0
Published studies measuring citation share for any St Helena source, on any AI engine
Source: all eight research documents cross-validated for this page
1 in 1,000
Odds of two identical prompts returning the same list in the same order across 2,961 runs
Source: SparkToro with Gumshoe.ai, via Search Engine Land, January 2026
5+
Distinct places and organisations sharing this name in live search and AI results
Source: entity audit conducted for this page, August 2026
2023
The year Arfadia began documenting its GEO practice
Source: Arfadia dated archives
About GEO & AEO in Saint Helena
This is the unusual market where the case for citation work is stronger than the case for ranking work, and the reason is arithmetic rather than fashion.
Where Rank Tracking Breaks and Citation Auditing Does Not
Rank tracking needs volume to mean anything. On a base of tens of sessions a three-place move is noise, and no keyword tool reports usable figures for this territory anyway. Citation auditing has no such dependency. You fix a set of prompts, run them across the engines on a schedule, and record what came back: whether the answer was accurate, which sources it named, whether it referred to the right place at all. That method works identically whether the underlying audience is four million people or four thousand. It is the one measurement discipline that does not degrade as the market shrinks, which is why for St Helena it belongs at the front of the engagement rather than as an add-on once rankings plateau.
The Failure Mode Here Is Confusion, Not Absence
Most destinations worry about being left out of AI answers. St Helena has a different problem: it gets answered about, confidently, and sometimes the answer is about a different place entirely. There is a wine town in Napa Valley, a parish in Louisiana, an island off South Carolina, a hospice in Colchester, and an airport in Montana whose name is close enough to swap. Retrieval systems mix them. Booking aggregators have published claims about non-stop flights to St. Helena from dozens of United States cities, which is false for the South Atlantic territory served by one weekly Airlink service from Johannesburg to HLE, with a seasonal Cape Town rotation and a roughly four-weekly Ascension connection. A traveller who acts on the wrong answer does not become a complaint. They just disappear.
A Small Source Pool Cuts Both Ways
Almost everything authoritative about this territory comes from a handful of places: the St Helena Government site, the official tourism site, one long-standing independent resource maintained by a private individual, the Wikipedia article cluster, and FCDO travel advice on GOV.UK. That concentration is a risk, because a single stale page carries disproportionate weight. It is also the opportunity. When the total corpus is this small, a few well-structured, clearly dated, unambiguous pages can shift what models say. In a market with a million competing sources, that kind of influence is simply not available.
The Same Question, Answered About the Wrong Place
A model does not tell you which St Helena it picked. It simply answers. The difference between these two responses is not tone or detail, it is whether the traveller ends up in the South Atlantic or in a Californian wine town.
How do I get to St Helena, and what should I know before booking?
An answer assembled from whichever St Helena the retrieval ranked highest:
Flight options that do not exist for this territory
Entry rules taken from a page in a different language and vintage
An investment contact wound up in 2021
Booking aggregators
Superseded PDFs
Legacy news archives
How do I get to St Helena, and what should I know before booking?
An answer built from dated, disambiguated, official sources:
One weekly Airlink service from Johannesburg to HLE
Entry permit rules cited to the current government page
Investment enquiries routed to the current coordinator
sainthelena.gov.sh
Official tourism site
GOV.UK travel advice
Five names, one territory
A Napa Valley town, a Louisiana parish, a South Carolina island, a Colchester hospice and a Montana airport all sit close enough in name to be retrieved in place of the South Atlantic territory. Entity grounding is not a refinement here. It is the first deliverable.
Illustrative of the conflation pattern documented across the cross-validated research for this page, including one case where a research pass returned passenger statistics from a similarly named airport in the United States. Not a captured screenshot of any specific query.
What Can Be Measured Here, and What Cannot
Our RoGEO framework measures citation frequency, reference depth and revenue attribution. On an account this size only two of those three can be evidenced, and we say so before the contract rather than after it.
RoGEO components against this account
Citation frequencyFixed prompt panel, scheduled re-runs
Reference depthWhich source is named, and does it support the claim
Revenue attributionAnswer exposure to a booking, at this volume
Attribution from an AI answer to a confirmed booking needs referral volume that does not exist on a base of 889 leisure visitors a year. We report it as unavailable and set the engagement's success criteria on the two components that can be evidenced.
26 vs 29
When the Official Source Contradicts Itself
The government's own fibre optic cable project page carries two different landing dates for the same event, 26 August and Sunday 29 August 2021, and two different branch lengths appear across official material. A retrieval system cannot arbitrate that. It picks one. Reconciling the owned corpus is cheaper and more effective than any amount of new content.
