ChatGPT chases sales, Claude exposes shares: July 28 brief

ChatGPT ads can now hunt for buyers instead of stopping at clicks, which gives advertisers a better test and a new measurement burden. Claude users also got a blunt privacy lesson, while Threads moved AI closer to public conversations. These shifts hit anyone buying attention, sharing AI work, or protecting company data.

ChatGPT ads start chasing actual sales

OpenAI added conversion-optimized campaigns to ChatGPT Ads by July 27. These campaigns try to show ads to people more likely to complete a chosen action, such as buying or submitting an inquiry. Advertisers still pay for each click, and the maximum bid remains the most they can pay for one. OpenAI's campaign instructions say an existing campaign cannot be switched into this setup.

That makes ChatGPT ads easier to judge against Google and Meta. A click is only a visit. A conversion is the sale, booking, or inquiry that pays the bill. OpenAI can now use those completed actions to guide where ads appear, but only when the tracking is sound.

The catch sits in the measurement. OpenAI's tracking can combine a small website code called a pixel with information sent directly from a company's system. It may also estimate some missing results. Owners should expect its total to differ from their own sales records, just as we explain in our plain-English attribution setup.

Public Claude links turn up in Google

Axios found publicly shared Claude creations in Google results on July 27, including business plans and infrastructure diagrams. Claude conversations and Artifacts, its name for documents or small apps created inside a chat, begin private. A user can create a public link, though, and Anthropic told Axios that a link posted somewhere visible to a search engine may be indexed.

The business lesson is immediate. “Anyone with the link” is a public setting, even when the link was made for one coworker. Search indexing simply made that gap easier to see. A proposal, customer list, internal plan, or account detail does not belong behind a public AI share link.

Audit the shared items in every company AI account today. Remove access to anything sensitive, then replace public links with files stored in a system where named people must log in. This is also a useful moment to set the same rule for ChatGPT and other assistants. Our customer-service AI guide applies the same principle: keep sensitive cases with people and controlled systems.

Threads puts Meta AI inside messages

Meta added one-to-one conversations with Meta AI inside Threads messages on July 27. On mobile, a person can send the assistant a Threads post, image, or video and ask questions about it. Meta says the conversation stays visible only inside that chat. The official announcement does not include the web version yet.

For a business active on Threads, this creates a new discovery step between seeing a post and visiting a profile. People can ask the assistant to explain a claim or add context without leaving the app. Clear, specific posts gain value because they give the assistant solid material to work with. Vague slogans give it almost nothing.

Treat Threads as a public reference shelf. Publish prices, dates, locations, and named services when they are relevant, then check that the profile points to a current destination. That matches the wider shift covered in how AI chooses which businesses to mention.

Microsoft gives security agents a bigger job

Microsoft announced Project Perception on July 27, an AI security system that looks for weak points, investigates risks, and can take corrective action while people retain control. The company says its first public preview begins August 3. Microsoft's launch post positions the system as a response to attacks that can now be created and scaled by AI.

Small firms do not need to buy a new security platform because of one launch. They do need to ask whoever manages their technology whether AI services are included in the current security review. Shared assistants, connected apps, and public links create openings that an annual password reset will miss. Start with access: who can see company AI chats, what those accounts connect to, and how former staff lose access.

Worth watching

• Google AI search is pulling more research into generated answers, raising the value of clear facts on product and service pages. Our AI Overviews briefing covers the traffic consequence.
• OpenAI says ChatGPT ad reports may include estimated conversions. Watch whether advertisers receive a clear split between observed and estimated results.
• Meta's Threads assistant is mobile-only at launch. A web release would make it more practical for teams monitoring brand conversations.