Week of August 18–24, 2026
Three things moved in AI this week that touch how your website shows up in search, how safe your data is with AI tools, and how you handle the customer texts piling up on your phone. Here’s what changed and what to do about it.
Google's August spam update went after AI-generated pages with no real value. If you've been using AI to pump out blog posts, check your traffic.
On August 18, Google released its third spam update of 2026, confirmed through the Search Status Dashboard. The rollout finished on August 21, according to Search Engine Land, a two-day, sixteen-hour rollout that applied globally and to every language. Search Engine Journal reported the update targets hacked content, scraped content, doorway pages, expired-domain abuse, and scaled content abuse, meaning bulk-produced pages, AI-written or not, that add nothing a reader couldn’t already get elsewhere. Google explicitly did not touch link spam or site reputation abuse in this round.
The bar Google uses isn’t whether AI wrote the page. It’s whether the page adds value. A blog post generated by an AI tool that answers a real question with specific, useful information isn’t a target. A page that exists only to catch a keyword, with generic filler and no actual local detail, is exactly what this kind of update looks for.
A fair number of Alaska trades and service businesses have started publishing AI-written blog posts or service-area pages, either doing it themselves or paying an agency that does it at scale, trying to rank for searches like "plumber Wasilla" or "roof repair Fairbanks." If those pages are thin, generic, and interchangeable with a hundred other contractors' sites, this is the kind of update that can tank them.
If your organic traffic dropped this week, check Search Console and see if the timing lines up with August 18–21. Then look at any AI-assisted pages on your site and ask honestly whether each one says something specific to your business and your service area, or whether it could be swapped onto a competitor's site with a find-and-replace.
OpenAI now guarantees zero data retention for its most capable models. If compliance was the reason you held off on AI tools, this is worth a second look.
On August 19, OpenAI announced it will continue offering Zero Data Retention, or ZDR, for its frontier models, meaning eligible customers’ prompts and responses are not stored after a request is processed, are not available to OpenAI staff for review, and are not used to train future models unless the customer opts in. Alongside it, OpenAI introduced Private Safety Processing, reported by Cybersecurity News, a system that flags potential misuse by category and severity without exposing the actual prompts or responses to a human reviewer.
OpenAI is explicit about who this is for: organizations working with financial records, health information, confidential business plans, or proprietary research, where how long data sits and who can see it are real constraints, not preferences.
Property managers holding tenant financial information, healthcare practice groups handling patient records, and distributors with vendor contracts all have a real reason to be cautious about feeding sensitive documents into an AI tool. This isn't a promise on a marketing page, it's an actual eligibility program. It doesn't apply automatically to every ChatGPT plan or every OpenAI product, so it's worth checking which tier you'd actually be using before assuming you're covered.
If data retention was the reason your business held off on AI tools for sensitive records, check whether the specific product you'd use, API access, ChatGPT Enterprise, or otherwise, qualifies for ZDR. It's not automatic on every tier, and it's worth confirming in writing before you send anything sensitive through it.
ChatGPT can now read and reply to your iMessages and texts on a Mac. Real option for anyone drowning in customer texts, with a permission setting worth reading twice.
On August 20, OpenAI released an Apple Messages plugin for the ChatGPT desktop app on Mac. 9to5Mac confirmed it lets ChatGPT read, search, summarize, and send messages across iMessage, SMS, and RCS threads, available on every ChatGPT plan including free, but only on Apple silicon Macs. Setup requires granting Full Disk Access along with contacts and automation permissions. By default, ChatGPT asks for approval before sending anything, an “Allow once” prompt per message. There’s also an “Always allow sending to this chat” option that lets it send to that thread going forward without asking again.
A lot of solo operators and small crews run scheduling and quick customer back-and-forth over text, because that's what customers actually use, and it piles up fast when you're also on a job site. Being able to ask "what did that customer on Spenard Road say about the estimate" instead of scrolling a thread is a real time saver.
If you're on an Apple silicon Mac and your texts are backing up, it's worth trying. Read the permission prompts carefully before turning on "always allow" for any thread, that setting lets ChatGPT send messages to that customer without asking you first.
What we're watching next week
-
Spam update fallout. Google says this round doesn’t touch link spam or site reputation abuse. We’re watching whether small business and local sites see sharper ranking swings than past spam updates, and whether “AI content is fine if it adds value” holds up once site owners start comparing notes.
-
ZDR eligibility rollout. OpenAI hasn’t published a full list of which specific products and tiers qualify for Zero Data Retention. We’re watching for that list, and whether Anthropic or Google match it with similar guarantees for their own enterprise tiers.
-
Apple Messages plugin adoption. The plugin is Mac-only and Apple-silicon-only for now. We’re watching for a Windows or Intel Mac version, and for early reports of the “always allow” setting causing a message to go out that shouldn’t have.