Does Generative Engine Optimisation (GEO) and Answer Engine Optimisation (AEO) Actually Work? A Live Experiment on Two Guernsey Pages

A documented test of GEO and AEO principles on a brand-new page and a retrofitted existing page, with baseline data, honest results, and the parts that didn’t work.

By Martin Loader, Guernsey Donkey SEO· Published 13th September 2026 · Last updated 13th September 2026

Update — 13 September 2026: Gemini has now cited the private jet charter page for the AEO-style pricing query, and Claude has cited the plumbing page three times on the pricing query. ChatGPT has cited both query types in Experiment A. The results tables, analysis, and FAQ below have been updated to reflect this.

A brand-new page with zero domain authority and zero backlinks earned citations in Perplexity, ChatGPT, and Gemini within [timeframe] after applying Generative Engine Optimisation (GEO) and Answer Engine Optimisation (AEO) principles.

A retrofit of an existing page produced citations on Perplexity, ChatGPT, Gemini, and Claude, with Claude citing it three times on the same pricing query. ChatGPT was the only platform to cite a GEO-style recommendation query.

This page documents the full methodology, baseline data, and results, including the null results.

Why This Experiment Exists

Most guides on AI visibility are written retrospectively. The successful experiments get published; the unsuccessful ones quietly disappear. The result is a body of advice built on survivorship bias, where every technique appears to work because only the wins are reported.

This experiment is different. It tests two pages in public, against a documented baseline, with a commitment to report whatever happens, including nothing. Both pages belong to the author of this guide, so there is no client confidentiality to hide behind and no incentive to exaggerate the results.

GEO and AEO: Two Disciplines, One Experiment

Before describing the experiments, it’s worth being clear about the two disciplines being tested, because they are not the same thing.

Answer Engine Optimisation (AEO) targets short, direct answers. It’s about winning the kind of quick factual responses AI systems give to simple queries, “how much does a plumber cost in Guernsey?”, “what is the best accounting software for small businesses?”. The content strategy for AEO is structure-first: question-format headings, concise answers in the first one to three sentences, FAQ sections, and definition lists.

Generative Engine Optimisation (GEO) targets complex synthesis. It’s about being included in the longer, more nuanced responses AI systems give when someone is doing research rather than looking for a quick fact. The content strategy for GEO is depth-first: original research, case studies, documented experiments, and content that cites credible external sources. AI needs substance to summarise; thin content gets ignored.

This experiment tests both, because the two pages under study generate both kinds of queries:

Experiment A (private jet charter from Guernsey) produces both AEO-style pricing queries and GEO-style recommendation queries.

Experiment B (Guernsey plumbers directory) produces AEO-style pricing queries and GEO-style recommendation queries.

The results below show that the two disciplines do not respond identically, which is itself one of the findings.

The two experiments also cover the two most common situations a business will face:

Experiment A tests whether a brand-new page can earn AI citations without any prior authority, backlinks, or citation history.

Experiment B tests whether retrofitting an existing page, and correcting a significant content error on it, can shift its LLM visibility.

Experiment A: Private Jet Charter from Guernsey

The Setup

The page targets a specific, high-intent query: people looking to charter a private jet from Guernsey Airport. It is a niche subject with clear commercial intent, and one where LLMs are likely to be asked for recommendations, “How do I charter a private jet from Guernsey?” is exactly the kind of question a wealthy traveller might put to an AI assistant.

The page started with zero domain authority, zero backlinks, and no prior content on the topic. It was built from scratch, with both GEO and AEO principles applied before a single word was published.

What Was Applied

Answer Capsule structure (AEO). Every major question about private jet charter from Guernsey was answered in a tight 40–60 word capsule, complete, citable, standalone.

Entity clarity (GEO and AEO). Guernsey Airport (IATA: GCI), the Channel Islands, and the relevant geography were named explicitly throughout, avoiding ambiguity.

FAQ schema markup (AEO). Structured data was embedded at the HTML level to signal machine-readable Q&A pairs to crawlers.

