Your agency needs to understand your buyers' questions, which is learnable in weeks, and somebody needs to be right about your subject, which is usually you. The one page-level lever anyone has measured at scale is whether the page's language matches the question asked: Chen and Koudas, across 158,099 AI answers carrying 468,037 citations, found that within a single website the titles and paragraphs matching the question get cited more, while page properties explain about 2.8% of the variation and the engine's own retrieval explains about 32% (arXiv preprint 2610.15033, submitted 10 October 2026). Their findings replicated in a market of small B2B vendors. Vocabulary and construction transfer between industries. Factual accuracy does not, and no engine will catch an error for you.
Your AEO agency does not need to understand your industry the way your sales engineers do, and it does need to understand two specific things that people lump in with industry knowledge and should not.
The first is how your buyers phrase a question when they are not talking to you. That is not product expertise. Nobody at your company knows it either, because your internal vocabulary drifts from the market's the longer you work there. It is collected from search data, from the prompts that already reach your site, and from the sales calls where somebody asked for the wrong thing by the wrong name. An outsider is sometimes better at hearing it.
The second is who else is in the answer. Knowing that three named competitors own the comparison pages for your buying question is market knowledge, and it takes an afternoon to collect rather than a career.
What does need real subject expertise is being correct. Specifications, tolerances, regulatory language, what your product does not do. An agency writing those from a summary of a summary will produce pages that read well and are wrong in ways your buyers notice immediately.
So the question to put to a shortlist is not whether they have worked in your sector. It is which of those three jobs they are taking, and who does the rest. An agency that says it will handle all three in an unfamiliar technical field is either unusually good or has not thought about it. An agency that says it will handle the mechanics and wants a named reviewer from your side has described how this actually works.
The mechanical part of AI visibility work is the same in every industry, and it is most of the work by volume. Three things carry across without translation.
Page construction carries across. Whether a question appears in the title, whether the answer arrives in the first sentence of the section rather than the fourth, whether a section can be understood by a reader who arrived from a citation and never saw the section above. None of that is industry knowledge. All of it is editing discipline, and it is what most pages fail on.
Retrieval mechanics carry across. Whether a crawler can read the page at all, whether the content is in the HTML or assembled by a script after load, whether a login sits between the engine and the answer. Google's own guidance on its AI features states that a page must be "indexed and eligible to be shown in Google Search with a snippet" and that "There are no additional technical requirements", adding that no new machine-readable files, AI text files or special schema.org data are needed. A pharmaceutical equipment maker and a payroll platform fail that check in identical ways.
And the evidence base carries across, which is the part that surprises people. Discovered Labs analysed about 2 million prompt, engine and citation observations over six months across ChatGPT, Claude, Google AI and Gemini, crawling 10,000 cited URLs, and found prompt-content alignment the strongest page-level predictor at a standardised beta of +0.37, 95% confidence interval +0.33 to +0.41, a result that had to survive at least four of nine robustness checks. Discovered Labs sells AEO services, so its choice of what to measure is not disinterested, and its method is more careful than most of this field. Both of those are worth knowing when you quote it.
Industry knowledge changes the outcome in one place that matters enormously and several that matter less than the sales pitch suggests. The place it matters is factual accuracy, because nothing downstream will catch an error.
Onweller and colleagues parsed the inline citations out of reports written by 14 deep research models, re-fetched every cited page, and scored each citation three ways (arXiv preprint 2605.06635, submitted 7 May 2026). Frontier models kept link validity above 94% and topical relevance above 80%. Factual accuracy against the cited source ran only 39% to 77%. The authors also found that accuracy degraded as retrieval deepened, with fact-check scores falling by about 42% on average as tool calls increased from 2 to 150.
Read that from the position of a company whose page is being cited. The layer reading your site is good at establishing that your page exists and is on topic, and much weaker at establishing that it supports the sentence it is being cited for. A wrong tolerance, a superseded standard, a claim your product cannot support: those can be lifted and repeated without anybody verifying them, and the correction arrives from a customer rather than from an engine.
Industry knowledge also changes the shape of the question set. In regulated categories, buyers ask about approvals and documentation before they ask about capability, and a page that leads with capability answers the second question. That is cheap to learn and expensive to guess.
Where industry knowledge changes less than claimed: the structure of the page, the technical fixes, the measurement design, and the judgement about which pages to work on first. Those are the same jobs in every sector we have looked at.
Agencies insist you need a vertical specialist partly because it is sometimes true and mostly because it is the strongest thing they can say about themselves. Read the claim as positioning before you read it as advice.
Search the question and the first page is almost entirely agency blogs, which all conclude that niche experience is important and several of which conclude that you should drop any agency without case studies in your exact sector. Those posts are written by firms with case studies in a sector. The argument is not dishonest and the incentive is obvious, so weigh it as practitioner opinion rather than evidence.
