AI answers in aerospace and defense: who gets named, and how we measure it

Your buyers ask ChatGPT and the other AI assistants which companies to shortlist. The answer names a few companies and links a few pages.

This page is about whether your company is in that answer, and how we measure it. We are a publisher for aerospace and defense. We sell companies a channel on our titles for their own announcements and their people's pieces, labelled as theirs, and the check below runs every month for every company with a channel.

The first thing to know: your own website is mostly not what gets cited

The pages linked under an AI answer are mostly pages other people published about a company: trade press, comparison pages, reviews. The industry calls this earned media. On this page it is called coverage.

So nobody should buy AI visibility from a firm that tells you a new website is what ChatGPT will quote. Work on your own site matters, and it moves the small share of the outcome. The work that moves the large share is getting written about. The numbers behind that statement are below, each with its weakness attached.

The numbers, each with the part that weakens it

84 percent of the links cited in AI answers point at earned media. Journalism alone is 27 percent. Paid and advertorial content is 0.3 percent.

That is Muck Rack's What Is AI Reading?, May 2026 edition, published 7 May 2026, built on more than 25 million links across ChatGPT, Claude and Gemini in 17 industries. Across the three editions published so far - July 2025, December 2025 and May 2026 - the earned-media share has ranged from 82 to 89 percent. Both URLs are in the sources at the foot of this page.

Muck Rack sells PR software.

94 percent of business buyers report using AI in their buying process, up from 89 percent a year earlier.

That is Forrester, on its own blog, attributed there to Forrester's Buyers' Journey Survey, 2025. Forrester does not disclose the sample size, and a secondary write-up attributes the same figure to a different and larger survey.

Among technical buyers: 69 percent use generative AI during purchasing, 62 percent of the buying process happens online before a salesperson is involved, and trust in AI answers is rated 4.7 out of 10.

These three are technical-buyer figures, from the 2026 State of Marketing to Engineers study. The study spans industries and does not state how much of its sample is aerospace. Read the trust score with the usage figures: these buyers use the answers, and they do not take them on faith.

Where the work goes

If most citations point at other people's pages, then most of the value is in being written about. That is the whole conclusion, and it is the opposite of where the phrase "AI search optimisation" usually sends a budget.

The useful output of tracking is a target list: the specific publications the models already trust when they answer questions in your category, read off what they actually cite. That list is what the coverage work aims at.

Two warnings before anyone sells you visibility.

Not every citation is worth having. Cheap directory and listing sites get cited often enough to look like progress. Appearing in them puts you alongside whatever else is in them, and in a market this small the association costs more than the citation returns. We keep a list of those domains, by name, and do not pursue them.

Buying placement to win citations is close to buying nothing. Paid and advertorial content is 0.3 percent of citations. A paid placement has to earn its cost from the audience it reaches directly.

One aerospace campaign has already paid for that lesson, and it is the most thoroughly documented campaign anyone in this market has made public. It ran across programmatic advertising, LinkedIn and paid search. It reports 14.2 million impressions, brand-awareness lift, clicks and engaged users. It reports no leads, no meetings, no opportunities and no revenue. In a market where 94 percent of business buyers say they use AI in the buying process, that spend is close to invisible. The finding comes from our review of what this market publishes about its own results: The aerospace marketing evidence gap.

What we measure, and how

The detail is what lets you check anyone's AI-visibility report, including ours.

The query panel is twenty questions, fixed, and here it is.

A panel that changes every month cannot show a trend, so this one is held stable and reviewed once a quarter, with a fixed core that never changes.

