Generative engine optimization for professional services means publishing enough clear, expert, crawlable content on your own domain that AI assistants read it, fetch it while answering buyers, and name your firm when someone asks who to hire. For law, accounting, tax, consulting and advisory firms, the work is less about technical tricks and more about surface area: many specific pages that answer the questions clients actually ask, written well enough that an assistant would rather quote you than a generic source.
The rest of this article covers how assistants treat professional services questions, what to publish, how to measure it, and the mistakes that keep firms invisible.
How is GEO different for a law, accounting or consulting firm?
Professional services buyers ask assistants questions that look like research long before they look like shopping. "How does VAT registration work for a new company?" "What is the difference between an asset deal and a share deal?" Only later do they ask: "Which firm should I hire for this?"
That sequence matters because assistants handle those two kinds of question very differently. Research questions get answered from whatever the model already knows or from authoritative sources. Selection questions are where firms get named.
The other difference is that in professional services, expertise is the product. A software company can be recommended for a feature. A firm gets recommended because its published thinking shows it understands a problem. Your content is not marketing around the service; it is evidence of the service.
When does an AI assistant actually name a firm?
On 23 September 2026 we put 20 German business questions to a model. It answered 10 of them without searching at all and cited nobody. Where it did search, it cited primary sources (laws, ministries, the OECD, large accounting firms) on 9 of 10. Firms were named only on "who should I hire" and "how do I choose" questions. And two runs of the same question named two different sets of firms.
Three practical conclusions follow.
- On informational questions, you compete with the statute and the regulator. You will not out-authority a ministry. You win by explaining application: what the rule means for a specific situation, with the primary source linked so the assistant can see you are grounded in it.
- Selection questions are where firms appear. If your site has no pages about how to choose an advisor for a given problem, what an engagement looks like, and who you work with, you give the assistant nothing to retrieve at the moment it decides whom to name.
- There is no fixed ranking. The same question produced different firms on different runs. The realistic goal is to be in the candidate pool more often, not to hold a permanent first place. Anyone promising a guaranteed citation is promising something no one controls.
| Question type | Example | What the assistant tended to do | What a firm should publish |
|---|---|---|---|
| General knowledge | "What is a holding company?" | Answered from memory, often with no search and no citation | Little to gain directly; cover it briefly as context for deeper pages |
| Rules and obligations | "When must a GmbH file its annual accounts?" | Searched and cited laws, ministries, OECD, large firms | Applied explanations that link the primary source and handle edge cases it does not |
| Selection | "How do I choose a tax advisor for a cross-border move?" | Named firms, with different firms on repeated runs | Selection guides, engagement process, scope, who you serve, credentials |
For a closer look at the selection-question side, see how consultants get cited by ChatGPT.
Why isn't good markup enough?
Many firms start with structured data, and it is worth doing. But we have seen a site with complete markup (server rendering, Article, FAQPage and Organization JSON-LD, a sitemap) and only 75 URLs stay absent from answers on its own article's exact topic. Meanwhile a site with about 190,000 crawlable URLs was crawled constantly. Surface area mattered more than markup.
A typical professional services site has a home page, an about page, team bios, six or eight practice area pages, a contact page and an occasional news post. That might be 30 URLs, most of them saying "we provide comprehensive advice" in different words. An assistant looking for an answer to a specific question finds nothing specific to retrieve.
Markup tells a crawler what a page is. It does not create the page. Treat schema as hygiene and content volume as the actual lever.
What should a professional services firm publish?
The aim is one page per real client question, written by or with people who answer that question for a living. A workable structure:
Question pages for each practice area
Take the questions prospects ask in first calls and turn each into a page. "Can I deduct home office costs as a managing director?" is a page. "Tax advice" is not. Answer in the first two sentences, then cover conditions, exceptions and what to do next.
Situation pages
Buyers describe situations, not service lines: selling a family business with a minority shareholder, hiring a first employee abroad, a founder leaving before vesting ends. Pages built around situations match how people phrase questions to assistants.
Selection and engagement pages
This is the content most firms skip and the content most likely to matter when an assistant decides whom to name. Explain how to choose an advisor for a given problem, what questions to ask, what an engagement involves, how scope and fees are typically structured in your practice, and which clients you are and are not a fit for.
