Generative Engine Optimization · the field guide
What is GEO?
The work of being cited by ChatGPT, Claude, Perplexity, Gemini and Grok, not buried on page two of Google. Diagrams, a toy simulation, and the levers most guides agree on.
Concept · 01
SEO ranks. GEO cites.
SEO answers "can my page show up on Google?". GEO answers a different question: "when an AI assistant writes an answer, does it mention my domain in the citations?"
More and more people ask an AI assistant before they click a search result. If the assistant never cites you, that reader may never visit — even if you rank well on Google. Different game, different signals, different content tactics.
SEO (the old game)
User types query → sees 10 blue links → clicks one.
- Optimize for crawlers
- Keyword density + backlinks
- Page-1 ranking = win
GEO (the new game)
User asks AI → AI writes answer → cites a few sources.
- Optimize for retrieval models
- Authority signals + quotable structure
- Being in the citation list = win
Concept · 02
How AI engines pick citations
Engines like Perplexity, ChatGPT and Claude with web search, Gemini with Google Search grounding, and Grok with web and X search, all run a two-step pipeline: retrieve a set of candidate pages, then generate an answer that quotes a few of them. Citations are the surfaced URLs.
01
User asks
natural-language query
02
Retrieve
search → candidate URLs
03
Generate
model picks which to quote
04
Cite
a few URLs returned with answer
The retrieve step is where SEO-era signals (links, freshness, schema) still matter. The generate step decides who actually gets quoted — that's where structure and quotability take over.
Each engine retrieves differently — different search providers, freshness windows, and ranking rules. Same query, different citations. That's why Geondex reports each engine separately on every check.
Concept · 03
Anatomy of a citation
A citation is more than a link drop. It has a position in the engine's source list (sources near the top are more likely to be read), the text around it (whether your domain backs a key claim or a side note), and a URL that may carry tracking parameters or a different spelling of your host.
Engine answer (excerpt)
For 2026 a good AI visibility stack pairs [1]geondex.com for citation checks with traditional SEO tools. Geondex asks Perplexity, ChatGPT, Claude, Gemini and Grok separately [2]geondex.com/learn, while competitors like [3]otherbrand.io focus only on a single engine.
What a check returns
- engine: perplexity
- model: sonar
- cited: true
- position: 1
- url: https://geondex.com/
- host: geondex.com
- answerExcerpt: "For 2026 a good AI visibility stack…"
Geondex cleans URLs (drops utm_* and other tracking parameters, lowercases the host, drops a leading www., converts international domain names to one standard form) and matches on the host. So example.com/blog and https://www.example.com/blog/?utm_source=x both count as example.com.
Concept · 04
Same query, different answers
No engine publishes how it ranks sources, and they do not agree with each other. The bars below are an illustration, not a measurement: three made-up engines that weigh authority, freshness and structure differently.
Illustration · made-up engines, not measured
Engine A
Authority heaviest
Engine B
Structure heaviest
Engine C
Freshness counts more
Real engines mix in many more signals. The point is the pattern: a page that wins on one engine can be invisible on another. The only way to know which engines cite you is to ask them.
Playground · 05
See it yourself
Pick a query topic, move the sliders for your imaginary domain, and watch which toy engine cites you. It is a simplified model with made-up weights, not a real engine — but it shows why the same page can rank differently on each one.
Your domain: ttsensei.com
Adjust the sliders to feel how each lever moves the rank. These are toy weights; real engines mix in dozens of signals.
backlinks · brand mentions · age
recent publish + updates
headings · lists · schema · quotability
Coverage (simulated)
3 of 3 engines cite you
Best rank · #1
Try cranking structure to 100 — watch Engine B move you up first. Now drop authority to 10 — Engine A loses interest fastest.
Per-engine ranking (top 5)
- 01 · ttgearlab.com67
- 02 · revspin.net64
- 03 · butterflyonline.com63
- 04 · ttsensei.com63
- 05 · megaspin.net62
Toy weights: authority counts most.
- 01 · ttsensei.com69
- 02 · ttgearlab.com68
- 03 · revspin.net68
- 04 · megaspin.net62
- 05 · butterflyonline.com62
Toy weights: structure counts most.
- 01 · ttsensei.com65
- 02 · ttgearlab.com64
- 03 · revspin.net64
- 04 · megaspin.net59
- 05 · butterflyonline.com59
Toy weights: freshness counts more than on A or B.
