Avakata Field Notes are short essays on agentic marketing, generative engine optimization, and what the engine has been doing this week. Some are written by Ryan. Some are written by the engine, edited by Ryan, and clearly marked.
This is the audit we run on discovery calls, converted into a timed solo exercise: 15 minutes on crawler access, 25 on extractability, 20 on identity signals, 20 asking the engines directly, and 10 building a punch list ordered by effort. Most fixes need an editor, not an engineer.
CASE Case note: the F-grade site that hit a B in fourteen daysAn HVAC contractor ran our free homepage grader and scored 31 out of 100, with sixteen of 22 checks failing. Fourteen days of fixes later the site regraded at 82, and Perplexity began citing it for local service queries. Here is the day-by-day log.
Every Monday: this week’s AI developments translated into the exact plays to run in your business. First issue free.
This is the audit we run on discovery calls, converted into a timed solo exercise: 15 minutes on crawler access, 25 on extractability, 20 on identity signals, 20 asking the engines directly, and 10 building a punch list ordered by effort. Most fixes need an editor, not an engineer.
An HVAC contractor ran our free homepage grader and scored 31 out of 100, with sixteen of 22 checks failing. Fourteen days of fixes later the site regraded at 82, and Perplexity began citing it for local service queries. Here is the day-by-day log.
If ChatGPT cannot say what your brand is, you have an entity problem, not a content problem. Here is the process we run: a five-engine baseline test, one canonical sentence, Organization schema with sameAs links, a parseable about page, and the 60-day timeline it honestly takes.
Prompt injection is not a prompt-writing problem. Every page, email, and PDF your agent reads is an unvetted dependency with a stranger as maintainer. Here is the supply-chain posture we run: mapped inputs, least-privilege tools, a quarantine airlock, and human gates on the irreversible three.
Bounce rose six points where our engine's hero animations got loud, so the system cut its own motion and palette theatrics across nine sites — and wrote the restraint into policy. A dispatch on calm as a learned behavior, with the numbers that taught it.
We spent six hours a month on a fourteen-page client report that earned 47 seconds of attention. We replaced it with a one-sentence memo — change, cause, next move — and reply rates hit 78%. Here is the formula, three real examples, and the two times a year the long report still matters.
When two pages make the same claim, an answer engine cites one and ignores the other. We ran 60 paired buyer prompts across three engines to find the tie-breakers. Four signals decided almost every head-to-head, and most are fixable in an afternoon.
A nine-person Shopify home fragrance brand turned on per-client learning June 1. Thirty days later: organic sessions up 18 percent, AI citations up from 2 to 19 a month, edit rate down from 70 to 12 percent, and the owner's content time cut from six hours a week to 50 minutes.
Publishing daily is an inventory problem, not a discipline problem. A 14-day queue of finished posts absorbs sick days, launches, and client crunches, refilled by two 90-minute batch sessions a week. Here is the full system, including the 20-minute weekly review that runs it.