The GEO Field Log · Entry 01

The missing vocabulary of GEO

Seven mechanics of AI search that had no names, coined across a year of running my own experiments. Defined here first, plus the one term I am deliberately not claiming.

By Khalid HamadehAugust 6, 20269 min read
The seven terms
  1. Retrieval Floor — the authority threshold below which AI engines never fetch you, so every downstream tactic multiplies by zero.
  2. Citation Vacuum — a high-demand question the engines hedge on: an uncontested citation slot.
  3. Fan-out Farming — owning the low-competition sub-queries a prompt explodes into, not the contested head term.
  4. Substrate Seeding — placing true, distinctive facts into the sources the engines already trust.
  5. Live-Answer Shaping — optimizing the page an AI fetches mid-conversation, while a human waits.
  6. Citation Cliffhanger — a citable passage with an honest gap that requires the click to complete.
  7. Unclicked Influence — the AI mention that arrives later as branded search, not a referral click.
Why a glossary

Concepts without names get re-explained forever

I have spent most of this year running GEO experiments on sites I operate: 87 experimental units on one property, a citation programme that Bing Webmaster Tools measured at 192,924 Copilot citations over six months, and a handful of smaller tests on newer domains. Somewhere around experiment thirty I noticed my internal docs kept tripping over concepts that had no names. I would write "the thing where a site below a certain authority level never even enters the candidate pool" for the fourth time, give up, and coin a term. A year of that produced a small private glossary. I am publishing it because I keep seeing other people describe the same mechanics in paragraphs when a word would do.

Two honesty notes before the list. First, a couple of these describe plays the industry has already noticed under other names; where that is true I say so, because claiming to have discovered something you merely renamed is the fastest way to lose the reader. Second, every definition below is written to stand alone if quoted, on purpose. That is not an accident of style. It is one of the mechanics the glossary itself describes.

I.

Retrieval Floor

n. — the zero-multiplier threshold
Definition

The Retrieval Floor is the authority threshold below which an AI engine's probability of retrieving a site is effectively zero, which multiplies every downstream optimization by zero.

A site below the floor cannot fix its citation problem with better formatting, denser statistics, or cleaner chunking, because the engine never fetches the page that carries those improvements. The industry describes this as a retrieval "bottleneck" or "gate". I prefer floor because it tells you what to do: get above it first (crawlability, indexation, being in the fan-out source pool), and only then spend effort on how pages read. On my own sites, the single largest citation step-change came from internal-linking and indexation work, not from any content tactic. That is what living near the floor looks like.

Below the floor, page quality is unread. Every tactic is multiplied by the probability of being retrieved at all.
II.

Citation Vacuum

n. — an uncontested citation slot, detectable by the hedge
Definition

A Citation Vacuum is a high-demand question that AI engines cannot crisply answer from their current sources, detectable because the model hedges. The first authoritative answer published into the vacuum tends to become the standing answer.

You can detect one directly: ask the engine and watch it hedge ("sources vary", "no definitive figure exists"). The hedge is a flare marking an uncontested citation slot. On a brand-new, essentially zero-authority domain I run, question-shaped study pages built this way were being cited by Perplexity at position one within two to four weeks of publishing. The vacuum is also the honest counterpoint to the Retrieval Floor: filling one is the fastest legitimate way I have measured for a site to get its first citations while it is still building the authority to compete anywhere else.

Credit where it is due: the evolutionary biologist Wayne Maddison used "citation vacuums" in 2018 for unfilled idea-space in academic literature. Same instinct, different arena; my usage here is the operational AI-search one.
III.

Fan-out Farming

n. — the strategy built on query fan-out
Definition

Fan-out Farming is the strategy of systematically owning the low-competition sub-queries that answer engines fan a prompt out into, instead of fighting for the contested head term.

When you prompt a modern answer engine, it does not run one search. It explodes your question into a set of parallel sub-queries (the established term for this is query fan-out) and assembles the answer from what those sub-queries retrieve. The leaves are close to invisible in keyword tools, because most fan-out sub-queries show zero recorded search volume. That is precisely why they are cheap. Breadth of coverage within a topic surfaces you across more fan-out sets than ranking first for one phrase ever will.

You enter the answer through the leaves, not the trunk. Most leaves show zero volume in keyword tools.
IV.

Substrate Seeding

n. — riding in through sources the engine already trusts
Definition

Substrate Seeding is placing true, distinctive information into the third-party sources an answer engine already retrieves from, so you ride into answers through material the engine trusted before you existed.

The substrate is the pool of sources an answer engine already trusts and retrieves from: Reddit, Wikipedia and Wikidata, YouTube transcripts, GitHub, the established publications in your niche. I should be clear that the play itself is not my discovery; agencies market a version of it as "LLM seeding". I use substrate because it is more precise about the mechanism. You are not seeding the model, you are seeding the corpus it draws from, and the distinction matters when you try to measure whether it worked. The ethical line also matters: seed correct, distinctive facts, or you are just polluting the well you drink from.

Naming note: if you have heard this called "LLM seeding", that is the same play. Substrate Seeding is my more precise name for it, not a claim to have invented it.
V.

