Open ChatGPT and ask it who your company is. Then ask Perplexity the same question. If the two answers differ in any material way, you do not have a search visibility problem. You have an identity problem, and it is removing pipeline you never see enter play.
That test takes ninety seconds, and almost nobody in MedTech has run it. The results are consistently worse than executives expect, and the reasons are structural rather than careless.
Ranking and citation are different mechanisms
Search engines rank documents. Generative engines cite entities. Those are not the same operation, and optimising for the first does not reliably produce the second.
When a generative engine composes an answer, it assembles and attributes claims. To attribute a claim to your company, it has to be confident about what your company is. That confidence is built from consensus across the sources it draws on. Where the sources disagree, confidence drops, and the engine attributes the claim to a competitor it can describe cleanly.
This is why companies that rank well in conventional search can be absent from AI-generated answers in their own category. The content is fine. The pages are indexed. The entity underneath them is not resolvable, so nothing attaches to it.
Why this is worse than a ranking problem
A ranking problem is competitive. You are behind someone, and you can measure the gap. An entity resolution problem is categorical. You are not in the consideration set at all, and there is no position to measure yourself against because you do not appear in the output.
It is also self-reinforcing. Every new piece of content published under a fragmented identity adds another slightly different signal, which lowers consensus further. Companies that respond to invisibility by publishing more are frequently making it worse.
MedTech has this worse than any other category.
The fragmentation is structural. A medical device company generates identity signals across an unusually wide set of sources, and each source has its own naming convention, its own submitter, and its own decade of accumulated history.
Your website carries a marketing name. Your FDA 510(k) filings carry a legal entity name, frequently in capitals and frequently abbreviated to fit a field. Your ClinicalTrials.gov registrations carry a sponsor name that may predate a rebrand by years and that clinical operations has no reason to update. The papers your key opinion leaders publish carry author affiliations typed by hand, differently, by different co-authors, across a decade. Your LinkedIn page carries whatever was entered when it was created. Press releases carry whichever version the agency used that quarter.
Five or six variants of one company is normal in this category. Each one splits the signal, and none of the people maintaining them has any reason to think it matters.
The two most commonly overlooked sources
Trial registries are the first. They are maintained by clinical rather than commercial, nobody has ever asked whether the sponsor name matches the brand, and the record persists indefinitely. A company that rebranded three years ago frequently still has its previous identity attached to every trial it has ever run.
Author affiliations are the second, and they are the more damaging of the two. In clinical contexts, peer-reviewed literature carries disproportionate weight in what an engine treats as authoritative. Affiliations on those papers are the mechanism by which your company is connected to that literature, and they are entirely outside your control unless you have asked for them explicitly.
The ten sources that matter
An entity audit checks name consistency and description consistency across the sources a generative engine actually weights in healthcare. Ten sources, ninety minutes, one page.
• Your own website, including the About page and the Organization schema block
• The FDA 510(k) or PMA database entry
• ClinicalTrials.gov sponsor and collaborator records
• PubMed author affiliations on papers involving your KOLs and your own clinical staff
• Wikipedia and Wikidata, if entries exist, and the separate question of whether they should
• Crunchbase or the equivalent business registry for your market
• The LinkedIn company page
• SEC filings, where applicable
• The three leading trade publications in your category
• Your own press releases across the last three years
For each source, two checks. Does the name match your canonical version exactly, character for character? Does the description of what the company does match your canonical description?
Score it honestly. Zero to one variant means you have already solved this, which places you in a small minority. Two variants is normal and fixable within a quarter. Three or more means no engine can form a consensus about you, and no volume of content will change that until the names are reconciled.
What a canonical description has to contain
One sentence, used identically everywhere, containing four things. The legal entity name. The category you operate in, phrased the way buyers phrase it rather than the way your positioning deck phrases it. The specific problem you address. The market you serve.
The most common failure is a canonical description written in positioning language. An engine cannot resolve “transforming outcomes through intelligent surgical innovation” into a category, because that sentence does not contain one. It can resolve “a medical device company producing intraoperative imaging systems for breast surgery.”
Write the boring version. The boring version is the one that gets cited, and the positioning language can live everywhere else on the page.
The test for a canonical description
Give the sentence to someone outside your industry and ask them to name your category and your buyer. If they can do both from the sentence alone, it will resolve. If they ask a clarifying question, an engine would have had to guess, and guessing is what produces the competitor citation.
Where the sameAs schema goes and what it actually repairs
The sameAs property declares that a set of URLs unambiguously refer to the same entity. Schema.org defines it as the URL of a reference page that indicates the item’s identity, such as a Wikipedia page, a Wikidata entry, or an official website. Google’s structured data documentation confirms that it makes general use of sameAs alongside other schema.org markup.
