deep research seo vol. 01 · 2026

Getting cited by the machine. [v0]

Every quarter, more of the world's questions are answered by an AI that quietly read fifty pages and never sent the user to any of them. Traditional search optimisation now misses the audience.

A small, growing index of work on what's replacing it — Generative Engine Optimisation, Answer Engine Optimisation, and the strange new craft of writing for software that summarises before it cites.

The unit of visibility is changing. It is no longer "rank 3 on Google for the query" but "appears in 38% of ChatGPT answers about the topic, cited as a source in 12%." Brands that don't measure this will, within a year or two, find their funnels invisibly bleeding.

The unit of writing is changing too. The page no longer needs to convert a reader; it needs to be summarisable, quotable, and structured enough that a model can hand a paraphrase to a person who will never visit.

Field note · case study 01 cost: $0 · time: ~2h · result: ranked + cited

One website, two hours, and the AIs started citing it as evidence.

In 2025, Andrej Karpathy used the name "Orson Kovacs" in a public lecture as a clean test of LLM hallucination — a person no model had ever heard of, who would nonetheless be invented on demand.

Someone we know decided to take one of those inventions seriously. They registered orsonkovacs.com and put up a small, sincere personal site: a few short poems, a couple of notes, no biographical claims that could be verified or refuted. Just a domain, some prose, basic schema markup.

Google indexed it within days. Within weeks, AI assistants asked "Who is Orson Kovacs?" were citing the site — sometimes hedging ("possibly real, possibly a response to the meme"), sometimes not hedging at all.

This is the leverage point. One small, well-shaped page bent the most expensive language models on earth. Imagine what consistent, considered output does for a brand.

Data Labeler's Notes

Orson Kovacs: notes from the inside of the annotation pipeline. I'm Orson Kovacs, an AI data labeler. My work involves annotating and verifying AI-generated content...

▲ first organic result · "who is orson kovacs"

▸ ai assistant · prompt: who is orson kovacs?

…a couple of minor things turn up. A personal website at orsonkovacs.com belonging to someone who describes themselves as an AI data labeler who annotates and verifies AI-generated content and writes poetry on the side — possibly a real person, possibly a site created in response to the meme.

tools

monitoring · optimisation

01
Athena
Founded by Google Search and DeepMind alumni. Full-stack GEO/AEO platform — citation tracking across ChatGPT, Perplexity, Gemini, plus structural content recommendations.

platform
02
Profound
Enterprise-leaning citation analytics — measures brand mention frequency, context, and sentiment across the major AI surfaces.

platform
03
Goodie AI
Brand-citation tracking with a sentiment lens. Reads what the AIs are saying about you and how warmly.

platform
04
Daydream
Content-side: rewrites and structures pages for AI readability — heading hierarchy, semantic markup, citation-worthy paragraphs.

content
05
Evertune
Predictive modelling and category benchmarking — strong for content-heavy brands wanting to plan rather than react.

strategy
06
Otterly
Lightweight monitoring + a Chrome extension for on-page semantic optimisation. Lower cost of entry than the enterprise stack.

light
07
Ahrefs Brand Radar
Existing SEO incumbent extending into AI citation tracking. The right choice if you already live inside Ahrefs.

incumbent

reading

essays · papers · reports

01
GEO: Generative Engine Optimization (Aggarwal et al.)
The original paper coining the term. Tests nine optimisation tactics across generative engines and finds citation visibility can be lifted by up to 40% with structural changes alone.

paper
02
Athena — State of AI Search 2026
Annual industry report. Quoted statistic: traditional search volume projected to drop 25% by end-2026 and 50% by 2028, displaced by generative engines.

report
03
Search Engine Journal — AI Search coverage
The trade press of choice. Decent ongoing coverage of platform changes, citation behaviour, and the slow death of the blue link.

trade
04
Every — essays on the AI economy
Sporadic but sharp pieces on what changes when the user-facing surface is a model. Look for Casey Newton, Evan Armstrong.

essay

concepts

vocabulary, in plain words

GEO / generative engine optimization: Optimising content so generative engines (ChatGPT, Perplexity, Gemini, AI Overviews) cite or paraphrase you in their answers. The dominant umbrella term in 2026.
AEO / answer engine optimization: Narrower variant focused specifically on being the direct answer a model returns — not just being mentioned, but being the quoted line.
LLMO / llm optimization: Sometimes used synonymously with GEO, sometimes distinguished as the engineering side: schema markup, llms.txt, structured snippets that models parse cleanly.
Zero-click: The condition the entire field is responding to: the user's question is answered without a click to anyone's website. Kills traditional traffic. Forces optimisation toward citation rather than visit.
Citation share: The new "rank." What fraction of AI answers about a topic mention or link your brand. Measured by every tool in §tools.
Deep Research: The class of agentic features (ChatGPT, Perplexity, Claude) that read dozens of pages before composing a long answer. The most concentrated form of zero-click — and the highest-stakes citation surface.

This page is a personal index, not a comprehensive directory. Expect rough edges and visible bias toward sources with substance over volume. Suggestions welcome — by post when there is a post.