The nine moves that actually earn citations from ChatGPT, Perplexity, and Google's AI — pulled from the 87 experiments that survived my testing. Tick them off as you go.
By Khalid HamadehUpdated June 20268 min read + checklist
Answer first — put the direct answer in the first 40–60 words of every section.
Publish original data — at least one number nobody else has.
Write to a specific person — "If you're a [persona]…"
Layer the depth — quick answer → detail → deep dive.
Make every passage liftable — self-contained ~150-word chunks; names, not pronouns.
State the verdict — plain recommendations, not hedges.
Cover the whole question — answer the follow-ups on the same page.
Earn off-site mentions — get cited where the models read.
Keep it fresh — citations decay; re-publish on a cycle.
Skip: schema as a citation lever, sentence-engineering, keyword-stuffed URLs, llms.txt, and assuming one engine's win carries to the rest.
If you've read the companion piece, you know most GEO advice is noise. So here's the signal: the short list of moves that actually moved citations across 87 controlled experiments — in rough order of impact.
None of this is exotic. It's mostly about structure and substance: writing so an AI engine can lift a clean, correct answer out of your page without the surrounding context. Work through the nine below — the checklist keeps score as you go — then go earn the mentions that compound them.
I call these the Citable-Page moves — the structural changes that make a page easy for an AI engine to lift a correct answer from. They're not theory; they're what survived 87 controlled experiments on a live site.
87
Experiments run
7
Promoted
4
Retired
How it works
How an AI engine decides what to cite
01
Your prompt
One question goes in.
→
02
Fan-out
Expanded into 8–16 sub-queries.
→
03
Retrieval
Pages pulled by keyword + meaning.
→
04
Re-rank
The best passages float up.
→
05
Cited
A few make the answer.
The catch: only about 15% of the pages an engine retrieves actually get cited. The nine moves are about winning that last gate — being the passage it picks, not just one it finds.
Show, don't tell
The same fact, invisible vs. citable
✕ Invisible to AI
"There are various funding options available to small businesses in Canada, and eligibility can depend on a number of different factors. It's worth exploring what might be a good fit for your situation."
✓ Citable
A Canadian small business can access three funding types: grants, tax credits, and forgivable loans. The federal Canada Job Grant covers up to 83% of training costs, to a maximum of $10,000 per employee. Eligibility is based on incorporation status, not revenue.
Same topic. The right column leads with the answer, names specifics, and hands the engine a number it can lift verbatim — moves 1, 2, 5 and 6 at once.
The data
The on-page tactics, by measured lift
Persona-addressed sections
+623%
Micro-comparison tables
+623%
Conversational calibration
+260%
Progressive disclosure
+260%
Geographic entity density
+219%
Eligibility decision trees
+217%
Verdict statements
+125%
Median citation change vs. the site baseline, from my tests. (Publishing original data was the single biggest one-page win: 294 → 1,931.) Read these as the shape of the program — 76 of 87 tests shared a page, so treat them as directional, not isolated effects.
And skip these — I tested them so you don't have to
Schema markup as a citation lever. No causal lift across 1,885 pages. Keep it as basic hygiene; don't expect it to win citations.
Engineering individual sentences to be "quotable." The most GEO-coded tactic I built — and a clean failure.
Keyword-stuffed URLs and padding for length. Clean URLs and tight pages out-cite long, stuffed ones.
llms.txt. Have one if you like; it has no measurable effect on visibility.
Assuming a win on one engine carries to the others. Tactics that help Google's AI can hurt ChatGPT. Optimize per engine.
Rewards: original data, clean structure, and an off-site footprint (Reddit, reviews, roundups) it can corroborate against.
Watch: FAQ-formatting can hurt here; length matters less than people claim.
Perplexity
Rewards: fresh, well-sourced pages with clear, attributable citations — it leans hard on retrievable sources.
Watch: its cited set barely overlaps with Google's rankings.
Google AI Overviews
Rewards: concise, answer-first pages; question-style content can help here. About 53% of cited pages are under 1,000 words.
Watch: in testing, schema gave a small negative signal — don't lean on it.
Microsoft Copilot / Bing
Rewards: being indexed in Bing plus clear, factual passages — this is where one of my sites earned 135,700 citations.
Watch: Bing's AI is growing fast on its own; ride the wave, don't credit only yourself.
Do the nine. Skip the five. Then the only thing left is the part you can't fake: being genuinely the best answer on the page.
Try it yourself
Check your own citations in 2 minutes
Paste these into ChatGPT or Perplexity (swap in your category and brand). If you're not in the answer, you've got work to do.
best [your category] for [your audience]
what is [your brand], and who is it for?
who are the top [your category] providers in 2026?
