This week's news: Small businesses using AI hiring faster, the insurance form that writes AI out of your cover, OpenAI's half-price models, the AI blog bundles Google just cracked down on, and a ChatGPT task that reads your Gmail before you do.
Read the whole thing in three minutes, or go deeper where it touches your business. We talk about the market here, not about ourselves.
Every article comes with a working tool. It runs right here in the page, it is free, and we do not need your information to use it. No signup, no download, nothing to fill in first. That holds for every article in every issue, back issues included.
Send it when it hits. Every article here has its own link; the send button next to each one copies it, ready to text or email to the person who needs it. Back issues stay in the archive and their tools keep working, so nothing you forward goes stale.
Want a reminder when the next one is up? Leave an email and we send one note a week. That is the whole program. We will not flood your inbox.
Gusto published payroll research on September 10 covering 2,262 of its small business customers. The 1,593 that adopted generative AI grew headcount about 7% faster over the next year than peers that did not. Firms under 10 employees grew about 10% faster. Teams of 10 to 24 showed no measurable change. The new hires were hands-on: technicians, cooks, front desk and care staff.
The gain did not come from AI replacing people. Gusto's economists tie it to owner time. AI takes the admin, and the owner spends the hours on growth. That is the part you can copy. Gusto says larger firms already have processes that protect the owner's time, which is why the effect faded above 10 staff. If your week still runs on inbox and paperwork, this is your move.
Do not read the 7% as proof that AI causes hiring. Gusto says early adopters differ in ways it cannot measure. Do not spend the freed hours answering more email; put them on sales, customers or the next hire. Do not hand AI the judgment calls first. Start with scheduling, drafting and summaries. Do not skip the tracking. Log your admin hours before and after for four weeks.
Look at who got hired. Not analysts or engineers. Technicians, cooks, therapists and front-desk staff. That pattern says the AI did not replace the work. It moved the owner off the paperwork. In a nine-person shop the owner is the scheduler, the bookkeeper and the email desk. Hand those jobs to ChatGPT, Claude or Gemini and the owner gets hours back. Those hours went into selling and serving, which created demand for more hands. Gusto's senior economist links the gain to owners getting back to their expertise. At 10 to 24 staff the lift faded, because someone already guards the owner's calendar.
Copy the mechanism, not the number. Spend one week logging every admin task you touch and the minutes it takes. Pick the three biggest. Look first at scheduling, quote and email drafts, and notes for your bookkeeper. Give each one a written prompt in ChatGPT, Claude or Gemini. Name one person who checks the output. Then block the freed hours on your calendar for sales calls and customer visits. The trade-off is the first two weeks. You spend time building and checking the prompts. After that the hours come back every week, and they go where the money is.
Verisk's ISO unit now offers three optional exclusions for general liability, all dated January 2026. CG 40 47 strips bodily injury, property damage and advertising injury claims arising out of generative AI. CG 40 48 strips just the advertising injury piece. CG 35 08 does the same for products and completed operations. A Gallagher study counts AI-related lawsuits up 978% from 2021 to 2025. Your carrier decides whether to attach the forms at renewal.
If your team drafts ad copy in ChatGPT, checks estimates with Claude or writes product pages in Gemini, those jobs sit inside the exclusion. The AI does not have to cause the loss. It only has to be in the chain. For a contractor or a retailer that is a real hole in the one policy that pays when a customer gets hurt or sues. You find out at renewal, or at claim time.
Do not accept the renewal on premium alone; get the forms schedule and read every endorsement number. Do not tell the underwriter you use no AI if your team drafts quotes, ads or emails with it. A wrong application answer is worse than the exclusion. Do not promise customers indemnity for AI-assisted work your policy no longer covers. Do not assume your cyber policy picks it up; ask in writing.
The forms turn on three words: arising out of. Courts read that phrase wider than caused by. So a claim only needs a link to generative AI, with no proof that the AI did the damage. Picture a takeoff checked in an AI estimating tool, ad copy that started as a ChatGPT draft, or a product page Gemini wrote. Under CG 40 47 the carrier can argue each one out of coverage. The forms are optional, so your carrier decides, and the decision shows up as a form number on the quote's schedule, where nobody reads.
Get the forms schedule 60 days before renewal. Search it for 40 47, 40 48, 35 08 and anything with artificial intelligence in the title. Ask for a second quote without them. Taking an exclusion off costs premium, so make it a priced decision, not a surprise. Then hand the underwriter a clean picture: which tools, which jobs, who reviews output before a customer sees it. Underwriters price what they can see. Last step: pull your top five customer contracts and read the indemnity clause, because your promise does not shrink when your policy does.
On September 22 OpenAI released GPT-6 Sol and GPT-6 Luna in its API and in ChatGPT Work. Per Benzinga and Technori, Sol runs $2 per million input tokens and $10 per million output, down from $4 and $20. Luna runs $0.10 and $0.50, down from $0.20 and $1.20. Quartz reports OpenAI calls the cut permanent. Sol targets coding and multi-step work; Luna targets extraction and summaries.
Every AI tool you pay for per seat, per document or per ticket sits on a model like these. The vendor's cost just halved and your invoice did not move. A thousand two-page invoices a month through Luna cost well under a dollar in model fees. Your price also covers setup, support and margin, fine, but the model share is now tiny. Any AI contract renewing this quarter is up for a rewrite.
