Blog · By industry

AEO by industry, where the work differs

Published 28 April 2026 · Updated 12 October 2026 · 8 min read

AEO by industry, where the work differs
Photograph: Declan Sun / unsplash

The short answer: The three layers of AEO, access, identity and corroboration, were the order of work in every sector we have audited. What changes is which layer is binding, and which class of source gets cited in the answer. Retail is the one sector with a published feed route into an assistant. In our regulated audits the citations skewed to registries and guidelines, and in our industrial audits to distributor catalogues. Find the cited source class first, then decide what to fix.

Most industry guidance in this field is a benchmark table: referral share by sector, citation share by sector, a leaderboard of domains. Useful once. It tells you where you sit. It does not tell you why the same fix that moved a payments company does nothing for a pump manufacturer, and that question is the one worth the afternoon.

The part that is the same everywhere

Can the engine fetch and read the page. Can it work out that the entity on the page, in the schema, and in the third-party listing are one company. Does anything it already trusts repeat the claim. Those three ran in that order in every sector in our audit sample, a few dozen audits to October 2026 weighted towards software, healthcare and industrial supply, and access was where most of the recoverable loss sat, whether the client sold insurance or industrial bearings. That is a sample of our own clients, not a census of the web.

Google is blunt about the corollary, for its own surfaces. Its documentation on AI features says there are no additional requirements to appear in AI Overviews or AI Mode, no other special optimisations necessary, and no special schema (Google Search Central, checked 12 October 2026). That rules out a vertical-specific AI schema for Google's AI features. We found no equivalent published statement from the other engines, so for them this is an inference, not a documented rule. Either way, the sector differences we can see are in retrieval: which sources get cited before the answer is written.

Source class, the variable actually worth measuring

Run a prompt set for a sector, which in our audit is about 45 prompts across five engines with retrieval on and off, and tag every cited domain by what kind of thing it is. You get a distribution. Across a few dozen audits to October 2026 it has been stable enough per sector to plan against. It is a distribution of citations, not of retrieval: we can see which domains an engine cited, not what it fetched and discarded.

In our audits, the mix for developer tooling leans on documentation, forums and repositories. For local services, it leans on map listings and review aggregators, with the brand's own site barely present. For regulated finance and health, it leans on regulators, professional bodies and reference databases. For heavy industry, it leans on distributor catalogues and spec pages that the manufacturer does not own.

The failure mode falls straight out of that. A team rewrites forty pages of its own site because rewriting pages is what the agency sells, in a sector where owned pages are perhaps a tenth of the cited mix. Nothing moves in the citation mix. The work was competent and aimed at the wrong corpus. This is the most expensive mistake we see in our audits.

Tag the domains before you write a brief. The first 90 days of AEO work puts that measurement in week one for exactly this reason.

Retail has a pipe the other sectors do not

Commerce is the one sector where we know of a published route to hand an assistant your catalogue directly rather than hoping a crawler infers it. OpenAI publishes a product feed specification for merchants who want items to surface inside ChatGPT, with nine required fields: a stable item id, title, description, product URL, brand, seller name, image URL, availability and price. Search eligibility defaults to true in that spec, checkout defaults to disabled and needs a separately enabled integration. The spec also says that eligibility does not guarantee display, so a clean feed buys you candidacy and nothing more.

Two consequences. First, a retailer with a clean feed is playing a structured-data game that looks far more like merchant-centre hygiene than like content work. Stale availability, a placeholder seller name, a price that disagrees with the landing page: those are now AEO bugs, not merchandising bugs. Second, we know of no equivalent published feed route outside commerce. A B2B software company cannot submit a feed of its product. It has to be inferred from prose, which is slower and noisier, and which is why AEO for B2B SaaS spends its effort on comparison pages and third-party listings instead.

Regulated sectors are answered out of somebody else's document

Health, finance, law and insurance share a shape in our audits. The answers hedge more and the citations skew institutional rather than vendor. One feature-level fact is worth knowing before you spend on it: Google has withdrawn the FAQ rich result entirely. Its Search Central changelog records that the FAQ documentation was removed because the feature is no longer shown in Google Search results (checked 12 October 2026). So FAQ markup is not a visibility lever on Google for anyone now, regulated or not. Whether assistants read that markup at all is something we cannot observe from outside, and we make no claim either way.