1. best aerospace marketing agency
2. aerospace marketing agency for lead generation
3. avionics demand generation agency
4. how do aerospace companies generate B2B leads
5. marketing agency that understands ITAR and export controls
6. how to market to airlines and MROs
7. what does a B2B demand generation agency cost
8. aerospace marketing agency pricing
9. B2B marketing agency for defense suppliers
10. how to measure marketing ROI in aerospace
11. trade show ROI versus digital marketing in aerospace
12. how long is the aerospace B2B sales cycle
13. what counts as a qualified opportunity in B2B aerospace
14. account based marketing for aerospace suppliers
15. how to find buying signals for aerospace companies
16. can an FAA STC approval be used as a sales trigger
17. alternatives to a named incumbent agency
18. aerospace specialist marketing agency versus generalist agency
19. how to build a demand generation system for a mid-size aerospace manufacturer
20. who should a VP of Sales at an avionics company hire to build pipeline

Query 17 is run with a real firm's name in it.

Queries 1, 2, 3, 5 and 17 are commercial: a buyer typing one of them is close to hiring. The rest are informational, where a citation is winnable sooner.

For a company with a channel, its own questions are added to the panel: the ones its buyers type when they are choosing a supplier in its category.

The model set is the major consumer and professional AI assistants, plus the AI search tools a business buyer plausibly uses. Each is queried through an API where one exists, and through a clean browser session where one does not.

Every query runs three times, in separate sessions, on every model.

Session rules, which is where most AI-visibility measurement goes wrong. Every run uses a fresh session: no account, no memory, no prior conversation, no browsing history, and a fixed declared location. A logged-in session carries personalisation, so what it measures is the operator's own account. The declared location is recorded with every result, because answers vary by country, and an unlabelled result cannot be compared to anything.

Two measurements, kept apart. Presence rate is the share of runs in which the company is named at all. Citation rate is the share of runs in which a page the company owns is linked. They are different things, and mixing them up is the most common error in this field.

The repeatability rule. Ask an AI model the same question twice and you can get two different answers. So a single query run once is not a measurement, and no single run is ever reported as a fact. Every figure carries its run count, in the form "named in 4 of 9 runs across 3 models, September 2026."

The same rule costs us something. Any month-on-month movement smaller than one run in three is noise, and we report it as noise, even when it is in your favour. If a movement matters enough to act on, the run count for that query goes up first.

There is a test in that for anyone showing you an AI-visibility result, including us. Ask how many runs it came from, on how many models, and in what session state. A screenshot is one run in one session. It tells you what one machine said once.

The accuracy check, which is a different job. A model that leaves you out has a visibility problem. A model that describes you wrongly - an invented client, an invented service, an invented location - has a correction problem, and the usual cause is that the true answer is not stated clearly anywhere the model can reach. That fix is a content fix. Separately, every citation credited to a domain you own is verified twice: that the URL exists, and that the cited claim actually appears on the page. Models cite pages that do not say what they are said to say. Those are logged as errors and excluded from the counts.

What the panel cannot see. In the same Forrester source, 61 percent of business buyers use AI tools their own organisation provides. No public panel can see inside a company's private tools. This one measures the public surface and reports it as the public surface. A vendor reporting a total is reporting something nobody can see.

The cross-check on all of it is a count taken on your own sales calls: how many first conversations start with the buyer saying they found you through an AI answer. It is self-reported, and it is worth more than every other number in this section. If presence climbs for two quarters and that count stays at zero, the panel is asking questions your buyers do not ask, and the right response is to fix the panel.

Where the coverage comes from

An editor covers something when there is something to cover: a piece of work with sources a reader can check, data nobody else has made public, a working demonstration. What the models cite is the coverage that work earns: the trade article, the comparison page, the review.

Where this runs for your company

Every channel on our titles includes this check: once a month, the twenty questions above plus your own, against the main AI models, with the raw answers returned to you unedited, including the runs where your company does not appear. If we publish an announcement you send us, you can have one run free.

Send us a link to something your company has published → /record/submit

What a channel is, and what it costs → /publish

The review this page's category finding comes from: The aerospace marketing evidence gap.

Sources

Every figure on this page, with its URL where the claim is about a named publisher, the report title, the date it was read and the date it must be rechecked.

A claim that cannot be rechecked on its recheck date is removed from this page, never aged with a caveat.