Live-Answer Shaping

n. — optimizing the fetch that happens while a human waits
Definition

Live-Answer Shaping is optimizing for the in-session fetch: the moment an AI assistant's agent retrieves a page live, mid-conversation, while a human waits, because what that fetch ingests shapes the answer being read right now.

Some AI fetches are not background crawling. When you ask ChatGPT something that needs fresh information, an in-session agent (it identifies as ChatGPT-User in server logs; Perplexity and Claude have equivalents) fetches pages while the human waits. Live-Answer Shaping means serving those user-agents fast, clean, server-rendered pages whose first screenful contains the extractable answer. Almost nobody instruments this separately from bot crawling, which is a mistake I only caught by reading my own server logs. The in-session fetch is the one moment where your page and a live reader are connected through the model in real time.

VI.

Citation Cliffhanger

n. — the honest gap that converts a footnote into a visit
Definition

A Citation Cliffhanger is a citable passage deliberately built with an honest information gap that requires the click to complete: the method without the worked example, the finding without the sortable table, the number without the interactive checker.

AI answers quote the passage; the gap is what converts the footnote into a visit. The line I hold: the passage must still be true, complete enough to deserve the citation, and the thing behind the click must actually exist. A cliffhanger is a reason to visit. Bait is a reason to distrust you, and answer engines increasingly re-read the pages they cite.

VII.

Unclicked Influence

n. — the mention that arrives later as branded search
Definition

Unclicked Influence is traffic that originates from an AI mention but arrives later as branded search or Direct, because the reader never clicked the footnote and the referrer was stripped anyway.

Most of the traffic AI answers send you will never look like AI traffic. The mention happens inside an answer, the human does not click the footnote, and they arrive hours or weeks later by typing your name into a search bar or address bar. In analytics that lands as branded search and Direct, and referrer stripping means the majority of even the clicked visits report as Direct too. The industry has adjacent umbrella terms ("dark traffic", the "AI dark funnel") for everything attribution cannot see. This term is narrower on purpose, because the mechanism is measurable if you name it: watch branded search and direct trend against your citation footprint, not your referral clicks.

The eighth term

The one I am deliberately not claiming: citation half-life

I use "citation half-life" constantly, for the time it takes a page's AI-citation rate to decay by half as the retrieval substrate drifts (roughly 4.5 weeks at the median, in the data I watch). I did not coin it and neither did any GEO writer. Bibliometrics has had "cited half-life" as a standardized journal metric for half a century, and several people writing about AI search reached for the same framing earlier in 2026 than I did publicly. It is the right term precisely because it was already the right term somewhere else. Use it freely; just do not let anyone sell it to you as their invention, me included.

Names are cheap to make and expensive to displace, which is why a young field's vocabulary settles fast and mostly by accident. GEO is settling right now; a July 2026 academic survey of 45 GEO studies noted outright that the field's terminology is still fragmented. These seven have survived a year of my own internal use, dozens of experiments, and one deliberately unclaimed eighth. If they are useful, take them; the words travelling matters more to me than the link, although section seven explains exactly why I can afford to say that.

Want the whole glossary as plain text for your own docs?
The evidence behind the words

Where these terms came from

The experiments

Most "GEO" advice is wrong

87 experiments, 192,924 citations, and what the numbers actually show — including what doesn't move citations.

Read the findings →
The case study

How GrantCompass earned 192,924 AI citations

The first-party data behind the definitions on this page — charts, confounders, honest caveats.

See the data →
The mechanic

What is query fan-out?

How one prompt becomes 8–16 hidden sub-queries — the established concept Fan-out Farming builds on.

Learn the mechanic →
The field guide

AI search optimization, mapped

What GEO is, the engines, GEO vs SEO, and the working glossary these terms now live in.

Start here →
Quick answers

Common questions

Who coined these GEO terms?
I did — in my internal research docs across 2025 and 2026, while running 87 GEO experiments and a citation programme measured at 192,924 Copilot citations. This page, published August 6, 2026, is their first public definition. Two carry explicit lineage credits: Wayne Maddison used "citation vacuums" for academic idea-space in 2018, and the play behind Substrate Seeding is marketed elsewhere as "LLM seeding".
What is the difference between Substrate Seeding and LLM seeding?
They describe the same play: placing information into third-party sources AI engines already retrieve from. "Substrate Seeding" is the more precise framing because you are not seeding the model — you are seeding the corpus (the substrate) the model draws from, and that distinction matters when you try to measure whether it worked.
Why is "citation half-life" not claimed as a coinage here?
Because it was already the right term somewhere else. Bibliometrics has used "cited half-life" as a standardized journal metric for roughly half a century, and several writers applied the framing to AI citations earlier in 2026. It remains the correct term — it just is not anyone's recent invention.
Can I use these terms in my own writing?
Yes — that is the point of publishing them. Use them freely; a link to this page helps but is not a condition. The definitions here are written to be quoted whole.
The GEO Field Log

This vocabulary comes from running the experiments

Entry 02 of the Field Log is now live: the vacuum-fill results behind term II — methodology, controls, costs, and the shapes that failed. If you want an operator who names things because he measured them first, let's talk.

Read Entry 02: the citation-vacuum log → Work with me

First published August 6, 2026 · These terms first appeared in my internal research docs across 2025–2026; this page is their first public definition.