The founder describes the problem to an assistant. An answer can only tie a firm to it if some page does.

The buyer is an operator with a named problem, not a category

The person asking is a founder, an owner or an operator who can already describe what is wrong: pricing is broken, the fifth location is not working, the founder wants out in three years. They type the problem, the kind of company and often its size. The AI assistant with web search can only answer by matching that description against what firm pages state.

Large firms' pages and boutiques' pages can fail the same way: both describe capabilities ("growth strategy", "operational excellence") rather than problems. A boutique that writes "we fix pricing for B2B software companies with 20 to 100 employees" has written the one sentence the assistant is looking for, and a capabilities page has not.

The second part of the question is trust: who will do the work, and have they done it before. An assistant can read a partner's page and a written engagement description; it cannot read a reputation. A firm whose site names no one, or names people without saying what each has done, gives it nothing to weigh.

The third part is the engagement: how it starts (a diagnostic, a workshop, a fixed-fee first phase), how long it runs, what it costs to begin. Stated in text, with or without a figure, it answers the "how much" question many firms leave to the first call.

What a consulting firm's site has to state before an assistant can name it

What an assistant reads is what your server sends, fetched the way an AI crawler fetches it; for ChatGPT, OpenAI documents the ChatGPT-User fetcher doing this when someone asks about your firm. It has not sat in your steering committee. It knows what the page states.

schema.org has no consulting type, and ProfessionalService, the LocalBusiness subtype older guides suggest, is marked deprecated there. Google's organization documentation lists the Organization fields it reads: name, url, logo, sameAs, contact points. Each partner can be a Person node linked from their page, and each publication can carry Article markup with a headline, dates and an author with a URL, as Google's article documentation describes. That is how a byline becomes a fact a machine can attribute, and an author profile becomes a page it can cite.

Aggarwal et al. (KDD 2024) tested source-side changes on a benchmark of generated answers and found that adding statistics, quotations and citations raised a source's visibility by up to about 40%. For a firm that already publishes, the change is not to write more. It is to publish on the page, with the author named and the figures sourced, rather than in a gated PDF.

What our crawl found, and the figure a consultancy should not be in

In our audit of 9,483 small business websites (40 trades, 90 metros, late August to early September 2026), a crawler that ran no JavaScript and obeyed robots.txt fetched the homepage, the contact page and the about page of each site. The audit publishes no separate cut for consulting firms; the figures describe the whole sample, and the method section says how it was drawn.

61.5% had no question-and-answer content of any kind, and 50.7% had a heading phrased as a question with no answer a machine could isolate under it. A "frequently asked questions" block whose three answers are each "it depends" is the consulting version.

Nearly one site in five (17.8%) carried no schema.org markup; where markup existed it mostly described the website, and 13.7% of sites had business markup with its useful fields left empty. 2.1% kept their real content in a PDF. A firm whose case studies and reports sit behind a download form has, for a crawler, published nothing.

43.1% were inconsistent about who and where they are across their own three pages; 36.9% published more than one phone number, and 64.7% showed no review signal a crawler could read.

In the audit's probe, one buyer question per business put to Claude with web retrieval on named 14.9% of 9,240 businesses: one assistant, one question, one day, a direction rather than a rate. On inspection, the businesses that were named were the ones whose sites had answer-shaped content and a consistent identity.

What to fix first, in order

1. Rewrite the first paragraph of the homepage as the problem you solve and for whom, in the operator's words, with a company size.
2. One firm name, one address and one phone number, identical everywhere and in the markup.
3. One page per problem, each naming who it is for and how the engagement starts.
4. One page per partner with what each has done, as text, and a Person node linked from it.
5. Case studies and publications out of PDFs and onto pages, with authors and dates, and Article markup carrying the author's URL.
6. Organization or LocalBusiness in JSON-LD with sameAs and a Service node per problem area. Google asks for JSON-LD that states nothing a visitor cannot also read on the page.
7. The questions operators ask on a first call (how it starts, how long it runs, what it costs to begin), as headings with a real answer under each.
8. Check what a crawler receives: the crawler access check for robots.txt, and the source of a problem page with scripts off.

No consultant, ourselves included, can promise a firm a place in an assistant's answer. The work above makes the firm nameable, and a fixed panel of operator questions, asked monthly, shows whether it is being named.

Sources: our audit of 9,483 small-business sites, with its method · Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024) · Google, organization structured data · Google, article structured data · Google, introduction to structured data · schema.org, LocalBusiness · schema.org, Person · OpenAI, crawler documentation