Factual specificity (GEO). Aircraft types operating from GCI, typical routing options, regulatory considerations, and realistic pricing ranges were all included, the kind of detail LLMs favour when constructing answers.

Clean semantic HTML (GEO and AEO). Heading hierarchy, paragraph structure, and page metadata were built to be crawler-friendly from day one.

Depth and originality (GEO). The page contains route-specific and aircraft-specific detail not available in generic private aviation content, giving LLMs something to synthesise that they cannot source elsewhere.

What This Experiment Tests

Can a page with no prior authority earn AI citations if its content is structured correctly from the start, and does the answer differ for AEO-style queries versus GEO-style queries?

The traditional SEO assumption is that domain authority and backlinks are prerequisites for visibility. Both GEO and AEO theory challenge this, at least partially. LLMs are trained on content quality and structure, not purely on link graphs. A well-structured answer on a low-authority site may still get cited if it provides information that other sources do not.

The Honest Caveat

New pages face a real disadvantage: they may not have been crawled and indexed by AI training datasets at all. There is an inherent time lag between publication and LLM awareness that no amount of good structure can eliminate entirely. This experiment may take months to yield meaningful results.

Baseline Results

Before any changes were made, both target queries were run across all four platforms. Baseline citation count: zero across every platform.


Baseline captured [date]. Full baseline archive: [link to citation-images page]

Current Results

Experiment A moved from zero citations to being cited by Perplexity, ChatGPT, and Gemini. ChatGPT cited both query types — the only platform in either experiment to cite a GEO-style recommendation query.


Result captured [date].


Result captured 13 September 2026.

What This Suggests

RAG-based platforms, Perplexity and ChatGPT’s browsing mode, responded to both AEO-style and GEO-style queries. These platforms retrieve live web content at query time, meaning freshness and structure matter more than legacy authority. The timeline for results on these platforms was measured in weeks, not months.

ChatGPT’s result is the most significant in this experiment. It cited the page on both queries, the factual pricing question and the recommendation question. No other platform cited a GEO-style recommendation query anywhere in either experiment. This suggests that ChatGPT may be responding to the depth and originality of the private jet content, not just its structure. The page contains route-specific and aircraft-specific detail not available in generic private aviation content, which gives ChatGPT something to synthesise that it cannot source elsewhere.

Gemini also cited the page, but only for the AEO-style pricing query and only after a longer delay. Gemini draws heavily on Google’s existing index and established entity signals, which take longer to influence than live retrieval.

Claude has not yet cited the private jet page on either query. This is notable given Claude’s strong response to Experiment B, where it cited the plumbing page three times. The difference may relate to the nature of the content, the topical coherence of the host domain, or simply the timeline. The experiment continues.

Experiment B: Guernsey Plumbers Directory Page

The Setup

The page under experiment is a directory-style page listing plumbers and plumbing services available in Guernsey. It had been live for months before the retrofit began.

The Problem: A Roofing FAQ on a Plumbing Page

During the audit that preceded this experiment, a significant content error was discovered: the FAQ section on the plumbing page contained questions and answers about roofing. Not plumbing. Roofing.

This was almost certainly a copy-paste error from another page on the same site. But its consequences for both GEO and AEO are serious. An LLM crawling this page would encounter contradictory entity signals, a page nominally about plumbing, containing authoritative-sounding answers about a completely different trade. This creates exactly the kind of topical incoherence that causes AI systems to either ignore a page or reduce their confidence in citing it.

The Lesson from the Error

Content errors that humans overlook, because they skim pages visually, are the same errors that confuse LLMs. AI systems read everything on a page, weight it for topical relevance, and make citation decisions accordingly. A roofing answer on a plumbing page is not a minor slip. It is a signal that the page’s information cannot be trusted, and LLMs respond to that signal by not citing it.

The Retrofit: What Changed

Removed all roofing FAQ content and replaced it with plumbing-specific Answer Capsules (AEO).

