The Technical GEO Checklist: llms.txt, Schema, and Being Machine-Quotable

Most GEO advice is content advice, and content does carry the weight. But I keep auditing sites whose content deserved citations and whose technical layer quietly disqualified them: AI crawlers blocked by a default security setting, schema absent, brand described three different ways in three places. This is the technical checklist I run, the same one behind my own site and my clients' sites.

Layer 1: Decide which AI crawlers get in

Each AI platform crawls with its own token, and your robots.txt controls each one separately. The ones that matter in mid-2026:

• GPTBot and OAI-SearchBot (OpenAI): training data and ChatGPT search results respectively.
• ClaudeBot (Anthropic): Claude's crawling.
• PerplexityBot (Perplexity): its answer engine's index.
• Google-Extended (Google): Gemini training. Note that Google's AI Overviews use regular Googlebot, so you cannot block AI Overviews without leaving search entirely.
• Bingbot feeds Microsoft Copilot alongside Bing search.

Check what you currently allow: open yourdomain.com/robots.txt and search for these names. Also check your CDN or firewall (some block AI crawlers by default at the network level, invisibly to robots.txt). Blocking is a legitimate business choice for publishers; for brands that want AI visibility it is self-sabotage. Make it a decision, not an inherited default. If you want to be recommended, the crawlers need to read you; that is the mechanical half of getting named by ChatGPT.

Layer 2: Add llms.txt, with honest expectations

llms.txt is a plain-text file at your site root that gives language models a curated map: what your site is, what you offer, and links to your most important pages in a format models parse easily. This site has one at rankedbyshai.com/llms.txt.

The honest part: llms.txt is an emerging convention, proposed in 2024, and adoption by the engines is partial and undocumented. I add it because it costs twenty minutes, cannot hurt, states your identity in your own words, and positions you for whatever the standard becomes. I do not promise clients it moves metrics today. Format: a markdown-style file with your name, a one-paragraph description, then sections linking your key pages with one-line summaries.

Layer 3: Ship the schema that models actually use

Structured data is how you state facts about your business unambiguously. For GEO purposes, priority order:

• Organization / Person: who you are, your logo, your official profiles (sameAs links). This anchors your entity.
• Product with offers, aggregateRating where real: the facts a shopping answer needs.
• FAQPage: literal question-answer pairs, the most liftable format that exists.
• Article with author and dates: freshness and authorship signals.

Validate with Google's Rich Results Test after every theme or plugin update; schema breaks silently. Full walkthrough in Google's structured data docs.

Layer 4: Keep your entity consistent everywhere

Models triangulate what you are from every mention of you they can find. The checklist:

• Same brand name spelling, same one-line description, same category wording on your site, Google Business Profile, LinkedIn, social bios, and directories.
• An About page that states plainly who runs the business, where it is, and what it does. Models quote About pages constantly.
• If people or publications describe you inconsistently ("marketing agency" vs "SEO consultant"), fix the sources you control and request edits where you can.

Layer 5: Structure pages into liftable blocks

The last technical mile is formatting. Engines lift self-contained blocks, so build pages out of them: a question as the heading, a 40-to-60-word direct answer immediately below, then depth. One idea per paragraph. Real lists for real enumerations, tables for comparisons. If a paragraph only makes sense with the three paragraphs above it, it will never be quoted alone.

This overlaps completely with answer engine optimization, which is the point: AEO and GEO share a spine. Structure serves both.

What technical GEO cannot do

The disclaimer that belongs in every GEO article: none of this guarantees a citation. The engines choose, their behavior shifts between model versions, and a perfectly optimized page can be ignored while a messy one gets quoted. What the technical layer controls is eligibility. My clients' AI answer appearances, shown anonymized in my case studies, all sit on this exact foundation, and none of them were promised in advance. Anyone who guarantees you AI citations is charging you for weather control.