This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Most of the WWDC coverage this week treated Apple’s ChatGPT tie-ins as a list of features. Image Playground gets ChatGPT styles such as oil painting, watercolor and vector art, which 9to5Mac walked through on Tuesday. Visual intelligence moves from the camera to the screen. Taken one at a time, they look like conveniences.
I think that reading misses what changes for companies. The interesting part is not that ChatGPT is a button press away. It is what the question arrives with.
What Apple announced
According to Apple, visual intelligence, which already helps people learn about objects and places through the iPhone camera, now extends to whatever is on the screen. Users press the same buttons they use for a screenshot, then can ask ChatGPT questions about what they are looking at, or search Google, Etsy and other supported apps for similar images and products. They can highlight a single object, Apple’s example is a lamp, and search for just that. It is in developer testing now and reaches users in the fall.
The usual model of AI and brands
The standard way to think about how AI describes a company starts with a typed query. Someone enters a company name or a product category, a system retrieves sources, an answer comes back. Reputation teams test those queries, track the answers and check which sources get cited.
Screen questions skip the typed query. The prompt is a screenshot plus a few words: what is this, is this any good, is this true.
Why the screenshot is the real prompt
A screenshot is never neutral. It is a crop of someone else’s page. It might be an influencer’s post holding your product, a competitor’s comparison chart with your logo in the losing column, a forum thread, a news headline, a parody. When a user asks ChatGPT about it, the model reads that framing as context. Your brand becomes the subject of the answer without being the subject of the question, and the question has already been shaped by whoever made the page.
A few consequences follow. The same product can get noticeably different answers depending on where it was encountered. A question asked over a review that calls a jacket overpriced starts somewhere different from the same question asked over the brand’s own product page. The search half of the feature routes to Google, Etsy and other apps, not to you, so the next screen is a set of similar items from wherever those apps send people. And there is no query log to study. The keyword is an image someone else composed.
The conventional take says integration is just distribution for ChatGPT. The more useful take is that the operating system now treats every screen as a potential prompt, and most screens showing your brand were not made by you.
What to do with this
It is too early for metrics. It is not too early to think in screenshots.
Start with the frames. List the third-party pages where your brand most often appears in someone else’s presentation: comparison sites, marketplace listings, review roundups, widely shared social posts, critical articles that still rank. Those are the screens most likely to be captured.
Then ask the questions a user would ask over each of them, in ChatGPT with the image attached, and read how much of the page’s framing survives into the answer. Where the frame is wrong or outdated, the fix is the familiar one: correct the source if you can, and make sure the facts that would settle a skeptical question, price, availability, safety, ownership, are easy for a model to find and confirm.
The brand is not the queried party here. It is the thing in the picture, described in the context of somebody else’s page. That is a weaker position, and it helps to know you are in it.