Strengthened topical entity signals (GEO). Guernsey-specific plumbing context, Gas Safe registration, island supply constraints, typical local pricing, was woven throughout the page.

Added FAQ schema markup (AEO) at the structured data level, making the Q&A pairs machine-readable for both search engines and AI crawlers.

Improved heading hierarchy (GEO and AEO). Section headings were rewritten to be semantically clear, ensuring the page’s structure signals plumbing at every level.

Added citable specifics (GEO). Pricing ranges, regulatory requirements, and emergency procedures, the kind of factual detail LLMs look for when constructing definitive answers.

What This Experiment Tests

Does fixing and optimising an existing page improve its AI citation rate, and does the improvement differ between AEO-style and GEO-style queries?

This matters because the vast majority of businesses start this process with pages already in existence, often pages that have been live for years, with accumulated content errors, topical drift, and zero GEO or AEO consideration. Experiment B models that reality.

It also tests the error-correction hypothesis specifically: does removing actively confusing content produce a measurable improvement, over and above any positive effect from the new content added?

Baseline Results

Before the retrofit, both target queries were run across all four platforms. Baseline citation count: zero across every platform.


Baseline captured [date]. Full baseline archive: [link to citation-images page]

Current Results

Experiment B produced citations on all four platforms, on the AEO-style pricing query. The GEO-style “best plumbers” recommendation query remained at zero across every platform.

Claude’s result is the standout. It cited the plumbing page three times on the same pricing query, more than any other platform on any query in either experiment.


Result captured [date].


Result captured [date].

What This Suggests

Every platform cited the page for the AEO-style pricing query. No platform cited the page for the GEO-style recommendation query. On the retrofitted page, the AEO/GEO split is absolute.

Claude’s three citations on the pricing query are particularly interesting. Claude emphasises entity clarity, content structure, and named authorship. The plumbing page, a directory page on a Guernsey SEO domain, retrofitted with clear entity signals and specific factual detail, appears to have matched what Claude looks for. This contradicts the common assumption that Claude is slower to respond than other platforms.

It also suggests that correcting a contradictory content error is necessary but not sufficient for every query type. Removing the roofing FAQ stopped the page actively confusing crawlers. It was enough to earn citations on factual queries across all four platforms. It was not enough to earn citations on a competitive recommendation query.

Results Summary: What Actually Happened

Across both experiments, the pattern is clear:

All four platforms have now cited at least one page. Perplexity, ChatGPT, Gemini, and Claude have each responded.

AEO-style queries cited more readily than GEO-style queries. Every platform cited the AEO pricing queries. Only ChatGPT cited a GEO-style recommendation query, and only on Experiment A, the purpose-built page.

ChatGPT produced the broadest result. It cited Experiment A on both query types, the only platform to do so. It also cited Experiment B on the pricing query.

Claude produced the strongest single result. Three citations on the plumbing pricing query, more than any other platform on any query.

Perplexity cited both pages, on three of four queries. It responded to both AEO and GEO query types in Experiment A.

Gemini cited both pages on the pricing queries but not on either recommendation query.

Claude cited only the plumbing page, but cited it three times. It did not cite the private jet page at all within the measurement window.

Platform Experiment A Experiment B
Perplexity 2 of 2 1 of 2
ChatGPT 2 of 2 1 of 2
Gemini 1 of 2 1 of 2
Claude 0 of 2 1 of 2 (×3)

What This Means for Your Own GEO and AEO Work

Audit Before You Optimise

The roofing FAQ was not a small oversight. It was actively damaging the page’s topical authority. Before applying any GEO or AEO techniques to an existing page, read the entire page, including sections you did not write, looking for content that does not belong. One incoherent section can undermine an otherwise strong page.

Local Specificity Is a Competitive Advantage

Both experiments operate in a Guernsey context precisely because local specificity makes content harder to replicate and easier for LLMs to anchor. “Plumbers in Guernsey” is a more tractable niche than “plumbers in London.” If your market has a geographic component, use it.