There is a real version of the argument underneath. An agency that already knows your vertical starts with the vocabulary, the competitor set and the objections, which saves you weeks of ramp. If two shortlisted firms are otherwise equal, that is a reasonable tie-break and you should take it.
What the argument hides is the cost of the filter. Requiring case studies in your exact category removes most of the market, including every firm whose relevant strength is editing discipline and measurement, and it selects for agencies that work with your competitors. Our own vertical research exists for this reason rather than as positioning: our report on pharmaceutical process skid suppliers ran buying questions through ChatGPT, Google AI, Perplexity and Claude across eight categories, and no supplier won all four engines. Doing that work in a category teaches you the category in a fortnight.
Ask any firm claiming vertical expertise one question: show me a page you wrote in a technical field you did not previously know, and tell me who checked the facts. The answer separates expertise from familiarity.
Testing whether an agency can learn your industry is a two-hour exercise, and it is a better predictor than any case study. Give them a real question and watch what they do with it.
Hand them one buying question your sales team hears weekly and ask for a one-page outline: the heading they would use, the first sentence of the answer, the three adjacent questions they would cover, and the five facts they would need you to supply. You are not grading the writing. You are checking whether they came back with questions about your product or with confident filler.
Two failure signals are reliable. An outline with no list of facts needed from you means they intend to write around the specifics, and specifics are what gets a page cited. An outline whose headings use your internal product names rather than the words buyers type means they skipped the vocabulary step, which is the single highest-weighted thing a page can get right.
| Capability | Transfers between industries | Who should own it |
|---|---|---|
| Page structure and answer-first editing | Yes | Agency |
| Crawler access and rendering fixes | Yes | Agency with your developer |
| Buyer vocabulary and question set | Learnable in weeks from data | Agency, confirmed by your sales team |
| Competitor and comparison set | Learnable in days | Agency |
| Technical facts, specifications, limits | No | You, with a named reviewer |
| Regulatory and approval language | No | You or a specialist reviewer |
| Measurement design and reporting | Yes | Agency |
Run the same exercise with both finalists on the same question. The comparison is more informative than their references, and it costs you one meeting each.
Facts should be owned by a named person on your side with a fixed turnaround, and getting that one arrangement right removes most of the risk of hiring outside your vertical. Vagueness here is what produces both the errors and the delays.
Name one reviewer per topic area rather than a committee. Give them a defined window, two working days is usually enough, and restrict their brief to whether the claims are true and whether anything is missing that a buyer would ask. Reviewers who are also asked for an opinion on wording will supply one, and the page will come back softened into the kind of prose that cannot be quoted.
Keep a short list of claims the agency may never write without checking: numbers, certifications, comparisons to named competitors, and anything about what your product cannot do. Everything not on that list can move at content speed.
Expect the arrangement to cost internal time, and budget it honestly. Our own published early case studies record an audit score moving from 17 to 68 on one engagement, and say plainly that the score is an input measure of citation readiness rather than proof of citations. The pages that moved it needed facts only the client had.
If you want a read on which of your pages are failing on mechanics rather than on subject knowledge, that is what a free visibility assessment is for, and it tells you how much of the problem actually needs somebody who knows your industry.
People ask these five questions about agencies and industry experience while building a shortlist and trying to work out which credentials are worth filtering on.
No, because that filter throws away most of the market and selects on the wrong thing. Ask instead for a case study where the agency had to learn an unfamiliar technical subject, and ask who wrote the copy and who checked it. An agency that can show its fact-checking process is safer than one that merely shares your vertical.
Knowing your market and being useful to you are different things, and an agency deep in your vertical usually has both the knowledge and a conflict. The knowledge is real: they have already learned the vocabulary buyers use. The conflict is that your positioning is now in a room where your competitor's positioning also lives, and nothing in a contract fully fixes that.
The error propagates, because the agents reading the web check reachability and topic far better than they check facts. In a 2026 study, Onweller and colleagues evaluated 14 models and found that link validity ran above 94% and topical relevance above 80% for frontier models, while factual accuracy against the cited source ran only 39% to 77%. A wrong number on your page can be repeated without anyone verifying it.
Regulated industries change the review process rather than the page mechanics, and that is worth paying for separately. What you need is someone who can work inside your approval cycle without stripping the specific claims that make a page worth citing. Experience with your regulator matters more than experience with your product category.
Name one internal reviewer per topic, give them a fixed turnaround, and limit their job to facts rather than style. Most delays in this work come from review that wanders into wording, and most factual errors come from nobody being specifically responsible for the numbers. Separating the two jobs fixes both.
Related guides
A free visibility assessment scores your pages on the things that transfer between industries, and tells you which gaps need somebody who knows your subject and which need an editor.
Request a free visibility assessment →