6 of 6
Categories Where the Honest Answer Is Unavailable
Prompt volume, on-island AI adoption, citation share, AI Overview trigger rate for this territory, referral traffic from AI answers, and return on GEO spend. Every one of the eight research documents cross-validated for this page reached the same conclusion on each. We do not fill those gaps with global averages borrowed from other markets.
Published figures on AI Overview prevalence and citation concentration vary widely by query set and method, and none of the studies reviewed here sampled queries about this territory. They are not presented as St Helena measurements.
Our GEO & AEO Services for Saint Helena
Getting the right island named, described correctly, and attributed to the right source, across every engine a prospective visitor or investor is likely to ask.
Baseline Citation Audit
A fixed panel of prompts covering access, entry permits, currency, safety, health and medical evacuation, relocation, work permits, company registration and investment incentives, run across ChatGPT, Perplexity, Gemini, Claude, Copilot and Google AI Overviews under declared United Kingdom and South African location conditions.
Each observation logs the exact prompt, engine and mode, whether live retrieval activated, the sources named, whether each source actually supports the claim attached to it, and whether the answer described the correct territory at all.
Entity Grounding and Disambiguation
Explicit, repeated signalling that this is the British Overseas Territory in the South Atlantic, in body copy, structured data, internal linking and external reference alignment.
This also means separating the island from the combined territory of St Helena, Ascension and Tristan da Cunha, because published figures for the two differ by roughly a factor of two and secondary sources merge them constantly.
Owned Corpus Reconciliation
Before any new content, we find and fix the contradictions already sitting in the authoritative owned sources: duplicate dates, superseded notices that remain indexed, orphaned pandemic-era files, and figures that disagree across departments.
Visible effective dates, one current page per topic, redirects from superseded ones, and consistent terminology across government and tourism domains. Unglamorous, and the highest-yield work available here.
Extractable Answer Architecture
High-risk pages open with a short direct answer, then detail, then exceptions, then the source link. Tables for logistics, one question per block, dated statistics with the measuring body named.
Audiences kept separate rather than merged into one page about the territory: short-stay visitors, returning Saint Helenians, prospective employees, long-term residents, foreign companies, approved investors, and yacht and cruise arrivals each have different rules and models blur them when the page does.
Third-Party Source Correction
When the owned footprint is small, third-party pages carry disproportionate weight. We identify the aggregator, directory and travel-trade pages that models are most likely to retrieve, and pursue correction of the ones publishing wrong access, permit or currency information.
Correction rather than manufacture. Building repetitive content purely to simulate corroboration is a tactic we do not use and would not recommend.
Scheduled Re-Audit and Reporting
Monthly or quarterly re-runs of the same panel, reported as verbatim answers, source inventories, error registers and change notes against the previous run. Not a share-of-voice percentage, because the denominator would be invented.
Every report states plainly that model outputs shift without vendor action, so movement is evidence to interpret rather than performance to claim.
the risk here is not being missed, it is being answered about wrongly
Why Choose Us as Your GEO Agency for Saint Helena?
A Method That Does Not Need Volume to Work
We have been documenting GEO practice since 2023, and the discipline we brought to larger markets is the same one that survives here: fixed prompts, logged sources, named errors, no invented denominators.
RoGEO
Our Measurement Framework
We Fix the Entity First
Everything downstream depends on a model knowing which St Helena it is describing. We treat disambiguation as deliverable one, not as a paragraph inside a content brief.
We Audit, We Do Not Screenshot
Identical prompts return different lists on repeat runs, at odds documented at roughly one in a thousand for the same order. A single screenshot proves nothing. We log repeated runs under controlled conditions and report the pattern.
We Publish Our Own Gaps
Six measurement categories on this account are genuinely unavailable, and they are named on this page rather than buried. An agency that cannot tell you what it cannot measure is not measuring.
We Ask for a Local Counterpart
Flight arrangements, immigration terminology, government portfolio names and which sources remain current are things a remote team gets wrong without a named validator on the island. We build that into the engagement instead of pretending distance is free.
Explore Related Services
Citation work and search work share one foundation here, and the entity layer serves both.
Start With the Audit, Decide Afterwards
We will run a fixed prompt panel across the major engines, log what they say about the territory and which sources they name, and hand you the error register. If it shows nothing worth fixing, we will tell you that. Contact our team to get started.
