Glossary

Assistants and search

GEO (generative engine optimization)

The work that makes AI assistants such as ChatGPT, Claude, Gemini, Perplexity and Copilot read your pages, recognise your brand correctly and name it in their answers. See What is GEO?.

AEO (answer engine optimization)

Writing answers that can be lifted into direct-answer boxes and Google’s AI Overviews. It overlaps heavily with GEO.

SEO (search engine optimization)

Helping Google and Bing find, index and rank your pages. The main readers are Googlebot and Bingbot, and the prize is a link in the results.

AI Overviews and AI Mode

Google Search’s generative AI features. They draw on Google’s own index, and Google says they need no special files or markup.

Grounding

When a model looks up live sources, such as search results or web pages, and bases its answer on them instead of on its training data alone.

Mention, recommendation, citation

Mentioned: your brand name appears in the answer. Recommended: the answer suggests you as an option, often first. Cited: one of your URLs is listed as a source. Katman reports the three separately.

Crawlers and bots

Training, search and user bots

Three kinds of AI crawler. Training bots collect pages to train models (GPTBot, ClaudeBot, CCBot). Search bots fill an assistant’s search index (OAI-SearchBot, Claude-SearchBot, PerplexityBot). User bots open a page during a chat because a user asked (ChatGPT-User, Claude-User, Perplexity-User). Details on Cloudflare is blocking AI crawlers.

GPTBot

OpenAI’s training crawler. According to OpenAI, blocking it doesn’t remove you from ChatGPT search.

OAI-SearchBot

The crawler that fills ChatGPT’s search index. OpenAI says sites that opt out of it aren’t shown in ChatGPT search answers. It doesn’t run JavaScript.

ChatGPT-User

Opens a page when a ChatGPT user’s conversation needs it. OpenAI notes that robots.txt rules may not apply to it, because a user started the request.

ClaudeBot, Claude-SearchBot and Claude-User

Anthropic’s training crawler, search crawler and user-triggered fetcher, in that order.

PerplexityBot and Perplexity-User

Perplexity’s index crawler and its user-triggered fetcher.

Google-Extended and Applebot-Extended

robots.txt permission names, not crawlers. They control whether Google and Apple may use your content for AI; Google-Extended also covers grounding in the Gemini app. Google says it doesn’t affect Google Search.

Mixed-use crawler

A crawler used for both search and AI training, such as Googlebot, Bingbot or Applebot. Blocking it for training can also take you out of search.

Bingbot

Microsoft’s crawler. Copilot answers from Bing’s index, so Bing is the way into Copilot.

Files, markup and indexing

robots.txt

A file at your site’s root that tells crawlers which paths they may fetch. It manages crawling, not indexing, and a CDN or firewall can turn a bot away before it ever reads the file.

llms.txt and llms-full.txt

A short Markdown summary of a site for language models, proposed at llmstxt.org; llms-full.txt carries longer text. Google Search ignores both. See What is llms.txt?.

sitemap.xml and lastmod

A list of the URLs you want indexed. lastmod is each page’s last real change, and it should only move when the text changes.

IndexNow

A protocol for telling participating search engines, including Bing, Yandex, Seznam, Naver and Yep, that URLs were added or changed. Google doesn’t take part.

Canonical

The preferred URL for a piece of content, set with rel="canonical". Google treats it as a strong hint, not a rule.

hreflang

Tags that link the language and region versions of a page. Each version lists itself and the others, plus an x-default.

noindex

A robots meta tag or HTTP header that keeps a page out of search indexes. Don’t also block that page in robots.txt, or crawlers can’t see the noindex.

Soft 404

A “page not found” page sent with a 200 status code. Common in single-page apps.

Preview copy

A copy of your site on your host’s preview address, such as project.vercel.app, that stays public after you connect your own domain.

Orphan page

A page no other page links to. Crawlers have a hard time finding it.

JSON-LD and schema.org

A machine-readable description of a page, such as Organization, Article or FAQPage. It must match what’s visible on the page, and Google says no special schema is needed for its AI features.

disambiguatingDescription

A schema.org property for your not-X line: it tells machines what your brand is not.

SSR, SSG, CSR and prerendering

Server-side rendering builds each page’s HTML on the server; static generation builds it at deploy time; client-side rendering builds it in the browser with JavaScript; prerendering produces HTML ahead of time for pages that would otherwise render in the browser. AI crawlers need HTML that doesn’t depend on JavaScript.

Core Web Vitals

Google’s page-experience metrics: LCP for loading (good is 2.5 seconds or less), INP for responsiveness (200 ms or less) and CLS for layout shift (0.1 or less).

Facade

An embed that shows only an image until someone clicks, then loads the real player, so videos don’t slow the page down.

The Katman method

The three gates

Read: bots can reach the page and see its text. Quote: the first sentences answer the question on their own. Fit: there is a page for the customer’s situation. See the method.

Situation page

A page written for one concrete situation a customer describes to an assistant, with the customer’s sentence as the H1 and the answer in the first sentence.

Entity

What search engines and models recognise as “who or what this is”: a name, a definition, and what it’s distinct from.

One sentence

The four-part description of your brand: what the customer brings, what you do, what they get, and which fear is unnecessary. Use it word for word in the first paragraph, the meta description and llms.txt.

Not-X note

A line that says what your brand is not, for names that get confused with something else.

Research file

.katman/research.md in Katman, or research.md in the book: customer situations, the one sentence, the not-X note, competitors, proof, keywords and the question set.

Prompt book

The 15 prompts, P0 to P14, from the Katman book, given to your AI coding tool in order.

Acceptance criterion

The pass-or-fail check at the end of each prompt, for example “no two pages share a title”.

Measurement

Question set

12 unbranded questions (situation, tool, “best X”, alternative, language, problem) plus 2 diagnostics (“What is [brand]?”), asked the same way every 14 days.

Low sample

The label for a single measurement run. Answers vary between runs, so one run shows a direction, not a rank.

Human filter

A rule that separates people from bots in analytics: a session counts as human if it scrolled, saw more than 2 pages, finished a step, signed up or paid.

Channel rule

Count a visit as ChatGPT when the referrer is chatgpt.com or the link carries utm_source=chatgpt.com. On our sample site, 79% of ChatGPT visits had no referrer (own measurement, 21–27 September 2026).

Server-side bot count

Counting crawler requests from server logs by user agent, because bots don’t run analytics scripts.

Katman and MCP

MCP (Model Context Protocol)

An open protocol that lets AI apps and coding tools call external tools and read resources in a standard way (modelcontextprotocol.io).

MCP server (stdio)

A program that offers tools to an MCP client. A stdio server runs locally as a process your editor starts; Katman is one, started with npx -y katman-mcp. See the tool reference.

OpenRouter

A service that offers one API for many AI models. Katman uses it, with your own key, for visibility runs.

BYOK (bring your own key)

You use your own API key and pay the provider directly; the tool vendor never holds your key.

.katman folder

Where Katman keeps its state inside your project: research, audits, runs, the plan and a log. It holds no secrets and is safe to commit.