Happy Tuesday!
This week, we are going super tactical with a screenshare session with Jim Wrubel, founder of Spyglasses.io. Jim came to the AEO space with fresh eyes, immediately spotting gaps some of us have been ignoring.
We covered brand auditing in ChatGPT, why your API-based tracking might be misleading you, and a framework called Category Entry Points that simplifies prompt monitoring dramatically. Here are 5 takeaways:
1/ Stop Tracking Every Prompt. Use Category Entry Points Instead.
Jim introduced a concept from traditional marketing called Category Entry Points (CEPs). These are the mental starting points a customer has when they first decide to engage with your product category.
Think “best catering service in Miami” or “what home fitness equipment should I get.” Not a specific keyword. The headspace someone is in when they wake up and decide today’s the day they solve a problem.
Here’s the powerful part: one well-chosen CEP actually represents dozens of prompt variations. You don’t need to track 20 different wordings. Track the one that maps to the entry point:
This is how Jolly approaches earned linkbuilding too. We target the core topics that matter, not every long-tail variation.
2/ The API vs. Web App Gap (Your Data Might Be Wrong)
This one stung because we lived it.
At Jolly, we built a Looker dashboard to monitor client visibility in ChatGPT. We were prompting through the API. Seemed like the right call: it’s neutral, every request is a fresh start, no personalization leaking in.
The problem? ChatGPT the web app has a massive system prompt running behind the scenes. Safety guardrails, context routing, cost controls, local detection, shopping channels. None of that exists in the API.
Jim showed this live. Same prompt, side by side. The API returned 4 results. The web app returned 8, with links, a map, and completely different recommendations:
If you’re auditing through the API, you’re not seeing what your customers see.
3/ Use a Scraper, Not the API
Jim’s recommendation for anyone serious about accuracy: use a service like DataForSEO.
They have machines set up that actually run ChatGPT web app sessions on your behalf. It’s not calling the API. It’s literally opening ChatGPT, typing the prompt, and recording the output. Like the old Mechanical Turk concept, but automated:
Cost is pennies per request. Top up a hundred bucks and you can run a thorough brand audit that reflects what real users actually experience.
Whether you use Spyglasses (which sits on top of Data For SEO) or build your own dashboard, the point is the same: get off the API for consumer-facing audits.
4/ Build Your Brand Snapshot First
Before you audit anything, Jim recommends building a brand snapshot: company name, tagline, pricing tier, category, audience, features, problems solved, differentiators, competitors.
Half an hour of work for anyone who knows the brand. From there, you assemble CEPs using a simple formula: “best [category] for [audience/feature/use case].”
It’s like Mad Libs for prompts. And it gives you a framework that scales without drowning in prompt variations.
(If you caught the newsletter from two weeks ago #72 with Dixon Jones, his Waikay tool takes a complementary approach, building knowledge graphs to score how accurately LLMs represent your brand.)
5/ ChatGPT Is Just a Thin Layer on Traditional Search
Jim closed with the most actionable takeaway of the episode.
When ChatGPT doesn’t have answers in its training data, it searches the web. Those are called grounding searches. The best way to show up in ChatGPT is to identify those grounding queries and build content that ranks for them.
This is where traditional SEO and AEO converge. The branded backlinks and content you’re building for Google are the exact same signals ChatGPT uses when it goes looking for answers.
If you can do that, Jim says you’re ahead of 95% of the industry right now.
Until next week,
Greg
P.S. If you want to see our updated AI Visibility report, fill out this form and we’ll send one right over!