The work behind the work
How I prompt
These are real prompts I've given AI while building my sites and doing SEO, word for word, with the date and the project. They show how I direct the work: the standard I set, the checks I ask for, and the calls I make on what comes back. AI speeds the work up; the judgment stays mine.
- 39prompts
- 11projects
- 6pillars of my system
Apr 7, 2026 to Oct 12, 2026. Names, keys and client details are removed; everything else is as I typed it, typos and all.
If I could show you one
another north star i'd like to say is that SEO is GEO, and any sort of AEO or other abbreviation, at it's core, is still seo [...] AI is good at systems, data, and research, but when it comes to finding "moat" opportunities, or insightful roadmaps that move the needle, you cannot lose the manual, creative process.
GEO is still SEO. AI speeds up the systems, data and research work, and the moat ideas and the roadmap still come from my own creative process.
The library, by pillar
Show
Start from the person
Every prompt here starts with a real person on the other side of the screen and the question they actually came with.
-
in the areas where people are tryign to understand what i even do, we need to think about simplifying it so everyone can understand, whether that's my 22 year old brother, my 80 year old grandpa, or my cousin from nebraska [...] don't act like they're dumb for not knowing
I picture real people I love reading the page and ask for plain words that don't talk down to them.
-
who's going to be on this page? on a blog, could be someone from linkedin or in the industry reading, [...] I also want to make sure my failures are known, because that is a big part of who i am, we just need to position me as someone who learns from mistakes, and gets better.
Every template on this site starts from who lands on it, and I asked for my failures to show next to my wins.
Planning the rebuild of this site.
-
pretend you're a 12 year old on mobile going to play, and try to play yourself. if it isn't easy, we aren't there yet. the basic mechanics we need to get right.
I make the test a real person on a phone, because when a kid can't play it, it isn't done.
I couldn't set a lineup on my own phone, so I handed the test to an imagined kid.
-
what do they get, people are selfish, they won't do something if we don't give them a reason to. we know our fans, we know our audience, we know what they care about. be funny, entertaininig, make them comfortable doing this program and being one of the first.
Before I ask anyone to do something, I say what they get out of it.
A fan submission page explained the ask but never gave a reason to say yes.
-
you are a nfl draft duck fan. what do we not have that you need to know?
I become the reader first, then ask what's missing, which is a gap analysis in one sentence.
Tell it as a story, in a voice worth reading
These are about how a page reads: one story in order, a human voice, and no design that looks machine-made.
-
could you do a copywriting and voice audit, like you were a veteran who's studied copywriting and brand voice for 60 years in the industry, with the goal of finding ai slop, if this then that statements, and anything else that is obviously not human sounding, with the exception of our ai voice i understand that's a theme. additionally, pick out things that feel or sound unnatural in H1s, h2s, paragraphs, etc. plan mode, don't touch code
I give the reviewer a persona, a named target and one deliberate exception, and I keep it from touching code.
-
id like you to follow the patterns of this duck team early in the season and start thinking critically about what this season's theme and personality is, and we need to be incorporating overall vibe and feel to our site more. it feels too stat/y and less emotion/y
Stats alone don't tell a season's story, so I asked for its personality.
-
do it, we can have opinions, i want you to decide based on [the author] and the book text, make it geo and llm readable, do all of them, do it in a smart way, make sure you say "this is just our opinion, [...]" or that type of thing
Opinion is welcome, as long as it's grounded in the source and labeled as opinion.
Deepening thin fan-theory pages.
-
can you also do a scan to make sure if the author saw our site, he wouldn't be mad [...] but happy the community is running with his great ideas?
One human test covers tone, accuracy, spoilers and respect at once.
Be the source worth citing
Here I'm trying to be the page that people and AI answers both reach for, built so a competitor can't copy it in an afternoon.
-
and on the about page, really think through what questions people will be curious about, what they search for, how we can dominate my name searches, and how we tell LLMs exactly who i am and what i do and what i'm good at and how i provide value not just with my work, but by how i think and who i am as a person.
I treat an About page as an answer for people and for language models, built from the questions they actually ask.
-
another north star i'd like to say is that SEO is GEO, and any sort of AEO or other abbreviation, at it's core, is still seo [...] AI is good at systems, data, and research, but when it comes to finding "moat" opportunities, or insightful roadmaps that move the needle, you cannot lose the manual, creative process.
GEO is still SEO. AI speeds up the systems, data and research work, and the moat ideas and the roadmap still come from my own creative process.
-
make sure LLMS can scrape all this, if you put something behind a wall, it needs to be llm friendly. that goes for the entire site, be careful from a geo perspective. again, plan mode.
Anything tucked behind a click still has to be readable by the AI crawlers that decide who gets cited.
-
what are we missing from our site that would be awesome? [...] are we showing the colors for the home games that the fans wear? this is the stuff that will make us successful from the ESPNs of the world and increase our moat, we need to think about what diehard duck fans care about
I don't try to beat the big outlets on news; I go after the details only real fans care about.
-
if we keep it, how can we improve it? someone can just copy this right away, so how do we make a little moat? in fact, think through the moat of our whole site,
I test every feature by asking whether a competitor could copy it in an afternoon.
-
the site launched yesterday i'm not expecting to show up in these that's not the whole point, the whole point is to look at what we have and see if we WILL EVENTUALLY get cited based on the work wev'e done, and what we need to do to improve now that we have proof of what these searches look like
I read today's AI answers as a map of what will get cited, not as a scorecard for a one-day-old site.
An audit was fixating on zero citations the day after launch.
