WEEKLY SIGNAL
A GrowthGary briefing on the week of September 8-14, 2026
1. Premium AI models should be assigned to premium work
GPT-6 Astra is a meaningful jump in computer use, long workflows and professional output. It can navigate software, work across large amounts of context and complete multi-step tasks more reliably than earlier models. That makes it attractive for deep research, complex website work, coding and processes that move through several tools.
The important business lesson is not that every task needs the newest model. Astra is more expensive through the API, and simpler work rarely needs its full capability. Using a premium model for routine rewrites, short summaries or basic brainstorming can raise costs without improving the result enough to matter. The better approach is to match model strength to task difficulty.
GrowthGary’s take: Stop choosing one AI model for everything. Use a fast, lower-cost model for everyday production, then bring in a frontier model when failure is expensive, the task is unusually complex or the work requires sustained tool use. The real advantage is not owning the most powerful model. It is knowing when that extra capability creates business value.
What to do now: Pick three recurring tasks and run the same inputs through your normal model and a premium option. Score accuracy, editing time, completion rate and cost. Keep the stronger model only where it materially reduces human work or produces a better final decision.
2. AI visibility needs a repeatable audit
This week, several newsletters moved AEO and AI-search visibility from theory into a practical checking process. The basic idea is to test the real questions a customer might ask, record whether your brand appears, note which sources are cited and compare your presence with competitors. Google Search Console can add useful Google-side data, but it does not show the entire AI discovery picture.
This matters because a traditional ranking report cannot tell you whether ChatGPT, Gemini or another assistant recommends your brand. At the same time, one favorable prompt is not proof of broad visibility. Answers can change by platform, wording, location and date. A useful audit needs a stable prompt set and a repeatable schedule.
GrowthGary’s take: Measure AI visibility like a new form of share of voice. Start small. Use ten to twenty buyer questions across awareness, comparison and purchase intent. Record mentions, position, citations and accuracy. Then improve the pages and outside authority signals connected to the gaps you find.
What to do now: Build a simple spreadsheet with the prompt, platform, date, answer, cited source and recommended competitor. Rerun the same questions monthly. Do not chase every answer change. Look for patterns that reveal missing content, weak authority or inaccurate brand information.
3. Visual prompting is becoming a real marketing workflow
OpenAI released ChatGPT Images 2.5 with faster generation, stronger editing and a Sketch feature that turns a rough drawing into a polished image. The important change is the interface. A marketer can now communicate placement, composition and visual hierarchy by drawing an idea instead of trying to describe every detail in a long prompt.
That lowers the barrier for people who know what they want but do not speak like designers. It can make campaign concepts, landing-page visuals, social graphics and product mockups faster to explore. Better selective editing also matters because older image tools often changed the entire scene when the user wanted one small correction.
GrowthGary’s take: Faster image generation should create more testing, not more random content. Good AI design still needs a clear brand system, a useful message and human review. Sketching can shorten the distance between an idea and a usable draft, but it does not decide which visual will earn attention or support the offer.
What to do now: Take one upcoming campaign and sketch three layouts for the same message. Keep the headline, offer and brand rules fixed while changing the composition. Generate the variations, review text accuracy and brand consistency, then test the strongest two instead of publishing the first attractive image.
The simple bottom line
The best AI marketing systems are becoming more deliberate. Use premium intelligence where it earns its cost, measure AI-search presence with a repeatable method and turn faster visual creation into disciplined testing. Capability is moving quickly, but clear goals, clean measurement and human judgment still determine whether the tools produce growth.