Do not sign a multi-year AI contract at a fixed unit price without a clause that resets it when model prices fall. Do not run clerical work on the top model; extraction and summaries belong on the cheap tier. Do not buy on seat price alone; ask which model the tool runs and what one task costs. Do not swap models in a live workflow without testing twenty real jobs first.
A lot of AI software you buy is a layer on top of someone else's model. A vendor wraps an interface and a workflow around OpenAI, Anthropic or Google, then bills you by the seat or the document. The model bill is the vendor's cost line, and at one of the biggest suppliers it just halved. Your contract price was set when you signed and does not move on its own. When a supplier's cost drops under a fixed price, the seller keeps the gap. Nothing shady about it. That is what happens when the buyer never reopens terms.
Reopen them. Ask each AI vendor three things in writing. Get the name of the model that runs your work. Ask what one task costs them now, and whether the September 22 cut reaches your invoice. At renewal, trade a one-year term for a price review tied to published model prices. The trade-off is real. A vendor that passes cuts through will want to pass increases through too. New top-tier models do not always come cheaper. If you build on the API yourself, move extraction and summaries to Luna. Keep Sol for the hard jobs, and test twenty real documents before you switch.
Google ran a spam update from August 18 to August 21. It enforced existing rules against scaled content abuse, meaning masses of pages made to game rankings. That covers AI content at scale, programmatic pages and thin affiliate pages. Case studies show sites losing rankings on 14,000 to more than 200,000 queries. Google did not ban AI writing. Pages enriched with real photos, data and human review held or gained ground.
If an agency sells you a monthly bundle of AI-written posts, you are paying for the exact pattern Google just hit. At best those pages do nothing. At worst they sink good pages with them, because the case studies show whole sites falling, not single posts. Recovery takes months. The same budget buys fewer pages built from what only you know: real jobs, real prices, real customer questions.
Do not renew a content retainer priced by post count. Do not publish city pages that swap only the town name; that is the programmatic pattern the update targeted. Do not delete everything in a panic; pull Search Console, find the pages that lost clicks after August 18, and fix or cut those first. Do not stop using AI to write; stop using it to write without facts.
Google's spam systems do not care whether a person or a model typed the words. They look for the pattern: lots of pages, same template, nothing new, published to catch searches. A bundle of 30 AI posts a month fits that pattern almost by design, because the agency is paid on count. Once enough of a site looks like that, the damage spreads past the weak pages. That is why the August case studies show sites losing tens of thousands of queries at once. Analysts tracking the update put recovery in months, not weeks.
Do this instead. Open Search Console, compare clicks for the four weeks before August 18 with the four weeks after, and list the pages that fell. Cut or merge the thin ones. Then publish two pages a month with facts only you have. Show a finished job with photos, the real price range and this week's customer questions. Use ChatGPT or Claude to turn your notes and emails into a draft, and have someone who does the work edit it. The trade-off is volume for proof. Fewer pages, each one worth ranking.
On August 25 OpenAI added event-triggered tasks to ChatGPT Business. A task can fire on new Gmail messages, new Slack channel messages or GitHub pull request activity. Gmail triggers filter by sender or subject, and Slack needs @ChatGPT added to each channel. Members can share tasks inside a workspace, and each copy runs on the recipient's own permissions. On September 22 GPT-6 Sol and Luna reached ChatGPT Work.
Your inbox already holds the urgent work: the supplier invoice due Friday, the customer who is angry, the quote request that goes cold by tomorrow. A Gmail-triggered task reads those the moment they land and hands you a brief with the ask, the amount and the deadline. You stop scanning and start deciding. No Zapier account, no developer, one task and one filter.
Do not point the task at your whole inbox; filter by sender or subject or it summarizes newsletters all day. Do not let it answer customers; have it draft, and you send. Do not share a task with private details in its instructions; every copy carries the full text. Do not skip the Friday check of what it flagged against what it missed.
The trick is the filter, not the model. A task that reads every email buries you in summaries of promotions and receipts. A task that fires only on the senders and subjects that cost money is a different tool. Think supplier domains, your top ten customers, and subject words like invoice, past due, quote, cancel and refund. It runs as the message lands, pulls out who, what, how much and by when, and posts one short brief. You get the slice of your inbox that moves cash without opening Gmail.
Set it up today. In ChatGPT Business, connect Gmail, create a task, choose the Gmail trigger, and filter on your supplier and customer domains plus the subject words. Paste the instructions from the task recipe below. Make it answer in five lines, sender, ask, amount, deadline, risk, plus a suggested reply you edit and send yourself. Run it for a week, then compare what it flagged with your own read. The trade-off: ChatGPT gets read access to that mail, so keep payroll, legal and HR threads out of the filter.
The Briefing arrives weekly: what moved, what it changes, what not to do, and a working tool for each. Every issue gives the tools away, free and built for the week they cover. No product tours, no victory laps. If it stops earning the read, unsubscribe.
If this issue earns its three minutes, it earns a forward. Every article above has its own link, the button below copies the whole issue, and the archive keeps every back issue with every tool still running.
If something in here lands on a problem you are already working, and you would rather not work it alone, write to hello@goudegroup.com and tell us what you are exploring. You get a reply from a person, not a sequence.