It belongs in the Organization schema block on your homepage and About page. Person schema for named executives should carry its own sameAs array pointing to their LinkedIn profile, their publication record, and any author page.
The critical sequencing point: sameAs does not fix inconsistent names. It connects consistent ones. If you declare that four different entity names are the same thing without giving an engine any independent reason to believe it, you have documented the inconsistency rather than resolved it. Fix the names first. Then declare the connections.
What this has to do with SEO, AEO and GEO
The three terms are used interchangeably by most agencies, though they describe three different optimization targets.
SEO ranks you
Document-level competition for position in a results page. Still real, still worth doing, and increasingly the layer that a smaller share of your buyers passes through.
AEO answers for you
Answer engine optimization is about being the source of a direct answer. It rewards extractability: a question stated plainly, an answer given in one self-contained block, no requirement to read around it.
This is why FAQ blocks still matter even though Google deprecated FAQ rich results in May 2026. The rich result is gone. The schema type is not, and an explicitly labelled question-and-answer block remains close to the ideal input format for a system assembling an answer. The two things are routinely conflated, and the distinction is worth being precise about.
GEO cites you
Generative engine optimization is about being attributed as the source of a claim inside a composed answer. It is the layer that depends most heavily on entity resolution, because attribution requires an entity to attribute to.
Most B2B companies are winning one of these three and losing the other two without knowing it, because they are measuring only the first.
Run this before you fund another content quarter.
Ninety minutes, ten sources, one page. The reason to do it before the next content budget rather than alongside it is that content published under a fragmented entity accrues to nobody.
You can produce genuinely excellent material for four consecutive quarters and generate no citation position, because each piece is attached to a slightly different thing and none of them reaches the threshold where an engine will attribute to it. The entity layer is the floor. Nothing above it works until it is fixed.
It is also, by a wide margin, the cheapest intervention available in this category. Reconciling names across ten sources is administrative work. Adding a sameAs array is a ten-minute change. Neither requires a content budget, an agency, or a strategy document.
Start with the ninety-second version. Ask two engines who your company is. If the answers do not match, you have your answer, and the ninety-second version is worth putting in this week rather than next quarter.
Frequently Asked Questions
Why is my company not cited by ChatGPT?
Generative engines cite entities they can resolve with confidence. If your company is described inconsistently across the sources an engine draws on, confidence drops and the engine attributes the claim to a competitor it can describe cleanly. This is an entity resolution problem rather than a ranking problem, and publishing more content does not fix it.
What is entity consistency in AI visibility?
Entity consistency means your company name and description are identical across every source an engine weights, including your website, regulatory filings, trial registries, author affiliations, business registries, and trade press. Inconsistency splits the signal and prevents an engine from forming a confident picture of who you are.
Does sameAs schema improve AI citation?
sameAs declares that multiple URLs refer to the same entity, which helps an engine consolidate signals. It connects consistent identities rather than repairing inconsistent ones, so naming should be reconciled before the markup is added.
Which sources do AI engines weight most in healthcare?
In clinical contexts, peer-reviewed literature and author affiliations carry disproportionate weight, alongside regulatory databases and trial registries. These are frequently the sources a commercial team has least control over and has never audited.
What is the difference between SEO, AEO and GEO?
SEO competes for position in a ranked list of documents. AEO optimises for being the source of a direct answer and rewards extractable, self-contained content. GEO optimises for being attributed as the source of a claim inside a generated answer, and depends most heavily on entity resolution.
Does FAQ schema still matter after the May 2026 deprecation?
Yes, for different reasons than before. Google stopped showing FAQ rich results in search on 7 May 2026, but the FAQPage schema type remains valid and parseable. An explicitly labelled question-and-answer block is close to an ideal input format for a generative system assembling an answer, so the markup retains value for AEO and GEO even without the SERP feature.
How long does an entity audit take?
Approximately ninety minutes across ten sources. A two-engine diagnostic takes ninety seconds and will indicate whether the full audit is necessary.
What should a canonical company description contain?
One sentence containing the legal entity name, the category phrased as buyers phrase it, the specific problem addressed, and the market served. Positioning language should be avoided, because an engine cannot resolve a category from an aspirational statement.
EXTERNAL CITATIONS
• schema.org sameAs property definition
• Google Search Central, intro to structured data
• Google Search Central, FAQ structured data and the May 2026 deprecation notice
• Google Search Central, August 2023 changes to HowTo and FAQ rich results
• Google general structured data guidelines