Founder @ AeroChat - AI agent platform to help you run customer service on autopilot. Gets paid for helping brands get cited on AI Search. Geeky about software.
AI support tools can now draft the knowledge articles your bot answers from. Useful. But if those drafts are built from old tickets, the bot learns last year's refund policy and repeats it with full confidence. Get each policy owner to sign off before go-live.
Google now lets shoppers buy from eligible US Shopify stores inside AI Mode, enrolled by default. Those buyers never see your site, your shipping page or your returns policy. Their first real contact with your brand is a support message. That reply is your new storefront.
AI QA tools can now score every support conversation, not a sample. Good. But most QA rubrics grade tone: greeting, empathy, polite close. A bot aces those every time. Make a wrong answer an auto-fail first, or you get 100% coverage of the wrong thing.
ChatGPT traffic "collapsed" in October? Check Direct. Its links reportedly stopped carrying UTM tags around Oct 8, so visits got filed as Direct. Annotate the dip, compare Direct on your old ChatGPT landing pages, and stop trusting a tag someone else controls.
Google says AI answers will now show more source links, higher up.
What it didn't share: click-through rates or any traffic numbers.
More links is a layout change, not a traffic promise. Baseline your clicks and engaged sessions now, then judge it on your own data.
Privacy for a support chatbot isn't about the model. It's about what the bot can reach.
Order status needs the order, not the customer's whole history. Scope access to the question, verify identity before revealing anything, log every lookup.
Search Console now splits out visual search (Lens, Circle to Search) with a new Multimodal filter.
If your text clicks dipped, check the new bucket before rewriting anything.
If people find you by camera, your images are search entry points now. Treat them that way.
Meta and Sierra proposed a standard for customers' AI agents to deal with businesses directly: identify, sign in, change an order. Early days. Ask now: which actions would you let a customer's agent take without a human? Can't answer? Your policies aren't written down.
Containment rate flatters your support bot. A customer who gives up on the chat and calls an hour later still counts as "contained." Track who comes back within 7 days, on any channel, about the same issue. That is your real resolution number.
Google, this month: scaled content is now becoming more of a problem than link spam.
The SEO risk audit used to start with your backlinks. Now it starts with your own CMS.
Every page you published because it was cheap is a page you've asked Google to judge.
New pitch this week: one AI agent across every channel. Fine. Test your own support first: start an issue on chat, finish it by email. If you have to explain twice, your problem is a split customer record, not the model.
Google's new model in AI Mode briefly shipped answers with no links or citations. Fixed now, but the lesson stands: your citation link is a UI choice you don't control. Track whether your brand is named in the answer text, not just whether you're linked.
Cresta surveyed 300 CX leaders: only 9% of support conversations are handled entirely by AI. 76% are a person and AI together.
So the moment that matters most is the handoff. If customers repeat their order number after the bot escalates, fix that first.
Field experiment, 1,100 Google users: AI Mode cut clicks to websites by 18.8 points, and people liked Search less.
Don't wait for Google to reverse course. Treat every organic visit as a chance to earn an email or opt-in.
The traffic is rented. The contact is yours.
Google's updated AI content guidance: manually fact-check everything before publishing, including titles, meta descriptions, structured data and alt text.
That's the part nobody reads.
The cost of AI content was never the writing. It's the review.
New Alchemer survey of 2,009 US consumers: 37% got AI support replies that didn't understand the question. 20% said AI made resolution slower.
Instant replies are table stakes now. Stop reporting time to first response. Report time to resolution.
Google is testing paying publishers for content used in AI Overviews and AI Mode. Early payouts look small.
Ignore the cheque. The useful part is the report: which of your pages Google's AI actually relies on. That's your real authority map, not your rank tracker.
Microsoft's new contact center voice models can start understanding a caller before they finish the sentence. Impressive.
But voice bots rarely fail on speed or tone. They fail because they can't see the order or issue the refund.
A faster dead end is still a dead end.
AI Overviews now show on 80%+ of branded Google searches (DemandSphere, late Sept). The first thing a buyer reads about you is Google's summary, built from reviews, Reddit and comparison pages. Not your homepage. Go search your own name today.
Gartner looked at 432 AI support deployments: 25% showed positive ROI, 42% couldn't measure it at all.
Savings show up this quarter. Churn shows up later, on someone else's budget.
If your bot report leads with containment, you're counting quiet quitters as wins.