Structure Does Not Guarantee Citation

Neither experiment produced universal citations. Both GEO and AEO improve the probability; they do not create a direct cause-and-effect relationship. Content quality, topical competition, domain trust, and LLM training cycles all play a role. This experiment does not claim otherwise.

Document Your Baseline

Before making any changes to a page, record its current state: what LLMs currently say when asked about your topic, whether your page or domain is mentioned, and what the content looks like. Without a documented baseline, you cannot accurately measure improvement.

Different Platforms Require Different Timelines

RAG-based platforms like Perplexity and ChatGPT responded on a timeline of weeks. Gemini responded more slowly. Claude, often assumed to be the slowest, produced the strongest single result on the plumbing page.

The lesson is not to write off any platform. Each responds to different signals and on different timelines, and the platform you least expect may be the one that cites you most.

AEO and GEO Are Not the Same Discipline

The results are clear on this point. Every platform cited the AEO-style pricing queries. Only ChatGPT cited a GEO-style recommendation query, and only on the purpose-built private jet page, not the retrofitted plumbing page.

This suggests the GEO result may depend on the page itself, not just the query type. The private jet page was built with depth, originality, and route-specific detail, genuine GEO substance. The plumbing page was a retrofit of a directory page. ChatGPT may be responding to the depth of the private jet content, not just its structure.

If you want GEO-style citations, you likely need GEO-style content: original, deep, and specific. Structure alone is not enough.

Retrofitting an Existing Page Can Work — Fast

Experiment B earned citations on all four platforms. The page had been live for months with a significant content error. Correcting the error and applying AEO principles produced citations across the board. The page was already indexed, which shortened the timeline considerably compared to Experiment A.

If you have existing pages with content errors, fixing them may be the fastest route to AI citations available to you.

New Pages Can Earn Citations Without Authority

Experiment A earned citations on three platforms despite having zero domain authority and zero backlinks. This is one of the most significant findings in the experiment. It does not mean authority is irrelevant, it means that on RAG-based platforms, and eventually on Gemini, content quality and structure can substitute for it. The window is open. It will not stay open forever.

Frequently Asked Questions

What is the difference between GEO and AEO?

AEO (Answer Engine Optimisation) targets short, direct answers, the kind of quick factual responses AI systems give to queries like “how much does a plumber cost in Guernsey?”. GEO (Generative Engine Optimisation) targets complex synthesis, the longer, research-style responses AI systems give when someone is doing research rather than looking for a quick fact. AEO is structure-first; GEO is depth-first.

Can a brand-new page earn AI citations without domain authority?

Yes. In Experiment A, a page with zero domain authority and zero backlinks earned citations in Perplexity, ChatGPT, and Gemini after applying both GEO and AEO principles. Traditional SEO authority was not a prerequisite for citation on these platforms.

How long does it take to get cited by AI systems?

It varies by platform and by query type. RAG-based platforms like Perplexity can cite a new page within days or weeks. ChatGPT browsing mode followed a similar timeline. Gemini cited both pages after a longer delay. Claude cited the plumbing page three times on the pricing query but did not cite the private jet page within the measurement window. AEO-style queries cited faster than GEO-style queries in this experiment.

Does fixing a broken page improve AI citations?

Yes, on factual queries. In Experiment B, correcting a contradictory content error (a roofing FAQ on a plumbing page) and applying GEO and AEO principles produced citations on all four platforms for the AEO-style pricing query. It did not produce citations on the GEO-style recommendation query. Fixing errors works, but not for every query type.

Which platform performed best in this experiment?

It depends on how you measure. ChatGPT was the broadest, it cited Experiment A on both query types and Experiment B on the pricing query. Claude produced the strongest single result, three citations on the plumbing pricing query. Perplexity was the most consistent, it cited both pages on three of four queries. Gemini cited both pages but only on the pricing queries.

Why did Claude cite the plumbing page three times but not the private jet page?