-
check that we don't have these already, if we don't, create the pages, think through GEO and how people buy and user intnetn and storytelling, think through natural bridges, internal link strategy, and topic cluster building, think through the homepage, menu, orphan pages, duplicate risk, other risks, readability, tables, and create them. go
One sentence carries my whole page checklist, starting with whether we already have the page.
Adding buyer-intent comparison pages to a water quality site.
-
I'd like you to think about how to win the LLM recommendation layer, anything else we can take advantage of?
I treat AI search as a recommendation the model makes, not just a ranking.
Never ship a wrong fact
I treat one wrong fact as a sign of more, and I make checks countable so "verified" means something.
-
only use industry research from the last like 3 months max, this industry moves so fast
I put a date limit on sources, because old SEO advice turns into wrong facts.
-
if you find pre-existing defects, and triple verifgy and confirm tghey are wrong on our site, and do not just lbvindly change , in the past youve done this but put wrong things on the site that were right that you thought were wrong
I warn against "fixing" correct content that a model only thinks is wrong.
-
fix anything that could harm someone
With limited time, reader safety goes first.
Triaging an outside audit of a fittings reference site.
-
our site is booming on bing and affiliate sales. [...] I would like you to not touch the site at all but audit the top 10 pages getting traffic and triple confirm every single word, number, calculation, formula, or naarative. this is life or death, and people need to trust us
When an electrical site starts winning, I tighten the checks, because a wrong number there can hurt someone.
Traffic was growing fast.
-
3 searches, 3 separate links. that's a requirement
I turn "verify" into something countable that anyone can check.
On an electrical reference site, where a wrong fact can hurt someone.
-
You found an error after checking just 6 pages. that means there are more errors.
One error in a sample means the whole set has errors.
Measure, then double down
I read the numbers before I trust a story, mine included, and then I put more into what's actually working.
-
i'm using peec.ai for ai visibility reports, ahrefs brand radar for citationa nd share of voice, GSC to pull csvs and analyze large qualities of data quickly with AI, ga4 for revenue and key events, bing for ai visiblity 1st party data to verify Peec and Ahrefs [...] this is a system, everything works together to bring the level up across the board. it compounds.
I use first-party data to check the third-party tools, and every source feeds the others.
-
when it comes to anything on the site from the past, i want you to pull the best elements out yes, but on a granular level, like data level, use current, most accurate data to see what is ACTUALLY working, not what i say is working, and focus on that. also focus on what isn't working.
I set the standard up front: the data wins, even over my own account of what works.
-
also research margins and best practices, and answer if our site is doing a good enough job. i don't like our width choice on desktop, for example, but if it's backed by data, fine. just prove it
I'll give up my own preference when the evidence disagrees, but the evidence has to be shown.
-
wider is way better, in fact, i'dl ike even wider
I changed my mind once I saw it, and said so.
After I'd asked for the reading width to be proven with data.
-
keep in mind it was a bye week as well
Context comes before conclusions: a dip in a bye week isn't a ranking problem.
Reading a week-over-week drop in impressions.
-
for standing problems, i hate it, this is worthless, consider dropping. the most essential updates are changes in gsc, ga4, clarity, psi, crux, bing, or daily/weekly/monthly changes that tell me where to look and investigate. i can't do anything about google indexing for example.
An alert has to point me somewhere I can act, so I cut the ones I can't.
Simplifying the alert panel on my dashboard.
-
the last 7 days have been awesome. i need to know why. Find the products that are being successful, and the urls
A good week is something to explain, so I can repeat it.
-
data doesn't add up, getting gsc clicks but nothing in ga4
I check one source against another and treat a mismatch as a bug.
It turned out to be a broken analytics tag on a new site.
Turn every rule into a mechanism
When something breaks twice I stop asking for fixes and build the check. Every finding, from a person or an AI, gets proven before it changes anything.
-
never trust agents and plugins naturally, everything needs to be checked with a fresh pair of eyes. [...] any handover documents or markdowns should lead with giving a message not to blindly run but to use as critical thinking direction and stress-tests to the current plan or mission [...] never trust each others findings blindly
Every finding gets proven against the source before I act on it, whoever it came from, and every handoff opens by saying so.
This became the "read critically" note every handoff on this site now opens with.
-
write a handoff to codex, where it isn't demands, but a thought document, to let codex explore, come up with new or better ideas, break the current plan you proposed, or find a unique angle or gap you haven't thoguht about.
I write handoffs as direction to explore and challenge, not a list of orders, because a plan should survive a real attack before I commit to it.
-
make the full plan, break the plan, make sure it's solid and we aren't repeating past mistakes, and then execute, push live, then audit the live site.
The whole loop in one line, and it ends on the live site, not on "pushed".
-
now, look at codex's review, look at yours, and make our final plan. what are we not thinking about? what could break? what is already breaking? what are we wasting our time on?
I weigh two reviews against each other and make the final call, starting with the question most reviews skip: what are we wasting time on.
The night before the season opener.
-
come up with a gate to test once claude finishes these
I decide what the work has to pass before it starts, so done means checked, not just finished.
A batch of electrical-code corrections was in progress, and I wanted the test settled before any of it shipped.
-
i just need to know when my scripts aren't 100% current, cause the gate can give green but that isn't true and i still need to backfill, for example, and it's driving me batty
A false green is worse than a red, because it stops me from looking.
-
break the plan, and what survives you do
My core loop: attack the plan while mistakes are still cheap.
-
what did you break? not think about? can you improve design? did you fabricate anything?
Four questions after every change, including a direct check for anything made up.