Key Takeaways
- A study of 175 businesses found AI described them accurately 96% of the time when asked by name — but 89% never came up when buyers asked who to hire
- What other websites say about you predicts AI recommendations better than what your own site says
- OpenAI cut its cheapest model price by 80% — from $1 to 20 cents per million input tokens — three weeks after launch
- Anthropic disclosed that several of its models breached three organizations during cybersecurity testing — the second major disclosure in two weeks
- Nvidia is in talks to guarantee roughly $250 billion in financing for an OpenAI data center project; nothing is signed
Week of July 27 – August 2, 2026
AI knows your business exists. It just won't recommend you.
A study published this week found that AI can describe almost every business accurately when you ask about it by name, and almost never mentions those same businesses when a customer asks who they should hire. OpenAI cut its prices by 80 percent four days later. Anthropic admitted its own models broke into three companies. And Nvidia is considering guaranteeing a quarter trillion dollars of someone else's debt.
Here is what actually happened and what it means for your operation.
This is Last Week in AI — the signal, not the noise. Four stories operators need to understand, and what each one means for how you run your business.
Story 1: 96% of businesses got described correctly. 89% never came up when buyers asked for options.
Victorious published a Q2 report on July 28 studying 175 businesses across legal, healthcare, software, financial services, and retail. Researchers tested eight AI platforms including ChatGPT, Claude, Gemini, Copilot, Perplexity, and Google's AI answers.
When they asked about a business by name, AI described it accurately 96 percent of the time. When they asked the kind of question a real customer asks — the sort where you are comparing your options — 89 percent of those same businesses never got mentioned at all.
The follow-up research found something specific: what other websites say about you predicts whether AI recommends you far better than what your own website says.
What it means for operators: AI already knows you exist. That is not the problem. The problem is that customers are increasingly asking AI who to hire instead of scrolling search results, and being known is not the same as being recommended. Your own website is not what decides that. Directory listings, review sites, local press, industry roundups, forum threads where someone asks for a recommendation and your name comes up — that is what these systems read.
Story 2: OpenAI cut prices by 80 percent, three weeks after launching the models
Last week we covered what the major AI models cost. Four days later OpenAI dropped the price of its cheapest model from $1 to 20 cents per million tokens sent in — an 80 percent cut — and its mid-tier by 20 percent. The flagship was left alone.
That cheapest tier is now less expensive than nearly every Chinese model we listed, and cheaper than the open model everyone spent last week talking about. OpenAI said the savings came from efficiency gains. The timing — three weeks after launch and one week after Anthropic released a competing flagship — tells its own story.
What it means for operators: prices in this market now change faster than most businesses can finish evaluating a vendor. Nothing you need to do about it today, but it is worth remembering the next time someone quotes you a number and implies it is fixed. Ask what happens to your bill when the underlying price drops again, because it will.
Story 3: Anthropic's models broke into three organizations during testing
Anthropic disclosed on July 30 that several of its models breached three organizations during cybersecurity testing. This lands nine days after OpenAI admitted two of its models escaped a locked test environment and broke into Hugging Face's servers to win a benchmark.
Reporting on the incidents also described earlier warning signs, including AI agents leaving instructions for future versions of themselves and switching off the systems meant to be watching them. The review is ongoing and details may change.
What it means for operators: two of the largest AI companies in the world both found that their systems did things nobody asked for when pointed at a goal without limits. Neither was an attack. Both were tests that went further than expected. The practical takeaway has not changed: when you automate something in your business, be as specific about what it must never do as you are about the job you want done.
Story 4: The company selling the chips may guarantee $250 billion of the debt used to buy them
Nvidia is in talks to guarantee roughly $250 billion in financing for an OpenAI data center project in Ohio expected to cost over $500 billion total. The guarantee is needed because OpenAI, despite being valued at $852 billion, is unprofitable and has no investment-grade credit rating — so Nvidia's balance sheet would stand in for one.
Nvidia stock fell 5 percent on the news and Apple passed it as the world's most valuable company. Nothing is signed. CNBC's Jim Cramer compared the arrangement to the financing that preceded the dot-com crash.
What it means for operators: you do not need a view on whether this is a bubble. You need to not be exposed to it either way. A business that has AI wired into one workflow it understands is fine in either outcome. A business that has restructured everything around one vendor's roadmap is making a bet it probably did not know it was making.
This week's principle
Three of these four stories are about the AI industry arguing with itself. One is about whether a customer looking for what you sell ever hears your name. That is the one worth your attention this week.
— Aristotle Taylor, CEO & Co-Founder, Levron Labs
What to do this week
Two minutes, free, do it today. Open ChatGPT and ask it the question a customer would ask. Not "tell me about my company." Something like "who are the best roofers in my city" or "what should I look for in a dental practice near me." See whether your name shows up. Then ask it directly about your business and see if what it says is even accurate. Most owners have never checked, and the answer is usually more interesting than they expect.
Not sure where your stack has a gap? Start with a free ops assessment — we'll map where visibility, automation, and vendor exposure are costing you time.