Claude cited the plumbing page three times on the pricing query, more than any other platform on any query in the experiment. It did not cite the private jet page at all. Claude emphasises entity clarity, content structure, and named authorship. The plumbing page, a retrofitted directory page with strong entity signals and specific factual detail, appears to have matched what Claude looks for. The private jet page, despite being well-structured, did not. This may be a domain authority signal, a topical coherence signal, or a timeline issue. The experiment continues.

Why did only ChatGPT cite a recommendation query?

ChatGPT was the only platform to cite a GEO-style recommendation query, “What are the best private jet options from Guernsey Airport?” in Experiment A. No platform cited the plumbing recommendation query in Experiment B.

The difference may be the page itself. The private jet page was purpose-built with depth, originality, and route-specific detail not available elsewhere. The plumbing page was a retrofit of a directory page. ChatGPT may be responding to the depth of the private jet content, not just its structure. This suggests that GEO-style citations require GEO-style content, original, deep, and specific.

What is the most important thing to do before starting GEO or AEO work?

Document your baseline. Before making any changes, run your target queries across Perplexity, ChatGPT, Gemini, and Claude, and screenshot the results. Without a baseline, you have no way to measure whether anything you do afterward actually works.

Does local specificity help with AI citations?

Yes. Both experiments used Guernsey-specific content, which is harder to replicate and easier for LLMs to anchor. “Plumbers in Guernsey” is a more tractable niche than “plumbers in London.” If your market has a geographic component, use it.

Do GEO and AEO require different strategies?

Yes. AEO is structure-first: question-format headings, Answer Capsules, FAQ schema, and concise factual answers. GEO is depth-first: original research, documented experiments, case studies, and content that cites credible external sources.

This experiment found that AEO-style queries cited universally across all four platforms on both pages. GEO-style queries only cited on the purpose-built private jet page, and only on ChatGPT. If you want GEO citations, you likely need GEO content, not just GEO structure.

How to Replicate This Experiment Yourself

The methodology used here is repeatable for any business, in any niche, at any scale.

Document your baseline. Run your ten most important target queries through Perplexity, ChatGPT, Gemini, and Claude. Run each query three times. Screenshot every result. Date every screenshot. Save them in a folder called “AEO/GEO Baseline.”

Separate your queries by type. Identify which are AEO-style (factual, pricing, definitional) and which are GEO-style (recommendation, comparison, synthesis). They will behave differently and should be measured separately.

Apply GEO and AEO principles to one page. Use Answer Capsules, FAQ schema, entity clarity, factual specificity, and clean semantic HTML for AEO. Add depth, originality, and citable specifics for GEO.

Test monthly. Run the same queries again. Screenshot everything. Compare to baseline.

Report honestly. Record what changed, what didn’t, and what surprised you. Null results are still data.

Build a measurement log. A simple spreadsheet with columns for Date, Query, Query Type (GEO/AEO), Platform, Times Run, Times Appeared, and Citation Frequency is enough.

About This Experiment

This experiment is run by Martin Loader, Guernsey Donkey SEO, based in Guernsey, Channel Islands. The full guide — Get Found in AI: The Definitive Guide to LLM Visibility — is available right here.

All baseline and results screenshots are archived at:

Experiment A: [link to citation-images page]

Experiment B: [link to citation-images page]

Results are updated as the experiments progress. Last updated: 13 September 2026.

Before You Publish

Replace all bracketed placeholders and add:

1. FAQPage schema using the FAQ questions and answers above
2. Article schema with you as the named author, datePublished, and dateModified
3. Organisation schema site-wide if not already present
4. Upload the screenshots and replace [IMAGE URL] with real paths
5. Add the page to your llms.txt with a plain English description
6. Submit to Bing Webmaster Tools for indexing

The page is now structured so that both GEO and AEO terms appear in the title, the opening Answer Capsule, every experiment section, the results tables, the FAQ, and the replication guide, without the terms feeling forced or repetitive.