THE AI EDGE · WEEKLY EDITION
The AI Edge, the weekly edition from Hub365 AI. Friday, September 11, 2026 · 6 min read · by Naty Ross and Todd Ross.
The read of the week
AI news shows up and most people do one of two things with it. They act, or they let it go. The trouble is there are three kinds, and the third one takes neither. So it just sits there, coming back every time somebody mentions it.
One question sorts them, it works on any headline, ours or anybody else's, and it takes about ten seconds: who has to move first?
You, today. Nobody else has to do anything for this one to reach you. September 3 was one of those: that morning you either had a second tool you had actually opened before, or the work you meant to do did not get done. It had nothing to do with the regulator, your competitor, or your client. Almost nobody gets these wrong, because the world pushes you, and because it stings enough that you remember.
Nobody. The news is real, it's interesting, it's everywhere, and it asks nothing of you. Watermarking in AI-generated text is one of those for almost everyone reading this. These get dropped, and the hard part is not knowing you should, it's giving yourself permission to.
Your competitor, before you. This is where people get stuck, and it's what wore everybody out this year. It opens a door that changes nothing for you this week, and starts changing things the day somebody in your category walks through it. Ads inside ChatGPT answers work like that. They take nothing from you right now. But the day a competitor of yours shows up paid inside the answer your client is reading, you're not being judged on your own anymore. You're being judged next to them, in the same paragraph.
Acting now means spending on something that might not reach your market until March. And letting it go does not work either, because you'll hear about it late and from somebody else.
What that third kind actually asks for is the ability to tell when something shifts. That's the real gap, and it's simpler and bigger than it sounds: you can't spot a change if you never saved what it looked like before.
Try it on your own year. When somebody told you AI was handing clients to a competitor, did you have anything to compare that to? Do you have, sitting in a folder somewhere, what an AI said about your service back in March? In June? Last week?
Nobody does, and it is not that people are careless. Everybody archives what went wrong. Nobody archives normal. We keep the complaints, the invoices, the disputes. Normal never gets filed, because we assume normal is going to stay where it is.
That's why every shift this year showed up as a shock instead of a trend. Not because things moved fast, though they did. Because there was nothing to measure them against, so each one felt sudden and each one got handled in a panic.
Being current and being ready are two different things. Current is information, and you've lost it somewhere between Monday and Thursday. Ready is a saved snapshot: you do it once, you never touch it again, and it's already sitting there on the day you need it.
This week's play (3 steps, 20 minutes)
We're going to do this differently, because selling you as new a task we already gave you would be dishonest.
This week's play is the one that closes Monday's roundup: take the before picture. It's been published, free, since Monday at eight in the morning. You probably have not done it, and that's not on you: it landed at the bottom of a seven-item news roundup, with nothing explaining why that task and not one of the other twenty competing for your week. Sitting there, it looked like a suggestion. This edition exists to put the missing reason in front of it.
The caution, which we gave you in August and which still holds. A single query measures nothing: ask the same question twice and you'll get two different answers often enough that one run proves nothing on its own. Real measurement is the three-day, two-engine protocol, and this is not that.
What you walk away with: a dated file. Not a measurement, not a diagnosis, and not a list of things to do. All of its value comes from existing before something changes, which is why it's the one thing this week that cannot be done later.
The three steps with the reasoning behind each detail are in Monday's roundup.
This week on the blog
Monday · This week they put rules and a price on the place where you show up Europe classified ChatGPT's search as a very large online search engine, the first thing to carry that label, and it did it using the numbers OpenAI itself had filed with the regulator. The same week OpenAI opened ad buying in more markets and reported an annualized rate of a billion dollars in under two hundred days. Seven items in all, each with what it actually asks of you, and the roundup closes with the one thing this week that cannot be done later.
Tuesday · Claude now signs what it writes: why authenticity is your most expensive asset AI-generated text now carries a mark you cannot see, and what's being said about the European rule behind it is not what the rule says. We read Article 50 with the source open: the duty to label belongs to whoever builds the model, not to you for using it, and what does fall on you is reviewing and signing what you publish under your own name. It's the August 18 roundup, reworked Tuesday with the part we had not verified back then.
Wednesday · Cited or bought Now that ads have opened inside the answers, there are two ways to be in what your client reads: they cite you, or they buy you. Only one of the two keeps working the day you stop paying, and the piece runs the twelve-month math on both with the same budget put behind each. That's where you see why the order you do them in matters more than how much you have, and why starting with the bought one costs more than it looks like in month one.
Thursday · You rent the tool. You own the asset. Updated with the morning of September 3, when ChatGPT, Claude and Grok stopped answering within minutes of each other and Downdetector logged more than 35,000 reports for ChatGPT alone at the peak. Two different risks live in there and blurring them gets expensive: the tool disappears and takes something you cannot rebuild, or the tool is fine and simply does not answer that morning. One four-row sheet covers both, depending on how you read the right-hand column.
Tool of the week
GPT-6 Astra, from OpenAI. It landed September 3 and started reaching paid plans over the following days. Monday covered the price; this is the other half, and it comes straight from OpenAI's own help pages.
If you pay for ChatGPT Plus, you don't have it in the ordinary chat window. GPT-5.6 Sol is still the top option there. Plus covers Astra in ChatGPT Work and in Codex; inside Chat the model is called GPT-6 Pro, and OpenAI lists it for the $100 and $200 Pro plans, Business and Enterprise only. The start was rough enough that Sam Altman called the rollout "messy" on September 4, and the company is banking one usage reset for every day a paying user goes without it, counting from the 3rd.
Where it is, if you're on Plus. In the desktop app, menu at the top left, and the toggle at the top switches from Chat to Work. On mobile, Work sits in the dropdown at the top. And Work is not only for developers: OpenAI describes it for research, analysis, and finished documents, spreadsheets and presentations. Using it to write takes no coding skill, just the right door.
What it's for this week. For the before picture, with the detail that decides whether it's worth anything in December: write down which model answered you. If you used Plus in the chat window, Sol answered, not Astra. Without that line, three months from now you won't be able to tell whether the answer changed because the market moved or because your own plan did, and those two call for opposite decisions.
⚠ What isn't known: OpenAI has not said whether Astra will reach the chat window. The question has gone unanswered in its own developer forum since September 5, so anyone giving you a date is making it up.
This week's worksheet
Wednesday's is two pieces and they come together: the sheet My Two Boxes, which tells you in a few minutes how much of your visibility is yours and how much you're renting, and the guide How to split your budget between the two slots, with your ceiling per inquiry, your monthly cap, and the five questions to ask before signing an AI ads contract.
The sheet tells you where you're standing. The guide tells you how much to put on each side, and in what order.
Ask for them by writing the word BOXES on WhatsApp and they arrive right away: https://wa.me/17866869497?text=BOXES
Thank you for reading on Fridays. We know that by this hour one more thing to read is the last thing anybody wants, so if you take one thing from this edition, make it the twenty minutes on your calendar: it's the only thing this week that cannot be done later.
At Hub365 AI we don't create for the sake of creating. We create to make an impact, and that means when the week's play was already published on Monday, we'd rather tell you and hand you the reason that was missing than invent a new task to look useful.
