How I think about SEO and AI search
I test SEO advice on my own sites before I trust it. Below, the six rules I work by.
- Test the advice on the site in front of you.A checklist that held pooled together and broke site by site
- Build the answer nobody else built.Three openings the big sites left, and what each earned
- SEO is GEO.What AI answers changed in the work, and what stayed put
- Change my mind when the data does.Four changes to how I work, each with its numbers
- Bring the builders in before the build.Who sees what SEO alone misses
- AI speeds up the work. The calls stay mine.What I hand off, and what never leaves my desk
All six serve the three outcomes of my SEO system:
- 1Get foundCan the person with this need find the right answer?
- 2Get understoodCan they quickly tell what the answer means for their situation?
- 3Get chosenWhat makes this answer worth trusting and acting on?
New to all of this? Start with what I do, in plain English.
Run the same test on your site
You don't need my data to check the advice, only your own. Here's the test from the top of this page in five steps, with my run beside each one.
| Step | What you need | Mine |
|---|---|---|
| 1Export every page Bing's AI answers cited | Bing Webmaster Tools, AI Performance, the pages report, one site at a time. | 1.8K pages cited at least once across my sites, through Oct 11, 2026. The export stops at 300 rows: on CheckMyTap it listed 300 of 876 and on WireRef it listed 300 of 386, so a big site's export stops at its top pages. |
| 2Count what the advice says matters | A crawl of every cited page, counting numbers, list items and tables in the main content and skipping the menu and footer. | One crawl of my cited pages, matched to the export by address. |
| 3Compare with every page pooled together | One comparison across all pages: does more of the thing go with more citations? | Pooled together, pages with more numbers looked far more likely to be cited. That's the result the checklists quote. |
| 4Run the same comparison inside each site | The same comparison, one site at a time. One big site full of number-heavy templates can make the pooled result on its own. | Site by site, that held on two sites and disappeared on three. |
| 5Decide what "held" means before you look, and save the join | A written rule (mine: the pattern has to show up inside a single site as well as across the pool), and the page-by-page table kept with its date. | The rule, yes. The table, no: I didn't save the page-level join from Aug 26, 2026, so that result can't be recounted here and no count is given. |
On my sites, steps 3 and 4 gave different answers. Run step 4 on your own site before you act on step 3.
What I look for in a new site idea
The big sites answer the obvious questions. The opening is the question people keep asking that nobody answered well. A page built for it is hard to copy, people trust it, and AI answers lean on it.
| The opening | What I built | What happened |
|---|---|---|
| Own the game-day questions the team site doesn't answerFor example, “providence park bag policy” | Title Town PDX: Game-day guides with the practical answers fans need, like bag policies and parking, instead of fighting the official team sites for schedule searches. | 1.50% click rate on the guides, against 0.46% on schedule pages. Search Console, Jun 12 to Oct 7, 2026. |
| Keep one page current when everyone else publishes onceFor example, “oregon ducks football injury report” | The Quack Report: A weekly injury report page, kept current through the season. | #4 average Bing position for those searches, against 22.5 on Google. Bing and Search Console, four weeks to Oct 8, 2026. |
| Rebuild a page around the job the reader is actually doingFor example, “how to prepare for a fan event (query paraphrased)” | An independent fan reference: A seasonal fan guide rebuilt around the reader's deadline, with the quick answer first. | 3.6x growth in clicks per day, while the rest of the site grew 1.6x. Search Console, 28 days before and after Aug 17, 2026. |
The SEO system these rules became
The three outcomes are reached through six pillars I work by on every site, each with an example from one of my sites. Why a system beats a one-off strategy: SEO strategy vs SEO systems.
- Start from the personA water page that can say no to a product
- Answer firstThe forthcoming book answer before a date existed
- Make it holdA technical mistake turned into a check
- Keep it trueThe checks that read every page before it ships, and what each one catches
- Watch it happenThe real prompts I give AI, dated
SEO is GEO
AI answers, answer engines, generative search: new names keep showing up. Underneath, it's still SEO.
The new surfaces change the priorities: clear answers near the top, the facts a machine can quote with a source and a date, a site that's plainly about one thing, a name other people mention. They don't change the job, which is being the most useful and most trustworthy result for the person asking.
