When a flight department manager types "which charter operators fly Teterboro to Palm Beach with ARGUS Platinum" into ChatGPT, the reply arrives as a short paragraph and a few names. It looks like an opinion. It is closer to an assembly: a retrieval step that pulls pages from a search index, a reading step that extracts entities and claims, and a generation step that writes the paragraph. Each step has measurable preferences, and none of them is "the nicest website".

Step one: retrieval favors lists, and it favors fresh ones

In December 2025 Ahrefs classified more than 10,000 third-party URLs that ChatGPT cited across 750 top-of-funnel prompts, including agency and vendor recommendations. "Best X" lists were 43.8% of all cited page types — the largest single category — and they were even more prominent in Google's AI Overviews. Ranking higher on those lists correlated with being recommended. Freshness mattered: 79.1% of the cited lists had been updated in 2025, and 26% within two months. Perhaps most usefully for a small operator, 35% of the cited lists sat on low-authority domains, and self-promotional lists carried no observed penalty. 1

For charter, the lists that exist today are mostly written by brokers, jet-card comparison sites and trade press. EpicEdits' July 2026 analysis of ten commercial charter queries on US Google found operators holding about 22% of page-one results against 74% for brokers and intermediaries. 4 Answer engines retrieve from the same web, so the intermediaries' advantage carries straight into the answer.

Step two: the model prefers brands it has seen everywhere

Also in December 2025, Ahrefs correlated ChatGPT visibility with off-site signals across 75,000 brands. The strongest correlations were YouTube mentions (Spearman 0.737) and branded web mentions (0.664). Domain Rating — the number most SEO agencies sell against — came in at 0.266, and the number of pages on the site at roughly 0.19. Ahrefs' own summary: there is almost no relationship between content volume and AI visibility. 2

This is why an operator with a two-page site and a well-documented reputation in trade press, forums and video can be named ahead of a competitor with 400 route pages. It also explains where ChatGPT's own citations come from: in Profound's 680-million-citation dataset, Wikipedia alone accounted for 47.9% of ChatGPT's top-ten cited domains, with Reddit at 11.3% and Forbes at 6.8%; Perplexity leaned on Reddit (46.7%) and YouTube (13.9%). 5 Different engines, different diets — which is why an operator can be visible on Perplexity and absent from ChatGPT on the same day.

Step three: generation quotes what is quotable

The Princeton GEO study (KDD 2024) tested what kinds of page edits raise a source's visibility inside generated answers. Adding statistics and adding citations produced the largest gains — the tactic review of that study reports a 41% improvement from statistics and a 115% gain for a page that had been ranked fifth when citations were added; keyword stuffing reduced visibility on Perplexity by 10%. 67 A 2026 study of "citation absorption" across ChatGPT, Google and Perplexity found that high-influence pages are longer, well structured, and rich in extractable definitions, statistics, comparisons and procedures — and that ChatGPT cites fewer sources per answer, each carrying more influence. 8

For an operator, "quotable" means a dated page that states the certificate, the ratings with audit dates, the fleet by type, the bases by ICAO code and the missions flown — in plain HTML, not on a badge image or a PDF. Business Jet Traveler's 2026 charter guide tells buyers that FAA Part 135 compliance alone is not enough and points them to ARGUS TripCHEQ and WYVERN PASS trip reports and the ratings behind them; 9 those tiers are exactly the verifiable statements an engine reaches for when the prompt includes "safe". 10

Step four: the list is not stable

In January 2026 SparkToro had 600 volunteers run recommendation prompts 2,961 times. The chance that the same brand list appeared twice was under one in a hundred; identical ordering, under one in a thousand. Narrower categories were more consistent. Their recommendation is to measure visibility as a percentage across dozens or hundreds of runs, not as a rank. 3 The practical consequence is that any "we got you into ChatGPT" screenshot is unfalsifiable, and any before/after based on one run is noise. We built our method around this.

What the majors already own

The one logged aviation snapshot we know of — EpicEdits' 14-prompt run on 31 May 2026 across ChatGPT, Perplexity and Google AI Mode — named VistaJet in 11 of 14 prompts, NetJets in 8, Flexjet in 7, Wheels Up in 6 and PrivateFly in 5. 11 That is one day and 14 generic prompts, so treat the numbers as directional; but the shape is consistent with everything above. Generic prompts go to brands with the broadest mention. An operator with nine aircraft should not spend money trying to change that.

What to do instead

  1. Make your facts machine-checkable. One entity page: certificate, ratings with dates, SMS statement, fleet, bases, missions. Structured data for the organization; no facts hidden in images.
  2. Get on the lists that exist. Directory entries (NBAA, ARGUS/WYVERN listings, charter directories), trade-press coverage and jet-card comparison sites are the retrieval layer. Keep them current; freshness is measured. 1
  3. Answer specific questions, one per page. Route, airport, aircraft type, rating. 1,000–2,000 words with dates and sources; long-tail queries trigger Google AI Overviews far more often than head terms. 12
  4. Widen mention, not page count. Video, LinkedIn, trade press, forums. 2
  5. Measure share of voice across many runs, and keep the log. 3

How we know

This article draws on published studies from Ahrefs (December 2025), SparkToro (January 2026), Profound (2025), Princeton/KDD (2024) as reviewed by blckalpaca, Zhang et al. (2026), Seer Interactive (2026), Business Jet Traveler (2026), Schubach Aviation (2025) and EpicEdits (May and July 2026). We have not yet published our own measured data; the Business Aviation AI Visibility Index will, from edition 1. We describe EpicEdits' index as a competitor's work and cite it as such.