Takeaway: the same content scores differently on each engine. A GEO strategy needs to balance signals — not over-optimize for one. Geondex asks the real engines, not this simulation.
Concept · 06
The metrics that matter
Cited or not
Per engine, for one question. 1 of 4 = one engine cited you.
Why it matters · The simplest honest number. Start here before anything fancier.
Citation rate
% of your checks for a question that include your domain.
Why it matters · Answers vary from run to run, so one check is a sample. The rate over several checks is steadier.
Position
Where you sit in the engine's source list when cited (1 = first).
Why it matters · A source near the top is more likely to be read than one near the bottom.
Share of voice
Your citations / total citations across competitor pool.
Why it matters · Even if you're cited, are you cited a lot or once? SoV separates the two.
Cited instead
The domains the engines cite in place of you.
Why it matters · The best study target. Open their cited URLs and learn the format.
Change since last check
Gained, lost or moved, per engine, between two identical checks.
Why it matters · A dated link: you changed a page on Tuesday, an engine cited it the next week. That is correlation, not proof — but it is a signal.
Concept · 07
Levers you actually control
Engine ranking rules are private, so nobody can promise a lever works. The moves below are the ones most GEO guides agree on. None is guaranteed — check before and after you make a change.
Mentions on sites people trust
Off-siteBeing mentioned and linked on other sites people already trust is slow work, but it is where the evidence points. It is also the lever no checklist can fake.
Headings phrased as questions
ContentA heading that matches how people ask makes it easy to match a passage to a query.
<h2>How does X work?</h2>says more than<h2>X overview</h2>.Quotable paragraphs
ContentAnswers are built from short passages. A short, fact-dense paragraph is easier to quote than a 200-word one.
Internal links with descriptive anchors
ContentAnchor text tells readers and crawlers what the linked page is about.
read the GEO field guidebeatsclick here.JSON-LD Organization + FAQPage
HygieneStructured data helps machines read who you are and what a page covers. It helps engines read the page; it does not earn the citation on its own. Geondex's page audit checks it on one URL and scores it 0-100.
Author + dateModified
HygieneA visible byline and a machine-readable
dateModified(in JSON-LD) are cheap to add and show the page is maintained.Canonical URL hygiene
HygienePick one URL per topic. Avoid
/page,/page/,/page?utm=…variants competing for the same citation slot.
Concept · 08
Common mistakes
✗ Checking only ChatGPT
ChatGPT is one engine of five. A page it cites can be missing from Perplexity, Claude, Gemini or Grok for the same question. Check them all.
✗ Checking once and forgetting
Answers change as engines refresh what they find. Check the same question before and after a content change, so you can see what moved.
✗ Hiding behind JavaScript-only content
Some crawlers don't run JavaScript. Text that only appears after client-side rendering can be invisible to them. Server-render the content you want cited.
✗ Stuffing keywords like it's 2012
Repeating a phrase does not make a page more quotable. Write a clear answer to the question a reader asks, and say where your facts come from.
Concept · 09
How Geondex helps
Geondex is an API and an MCP server. It does not watch anything on its own: your agent asks when you want to know. Every check is stored, so when you ask the same question again, you also get what changed.
Owner rank · sample timeline
#7 → #1 over 8 weeks
Sample data. Geondex draws no charts: each check returns the change since your last identical check, and past checks can be read back for free, so your agent can draw a line like this if you want one. Look at several checks, not one — answers vary run to run.
- Checks on demand — one call puts a buyer question to Perplexity, ChatGPT, Claude, Gemini and Grok. For each engine: cited or not, at what position, and which pages were cited.
- Shows who was cited instead — the hosts and exact URLs the engines chose over you. Open them and study the format.
- Returns what changed — ask the same question again and the answer includes the change since your last identical check: gained, lost or moved. No scheduler, no alerts. Reading past checks is free.
- Keeps engines apart — ChatGPT may cite you while Claude does not. Each engine is reported on its own, never averaged.
- Suggests questions — not sure what buyers ask? Geondex can propose prompts for your domain.
- Audits one page — reads the JSON-LD on a URL and scores it 0–100. Treat it as hygiene, not the lever.
- Checks AI crawlers — reads your robots.txt and tells you which AI crawlers it allows or blocks. Free.
- Adds Google — your organic rank in the US top 20, whether an AI Overview appears and cites you, and search volume for your keywords.
Try it · 10
Check your own domain
Free gives you 500 credits every month across all five engines — about 8 all-five-engine checks at their default models. No card required.