Source-anchored explanations
Because assistants lean on primary sources, cite the law, regulation or official guidance you are interpreting, then add what the source does not say: practical consequences, common errors, timelines. That combination is what makes a page worth quoting.
Clear entity information
Keep your firm name, locations, practice areas and the credentials of the people writing consistent across the site. An assistant naming a firm needs to be confident what the firm is and where it operates.
How do you know whether it is working?
Not through referral traffic. In about two weeks of logs from three sites, roughly 1.2 million requests in the two busiest logs, human visitors arriving from AI assistants totalled one from Copilot and zero from ChatGPT, Perplexity, Gemini, Claude, Meta AI and others. Assistants read pages and answer in place; they rarely send a click. If you judge GEO by your analytics referral report, you will conclude nothing is happening.
The evidence is in your server logs, in two layers.
Crawlers. These read your site to build knowledge and indexes. Across the full retained nginx logs of an edtech platform we work with, from 28 July to 11 August 2026, ClaudeBot made 19,272,125 requests, Meta-ExternalAgent 3,854,076, GPTBot 2,744,985, Applebot 2,655,555 and Amazonbot 1,170,268, against 52,292 from Googlebot. AI crawlers can be far more active than the search crawler you are used to watching.
Live retrieval agents. ChatGPT-User, OAI-SearchBot and Perplexity-User fetch a page while an assistant is answering a real person. In the same fortnight, the edtech platform received 1,918 such fetches on content pages. A client blog of a few dozen articles received 56 across 24 different articles. That second figure is the more useful benchmark for a firm: a modest library, and real questions from real people still caused assistants to pull specific pages.
If you have never looked for these user agents, start with how to find AI crawlers in server logs.
What mistakes keep professional services firms invisible?
- Blocking the crawlers without knowing it. CDN bot protection, security plugins and old robots.txt rules can turn AI crawlers away. This guide on whether your site is blocking AI crawlers covers the usual causes and how to check.
- Locking expertise inside PDFs, gated downloads and client portals. The best thinking at many firms lives in newsletters and briefings a crawler never reads.
- Writing positioning instead of answers. "Pragmatic, partner-led advice" gives an assistant nothing to quote. A clear answer to a specific question does.
- Relying on a handful of practice pages. Thin sites are rarely retrieved, however polished.
- Measuring by clicks. Clicks from assistants are close to zero by design. Measure crawler reads and retrieval fetches.
- Publishing a burst, then stopping. Surface area is cumulative. A dozen articles in one month and nothing after leaves you with a small site again.
What does it cost, and who should do the work?
There are three common routes. Partners and associates can write the content themselves, which produces strong material but competes with billable time. An agency can run it, which works if they measure in server logs rather than in rankings. Or a specialised service can handle research, writing and publishing on a steady cadence.
ItGrows for Business writes and publishes research-grade articles to your own domain, structured so answer engines can quote them, and reports from your server log which AI crawlers read them and which assistants retrieved them while answering someone. It costs $499 a month (plus a one-time $499 setup, free with 3 months or a year), and it does not guarantee citations, because nobody can. For a wider view of pricing models, see how much generative engine optimization costs.
Whichever route you take, set the expectation internally: success looks like a growing library, rising crawler activity, and retrieval fetches spreading across more of your pages, not a spike in website visits.
Frequently Asked Questions
Do AI assistants ever recommend specific law or accounting firms?
Yes, but mainly on selection questions such as "who should I hire" or "how do I choose an advisor". In our test of 20 German business questions, firms were named only on that type of question, and repeated runs named different firms. Selection and engagement pages on your site are therefore the most important content to have.
Is structured data enough to get my firm into AI answers?
No. We saw a site with complete markup and only 75 URLs stay absent from answers on its own topic, while a site with about 190,000 crawlable URLs was crawled constantly. Markup helps crawlers understand pages, but the number of specific, useful pages matters more.
Why don't I see traffic from ChatGPT in my analytics?
Because assistants usually answer in place rather than sending a click. Across about two weeks of logs from three sites, human visits from assistants totalled one from Copilot and none from the others. Check server logs for retrieval agents such as ChatGPT-User and Perplexity-User instead.
Should a professional services firm compete with official sources on regulatory questions?
Not head on. When assistants searched in our test, they cited primary sources such as laws and ministries on 9 of 10 questions. Link those sources and add what they do not cover: application to specific situations, exceptions and practical next steps.