In the same BrightonSEO session as my local AEO talk, Jeremy MacDonald of We Are Pioneers made the case that AI moved the bottleneck. Producing output used to take hours. Now you can produce 1,000 outputs where you once produced five, but checking them is harder than ever. His line: testing isn’t the tax on scale, it’s what makes scale possible.
I covered his talk in a six-part series, from what a moon rocket teaches AI teams to the RISE eval framework and the self-improving feedback loop. This post connects the two talks: what agentic evaluation looks like for local AI search.
Applying evals to local AEO
This isn’t something Jeremy presented for local search; it’s our adaptation. But it maps cleanly onto the local tactics I’ve been writing about. Here are evals a local AEO team could run, each scored on a simple scale:
| Asset | Eval criteria | Why it matters |
|---|---|---|
| Deep service page | Covers every specialty named in reviews; specific to the market; no near-duplicate sections | Guards against thin-page risk (go deep, not wide) |
| Pricing page and schema | Prices match across page, schema and listings; scope stated | Pricing is machine-read (publish your prices) |
| Review-request copy | Prompts customers to name the service, not just rate | Review text carries topical signal (the two rulebooks) |
| Reddit reply drafts | Discloses affiliation, answers the question, follows subreddit rules | Keeps participation honest (earn it, don’t farm it) |
| Brand description across AI platforms | Accuracy, sentiment, consistency across models | Early warning for negative GEO (black-hat GEO) |
Where the evals sit in the workflow
The monitoring step is the same observe, diagnose, act loop from tracking AI search isn’t a strategy. The difference is that the eval catches problems before they go live, not after a competitor wins the shortlist.
The shift underneath
Jeremy called the emerging role the marketing engineer: someone who builds the system that does the work instead of doing it. His closing point applies to local AEO too. The goal isn’t to do the old job faster. It’s to do work that wasn’t possible before, and to test it well enough that you can trust it at scale. And remember to check the grader, too.
Want help building quality checks into your local AI search work? Get in touch.
Leave a Reply