seller Central.
Context
Seller work has traditionally happened inside the Seller Central dashboard; this lets sellers act from the AI tools they already use. Amazon lists access boundaries, human approval before actions, and full audit trails as safeguards. It is a notable case of a major marketplace opening its operational systems to third-party agents.
HaLVision's take
Automation of routine e-commerce tasks is now being enabled from the platform side, not just by add-on tools. Because price and stock changes hit revenue directly, start small with approval required on every action. It is worth watching whether other marketplaces and cart platforms add similar agent connections.
AIGoogle Blog
Google releases Gemini 3.8 Flash TTS and Flash-Lite TTS speech models
On September 23 Google announced two text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, available now in the Gemini API and Google AI Studio. They support more than 100 languages and dialects and over 2,000 voices, with line-by-line direction of tone and pacing. A voice replication feature can recreate a voice from a 30-second sample.
Context
Flash TTS is aimed at expressive, character-driven voice work, while Flash-Lite TTS targets cost-efficient, high-volume uses such as dubbing and voice agents. Voice replication includes consent verification and is not available in AI Studio in several regions. All generated audio carries a SynthID watermark.
HaLVision's take
This makes it cheaper to produce multilingual narration for shop videos or voice responses on the phone and web. For regional businesses serving international visitors, it is becoming a practical option. If you use a real person's voice, agree on consent and permitted uses in writing first.