Request a Baseline Citation Audit
Frequently Asked Questions About GEO & AEO in Saint Helena
Is GEO worth doing for a territory of fewer than four thousand people?
Not as a service sold to residents or to island businesses targeting residents. The buyer base is a few dozen firms and the internal audience is too small to justify any recurring retainer. It is worth doing in one specific form: managing how the outside world's AI systems describe access, entry rules, currency, safety and investment for people in the United Kingdom, South Africa and further afield who are deciding whether to visit, relocate or invest. Start with a fixed-scope audit and let the findings decide whether anything ongoing is justified.
How can you measure results when no tool reports prompt volume here?
By changing the unit of analysis from the visit to the generated answer. A fixed prompt panel, run across the engines on a schedule under declared location conditions, produces a comparable record every cycle: what the model said, whether it was accurate, which sources it named, whether the source supports the claim, and whether it identified the right place. None of that requires volume. What it will not give you is a share-of-voice percentage, and any vendor quoting one for this territory has chosen a denominator out of thin air.
What is the single biggest AI risk facing this territory right now?
Confusion with similarly named places. A Napa Valley town, a Louisiana parish, a South Carolina island, a UK hospice and a United States airport all sit close enough in name to be retrieved in place of the South Atlantic territory. The practical damage shows up as wrong flight claims, wrong entry rules and wrong contacts presented with full confidence. During research for this page, one automated pass returned annual passenger statistics from a similarly named airport in Montana and presented them as this island's figures. That is the failure mode, in miniature.
Do you guarantee that AI engines will cite us?
No, and treat any guarantee as a procurement warning. No vendor controls retrieval or generation across platforms, model versions change without notice, and identical prompts return different results on repeat runs. What we commit to is method: a defined prompt panel, controlled conditions, logged outputs, a named error register, prioritised corrections, and a re-audit that shows what moved. Movement in citations is evidence to interpret, not a result to claim credit for automatically.
Should the strategy be built around Wikipedia?
Wikipedia matters for entity understanding and the article cluster covering this territory is unusually thorough, so keeping it accurate and properly sourced returns more than almost anything else, at very little cost. But it is not owned media, it is not a promotional channel, and it cannot be treated as a guaranteed citation source. Different engines weight it very differently, and one published analysis across a large mixed-topic citation dataset found a gap of more than a hundredfold between engines in how often Wikipedia appeared. Build on it, do not build only on it.
Is schema markup enough on its own?
No. Structured data reduces ambiguity and helps machines extract cleanly, which matters a great deal when the entity is easily confused. It does not substitute for accurate content, corroboration from independent sources, freshness, or an authoritative page that actually exists. A 2026 critical survey of the GEO research literature warned specifically that citation-oriented changes do not automatically improve retrieval or downstream outcomes, and that discoverability, retrieval, citation, prominence and commercial results are separate variables that should not be collapsed into one claim.
What should be audited first?
The categories where a wrong answer changes a real decision: how to get here and on which day, entry permit requirements and validity, travel and medical insurance, money and card acceptance, medical evacuation, work permits, company registration and investment incentives. Marketing copy about scenery can wait. A model that describes the view accurately but the flight schedule wrongly has still cost the territory a visitor.
Does a foreign vendor create a data protection problem?
Citation auditing is unusually low risk on this front because the work sits almost entirely on public web content and model outputs rather than personal data. We use standardised research accounts, keep real traveller, employee and resident details out of prompts, and exclude session identifiers from reports. The picture changes if analytics, survey responses or enquiry records are shared, in which case a controller and processor agreement is required. St Helena holds no UK adequacy decision, so any transfer is treated as a third-country transfer with an appropriate safeguard and a documented transfer risk assessment.
How does this differ from GEO for a normal client?
Three ways. Content velocity is low, because a territory this size does not generate frequent news, so volume-based content strategies have nothing to feed on. Third-party authority matters more than owned content, because the owned footprint is small. And the reporting is qualitative by necessity: verbatim answers, source inventories and error types rather than dashboards. Structurally it looks less like a retainer and more like an audit, a correction cycle, and a periodic re-check.
What would make us recommend stopping?
If the baseline audit shows accurate answers, official sources being named, and no meaningful conflation, then the problem we would be selling against does not exist and we will say so. The honest continuation triggers are the opposite: repeated factual errors in categories that carry traveller or public service risk, persistent citation of superseded sources, or repeated omission from prompts where the territory is clearly relevant. Absent one of those, the correct recommendation after an audit is to stop rather than to manufacture an ongoing engagement.
Ungrounded entity
Grounded entity