From checklist to real number
The list tells you what to do. A measurement tells you if it worked.
A single "did ChatGPT cite me?" check is wrong about one time in nine — answers shift on every run. The honest way to track this is your share of answers across many runs and every engine, over time. That's what I built LumenGEO to do.
Shaping your content so generative AI engines — ChatGPT, Gemini, Perplexity, Google's AI — cite it in their answers. Distinct from ranking blue links in classic search.
Answer Engine Optimization (AEO)
Often used interchangeably with GEO. It emphasizes becoming the direct answer an engine returns, not just a listed source.
AI Overviews
Google's AI-generated answer box at the top of the results page, which summarizes and cites sources instead of only linking them.
Share of answers
The percentage of repeated AI queries in which your brand appears. Because answers shift run to run, this is the reliable alternative to a single "did it cite me?" check.
Citation half-life
How long an AI citation lasts before it's replaced. The observed median is about 4.5 weeks — which is why GEO is a continuous program, not a one-off.
Query fan-out
How one user prompt is expanded into many sub-queries the engine retrieves against. It's why covering the whole question on a single page matters.
Retrieval vs. citation
Being retrieved (found) is necessary but not sufficient — only a fraction of retrieved pages are actually cited in the final answer.
Quick answers
Common questions
How do I get my site cited by ChatGPT?
Lead each section with the answer in the first 40–60 words, publish original data, write self-contained passages that read out of context, and earn mentions on the third-party sites the model reads — then keep the content fresh, because citations decay.
How long does it take to show up in AI search?
Individual citations can appear within days once a page is indexed and retrievable, but durable visibility is continuous — citations have a median half-life of about 4.5 weeks, so you have to keep publishing and earning mentions.
Do I need schema markup to rank in AI search?
No. A controlled study of 1,885 pages found schema markup produced no positive lift in AI citations. Keep it as basic hygiene, but spend your effort on answer-first content, original data, and off-site mentions.
Does GEO replace SEO?
No — they overlap but they've diverged. The link between Google's top-10 results and what AI cites has fallen to roughly 38%. Keep doing SEO, but don't assume your rankings carry over to AI answers; optimize for both.
What is answer engine optimization (AEO)?
AEO is a near-synonym for GEO. It frames the goal as becoming the direct answer an AI engine returns, rather than a link the user has to click. The tactics are the same nine moves above.
How do I optimize for Perplexity specifically?
Publish current, well-sourced pages with clear citations, and get referenced on third-party sites. Perplexity leans heavily on retrievable, attributable sources — so source quality and off-site mentions matter more than on most engines.
Do I need to allow AI crawlers?
Yes. Check that your robots.txt — and any CDN or edge rules — don't block retrieval bots like OAI-SearchBot, PerplexityBot, and Google-Extended. Many sites are invisible to AI engines without realizing they block the crawlers that feed them.
Living document
Changelog
v1.1June 2026
Added the citation pipeline, a measured-impact chart, a per-engine breakdown, a GEO glossary, and the interactive readiness ring. This page is itself a GEO experiment — it'll keep changing.
v1.0June 2026
Published the nine-move Citable-Page playbook.
Part 3 is coming.
The next field note covers the off-site 85% — earning the mentions that compound everything above. Want it when it drops?
✓ You're on the list — I'll send part 3 when it drops.
No spam, no list-rental — just the next playbook.
Khalid Hamadeh
I'm a growth lead for Invoice Simple and Joist at EverCommerce, a 2× founder, and ex-Meta. Over 11 years I've scaled DTC, SaaS, and subscription businesses — and built the tools, like LumenGEO, that measure them. I'm talking to early-stage teams about growth leadership and fractional work.
The 87-experiment dataset behind these moves (original data, retired tactics, platform splits). Khalid Hamadeh: "Most GEO advice is wrong." link
Schema markup shows no positive citation lift (difference-in-differences, 1,885 pages). Ahrefs (May 2026). link
Brand mentions predict AI visibility ~3× more than backlinks (0.66 vs 0.22, 75k brands). Ahrefs (May 2025). link
Median AI-citation half-life ≈ 4.5 weeks (3.5M citation events). Scrunch × Stacker (Mar 2026). link
The "1-in-9" instability of AI answers & share-of-answers method. First-party research — Khalid Hamadeh / LumenGEO. link
The experiment results and citation counts are first-party (GrantCompass, via Bing Webmaster Tools and GA4). External figures are vendor-sourced and best treated as directional.
