“GEO” gets sold like a new religion. Strip the slides and you’re left with something older: make pages easy to fetch, easy to understand, and easy to trust — then decide, deliberately, what AI crawlers are allowed to do.
That’s the version I care about.
Don’t mash three problems into one
| You’re worrying about… | Different job |
|---|---|
| Blue-link SEO | Rankings / clicks in classic search |
| Answers & overviews | Getting cited or selected inside generated answers |
| Training crawls | Whether trainer bots may fetch you for corpora |
GEO, as people mean it in 2026, is mostly the middle row. Training policy is robots + legal. Confusing them is how you Disallow: / the whole site and then ask why nobody cites you.
What actually helps citability
No peer-reviewed “do this meta tag, win ChatGPT” recipe. The boring patterns show up again and again:
- Fetchable (robots isn’t accidentally blocking you)
- Answer up top, nuance after
- Specifics: numbers, dates, constraints, failure modes
- Real publisher identity — About, contact, dates you didn’t fake
- Same facts on the page and in any schema you emit
- A stable canonical URL
If that sounds like good documentation, yes. Draft pass: GEO Content Checker. Site-wide: audit.
Skip cloaking a different body to “AI user agents.” That’s the kind of clever that ages into a penalty story.
llms.txt — useful when it’s honest
The convention is documented at llmstxt.org. Idea in one line: put a Markdown brief at /llms.txt so tools that care can learn what your site is and which URLs matter.
It is not something Google has announced as a ranking factor. Some assistants and scrapers look for it; many ignore it. Still worth writing — mostly because it forces you to describe the product without hallucinating features.
What I put in ours (theseosoul.com/llms.txt):
- One sentence definition a stranger would accept
- Canonical links (home, tools, guides, contact)
- Explicit “what we don’t do”
- A monitored email
- A short list of guides worth citing — not every URL ever
What I delete on sight in other people’s files: slogan salad, “#1 platform” cosplay, affiliate dumps, outdated product names.
Optional /llms-full.txt only if you’ll maintain it. Stale full dumps are worse than a tight root file.
AI crawlers and robots.txt
Product/legal decision first. Robots second.
Google documents its crawlers — including Google-Extended — in the common crawlers list. Their note is worth reading before you paste a viral “block all AI” robots snippet:
Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search.
It does control whether crawled content may be used for certain Gemini / Vertex training and grounding cases (details on that same page). Search crawling and that control are different levers — don’t confuse them.
Other agents people name explicitly (policies change — verify current docs when you edit robots): GPTBot, Anthropic’s crawlers, PerplexityBot, Bytespider, etc.
Blocking trainers can be the right call. Blocking everything still nukes classic SEO. See what’s live today: Robots.txt Checker.
Also: robots.txt is not DRM. Cooperative bots honor it; hostile scrapers don’t. It’s still the right place to publish intent.
A week of work, not a quarter of theater
Mon — Crawl/index basics. GEO on a site that’s accidentally noindex is cosplay. Use the launch checklist if you’re mid-migration.
Tue–Wed — Ten money pages. First screen answers the query in two sentences, then evidence. Kill fluff intros.
Thu — About, contact, Organization schema, honest updated dates.
Fri — Write llms.txt. Link only URLs you’d be proud to see quoted.
Whenever you’re ready — Document AI crawler policy in the repo; deploy robots on purpose, not from a tweet.
Re-audit quarterly. Models and crawlers move faster than most CMS themes.
Where we fit
We stuck GEO-ish checks next to meta and technical signals because that’s how the work shows up in real audits — not as a separate “AI traffic” chart. We won’t invent one.
Strategy is this page. Inspection is a free audit. Primary references: llmstxt.org, Google crawler docs, and ordinary Search Central crawl/index material linked above.