In September 2026, Bing's AI answers (Copilot, Bing's AI summaries and partner sites) cited pages from my sites 634K times, by Bing Webmaster Tools' AI Performance report (data through ; of the sites I show here, WireRef has the biggest share). Which page got picked on each site, and what the winners share: the most-cited page, site by site.
How an AI answer actually reads a page: how AI search understands your content.
Others in the field Lily Ray found that ChatGPT now checks trusted sources through ordinary search before it answers, which she calls fundamentally an SEO problem (August 17, 2026). Wil Reynolds pushes back that GEO isn't a performance channel and needs its own measures (October 2, 2026). An answer can name a page and still send people elsewhere, so this site counts citations and visits separately.
"From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
What changed in how I work
Four changes, each with the data that drove it.
I track Google and Bing side by side now.
My reference sites get almost all their search traffic from Bing. The WireRef story
The checks go in before more pages do.
My water-quality catalog went from 1.6K pages in Google's index to 218, and in July the error types my audits had flagged on it became rules the build enforces. The CheckMyTap story
Nothing ships until it passes a check.
AI output that looked right went live on three reference sites. I ranked the nine ways it failed, and nothing ships now until it passes checks I wrote the rules for. The personal audit
My framework comes from what I actually do.
I sorted 2.2K of my own messages about how I work, and they didn't fit the framework I had published. I rebuilt it around six pillars drawn from those messages. The system
What I'm still figuring out
- Whether WireRef's growth on Bing would have happened without anything I did. What the data can and can't say
- Why Google dropped most of CheckMyTap's pages. I can measure the decline; I can't explain it yet. The miss, in full
- How many AI citations turn into visits. A citation isn't a click, and nobody reports the link between them well yet. How I measure it so far
- Whether six pillars hold up against another six months of real work. The system as it stands
What I'm writing now
When I bring other people in
SEO rarely wins on its own: the fix usually needs a developer, a designer, data, PR or paid search. Those people come in before anything gets built.
A fix that arrives after launch costs a ticket, a sprint and somebody's patience. The same idea in the planning meeting costs a sentence. So the plan starts from the ground up with the people who will build it.
| Who | What they see that SEO alone misses |
|---|---|
| Developers | What's actually buildable, what the templates can carry, and where a check can stop a mistake for good. They come first. |
| Design | Whether the page helps the person on their phone, in the moment they searched, before anyone reads a word. |
| Data | What to measure before launch, so "did it work" has an answer later. |
| Digital PR | The stories and original data that get a brand mentioned and linked, which search engines and AI answers both notice. |
| PPC | Which searches already convert, what people click on, and what to test before betting a whole section of the site on it. |
How I use AI in SEO work
AI is my tool for systems, data and research. It's good at systems, data and research. The creative work stays human, on purpose.
What I use AI for, to my spec
- Writing code to templates I designed
- Pulling and cleaning data
- Gathering research, which I check against the source
- Writing the check code to my spec
What stays with me
- Finding the opportunity nobody else saw
- The roadmap that actually moves the needle
- The voice, and the opinions
- Deciding what's worth building at all
- What every check has to catch, and when something is ready to ship
One job per tool
Each AI tool I use has one job, in the order the work happens. Between steps, I decide what moves forward.
-
Peec AI: set up the AI visibility project
Peec AI runs a set list of questions in AI tools like ChatGPT and Google's AI Overviews every day and records which brands the answers mention.
From Peec AI's quickstart guide, read Oct 14, 2026.
My part: I write the questions people actually ask, and I check Peec's read against Bing's own AI citation data.
-
Ahrefs: set up the site project, then dig into one question
An Ahrefs project can bring in a site's Search Console data and crawl the site on a schedule, and Brand Radar shows which pages AI answers cite and how a brand's share compares with its competitors'.
From Ahrefs on projects (Sep 3, 2026) and Ahrefs on Brand Radar (Sep 3, 2026), read Oct 14, 2026.
My part: I pick the one question worth a deep dive, and decide what the answer changes.
-
Claude and Codex: build, check, and sharpen the prompts
I run two AI tools, Claude and Codex, in separate copies of this site, and give each one the brief and the standard. Each change names what the other should verify.
My part: Their findings are leads. I check the result against the site or the source before I treat it as settled.
Automate the left-brain work well, and the right-brain work gets the time it needs. Lose the manual, creative part, and every site starts to look like every other site, which is exactly what search engines and readers have learned to skip. This site is built that way, and the checks behind it are here.