That is a Search feature rather than an assistant behaviour. What we do see in regulated audits is a split in what counts as evidence. Your pages are evidence of what you claim. The registry entry, the licence record, the professional body listing and the guideline document are evidence of whether the claim is allowed to be true. Work on identity in these sectors is largely about making those two sets of records agree, down to legal name and registration number.

Compliance review also changes the writing. Claims need a source attached and a reviewer's sign-off, which lengthens the cycle and rules out the fast-publishing tactics that work elsewhere. AEO for healthcare, with the compliance in mind goes through that properly.

Industrials, where the facts live in a PDF nobody can read

Manufacturing and industrial supply have a specific, fixable problem. The buying facts are real: dimensions, materials, tolerances, certifications, part numbers. They sit in a datasheet PDF, and often the spec table inside it is a scanned image or a vector drawing rather than text. When an answer in our runs cited one of those PDFs, it reproduced the title and almost none of the spec table. We cannot see what the engine extracted, only what the answer contained.

Meanwhile the distributor who carries your part has the same specs typed into an HTML table, with a schema.org Product block around it. In one audit in our sample, the manufacturer was absent from every answer we ran about its own flagship product while three distributor domains were cited, and the distributors had published the numbers as text while the manufacturer had not. One audit is one case, and we did not test the counterfactual.

The fix is unglamorous and cheap: publish the spec table as HTML on a page per part number, keep the PDF as a download, and make the part number a visible string rather than an image. Then check whether the distributor pages name you as the manufacturer in text. The self-check has a version of this you can run without tooling.

What differs, in one paragraph

Google's structured data gallery documents a long list of feature types, and most are bound to a vertical: job posting, recipe, local business, vacation rental, course list, dataset, product (gallery last updated 15 June 2026). The list moves, which is why we give no count here: the FAQ entry has since come off it. The markup your sector is eligible for is a fact you can look up in an afternoon, and it varies enormously. Prompt shape varies too. In our prompt sets, consumer queries are short, comparative and asked constantly, and the answers moved between runs often enough that a single run proved little. Industrial and enterprise queries are long, specific and rare, so one bad answer can sit unchallenged for a long time, which is an inference from how rarely those prompts are asked rather than something we have tracked. None of this changes what AEO is, which is why the naming argument between AEO, GEO and SEO stays a side issue.

Honest limits

Our source-class distributions come from a few dozen client audits to October 2026, weighted towards software, healthcare and industrial supply, so they are a sample of who hires us and not a census of any sector. Retail and local services are thinner in that sample, which makes the retail-moves-faster claim directional rather than measured. We report which domains an engine cited, never which it fetched, so nothing here establishes why a source was chosen or what was read and discarded; the industrial PDF finding in particular is a correlation from one audit, not a tested cause. The only documented no-special-schema statement we could open is Google's, covering its own AI features, and we have no equivalent statement from the other engines. We also cannot yet tell you how fast a mix drifts, only that it has drifted inside a single engagement, so re-measure rather than inherit.

FAQ

Does my industry need a different kind of AEO? It needs a different target, not a different method. Access, identity and corroboration applied in that order in every sector we have audited. Which one is binding, and which third-party corpus matters, is what changes.

Is there vertical-specific schema for AI answers? Not for Google's AI features: Google states that no special structured data is needed and that there are no additional requirements. We found no equivalent published statement from the other engines, so treat their behaviour as untested rather than settled. Ordinary vertical markup still helps for Search features, and those are genuinely sector-bound.

Why do retail brands see results faster? In our sample they did, and retail is the thinnest part of that sample, so read this as directional. Partly commerce has a published feed route into one assistant, with defined fields. Partly consumer prompts are asked far more often, so a change shows up sooner in a sample of that size.

How do I find my sector's source class? Run thirty to fifty buyer prompts across more than one engine, record every cited domain, and sort them into owned, review, forum, regulator, distributor and press. That gives you the citation mix, which is the only part visible from outside, not the engine's retrieval. The shape of that sort is your brief.

Sources

Related reading

AEO for B2B SaaSAEO for healthcare, with the compliance in mindThe first 90 days of AEO workAEO vs GEO vs SEO: what the three terms actually meanAEO for agencies, adding it to a retainerAEO for ecommerce, when the product is the answer