AEO and GEO, How Accounting Firms Get Cited by AI

Last reviewed: September 2026

What is the difference between SEO, AEO and GEO?

SEO optimises a page so it ranks in a list of links. AEO structures the content on that page so an assistant can lift a direct answer out of it. GEO builds the signals that make a model name and recommend the firm rather than merely quote it. Same content operation, three different definitions of winning.

The reason this matters commercially is that the click is no longer the unit of success. Pew Research Center, analysing the browsing of 900 US adults across 68,879 Google searches in March 2025, found users clicked a traditional search result in 8% of visits where an AI summary appeared, against 15% where none did. Clicks on a source link inside the summary occurred in 1% of visits. That is United States data, not Australian, so treat the direction as instructive and the figures as imported.

Does an accounting firm need a separate AEO budget?

No, and the strongest evidence comes from Google itself. Its AI features and your website documentation, last updated 10 December 2025, states that there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary. It goes further: you do not need to create new machine readable files, AI text files, or markup, and there is no special schema.org structured data that you need to add.

That single paragraph deflates most of what is sold to Australian practices as an AEO package. What is genuinely different is not the technology, it is the shape of the writing. A page built to hold attention for four minutes and a page built to yield a clean forty-word answer are not the same page, and most accounting firm websites are neither.

The organisational gap is the real finding. Forrester principal analyst John Buten, writing on 19 May 2026 in Stop Replacing Traffic, Start Replacing Visibility, reported that 70% of marketers said AI visibility is a top priority for their CMO or CEO, while only 30% of companies have defined a discrete owner for answer-engine visibility. That survey covers Forrester's B2B marketing community rather than Australian accounting firms specifically, but the pattern will be familiar to any practice principal: everyone agrees it matters, nobody's name is next to it.

What should an accounting firm publish to get cited by AI?

Publish the questions clients actually ask, phrased the way they ask them, with the answer in the first forty to sixty words. This sounds trivial. It is the single change that separates a firm's content from the undifferentiated mass, because almost every accounting website answers the question it wishes clients asked instead.

Australian bookkeepers and BAS agents hear the real questions every week: can I claim my ute, why is my BAS different this quarter, do I need to register for GST yet, what happens if I lodge late. Those exact strings get typed into ChatGPT. A page titled "Taxation Advisory Services" answers none of them. A page titled "Do I need to register for GST yet?" with a direct answer, a threshold, a date and a source answers one completely, which is what an answer engine needs in order to quote anything.

The second half is harder and matters more. Answer engines synthesise consensus, so content that restates consensus adds nothing worth citing. A firm needs at least one thing on its site that no other firm could have written: an anonymised distribution of errors found across the client book, a dated position on a contested change, a worked example with real numbers from a real engagement. This is the same reason a practitioner's genuine position outperforms a summary; the argument runs through the skills shortage post in this series, where the differentiator turns out to be judgement rather than throughput.

Does FAQ schema and structured data still work in 2026?

FAQ schema no longer produces anything visible in Google. Google's FAQPage structured data documentation carries a deprecation notice reading: "This feature will no longer appear in Google Search starting May 7, 2026." Any agency still selling FAQ markup to an accounting firm on the promise of expanded search results is selling a 2019 product.

Keep the markup anyway, for a smaller and more honest reason: it labels a question and its matching answer as a pair for any system parsing the page, and costs nothing once templated. Just do not budget for it as a visibility line item.

The schema that does still earn its place is entity markup, and here most advice given to accounting firms is simply wrong. There is no schema.org type called Accountant. The correct type is AccountingService, defined as an accountancy business, sitting at Thing > Organization > LocalBusiness > FinancialService > AccountingService. Because it inherits from LocalBusiness, a practice gets address, telephone, opening hours and area served without inventing anything. Add Organization markup for the firm, and author markup that names the practitioner with their actual registration and qualifications rather than "the team".

Why do your firm's details need to match everywhere?

Because retrieval systems resolve an entity by corroboration, not by assertion. A model deciding whether a given Ballarat practice is real, current and findable is weighing agreement between independent sources. When the website says one trading name, the Google Business Profile says another, LinkedIn lists a suburb the firm left in 2023 and three directories carry a disconnected landline, every source undermines the others, and the safest thing a model can do is name a competitor whose story is boringly consistent.

Google's Business Profile guidelines put the same requirement in plainer language: a business name should reflect the business's real-world name, as used consistently on its storefront, website, stationery, and as known to customers. For an Australian practice the corroboration set is specific and short: the website, the Google Business Profile, LinkedIn, the ABN Lookup record, the professional body directory listing, and whichever review platforms already carry entries. Reconciling those is an afternoon of tedium that no one volunteers for, which is exactly the kind of task the marketing skill exists to absorb.

What did Agentive learn running this playbook on its own site?

Three structural changes on agentive.au held up, and one did not. The three that held: every article opens with a short, self-contained direct answer before any preamble; every article carries a structured question and answer set emitted as FAQPage JSON-LD from the article's own source file rather than bolted on by a plugin; and robots.txt explicitly names and admits the assistant crawlers, including OAI-SearchBot, GPTBot, PerplexityBot, ClaudeBot and Google-Extended, alongside a Content-Signal declaration. That last one is not an optimisation so much as removing a self-inflicted block, and it is worth checking, because plenty of firms have a security-minded webmaster who quietly denied all of them.

The honest negative: Agentive published an llms.txt file at agentive.au, and cannot show that it moved anything. It was an afternoon of work on a proposed convention that no major engine has committed to reading, and Google's AI features documentation now says outright that you do not need to create new machine readable files or AI text files to appear in its AI experiences. The file is still there because removing it costs more than leaving it. Anyone quoting llms.txt as a reason to sign an AEO retainer is quoting a hypothesis.

A second observation worth stating plainly, from building this category of software for Australian finance teams: the constraint on AEO in a small practice is never the technique, it is that nobody writes the answers. Schema is a template change made once. Forty genuinely useful answers is forty acts of writing by people who bill their time, which is where the work stalls in nearly every firm.

How do you measure AI visibility for a practice?

Use three instruments, because no single one is complete. First, Google Search Console's generative AI performance report, which covers AI Overviews and AI Mode and rolled out to all sites by 31 August 2026. It breaks down by pages, countries, dates and devices, and the limitation matters: there is no query dimension, so it tells a firm which pages are being surfaced but not which questions triggered it.

Second, referral traffic. OpenAI's crawler documentation distinguishes OAI-SearchBot, which surfaces sites in ChatGPT's search features and is not used for model training, from GPTBot, which crawls for training, and ChatGPT-User, which fetches pages when a user asks. Allowing OAI-SearchBot is the prerequisite for appearing at all, and assistant referrals arrive tagged in analytics, so they can be segmented rather than guessed at.

Third, and the one nobody wants to run: a manual prompt panel. Write down the twenty questions the firm wants to be the answer to. Once a month, ask each of ChatGPT, Gemini, Perplexity and Copilot, and record whether the firm was cited, named, or absent. It takes about ninety minutes and is the only instrument that actually measures GEO. Assistants personalise and vary, so treat it as a trend line, not a scoreboard. The same discipline of logging what was asked and what came back applies inside the practice too, which is the argument made in the post on hallucinations in financial reporting.

The 10-step AEO and GEO checklist for this quarter

1. Name the owner. One person in the practice, by name, accountable for AI visibility. Forrester's 70/30 gap is entirely this step. Everything below is unallocated until it is done.
2. Write the twenty questions. Pull them from your inbox and your file notes, not from a keyword tool. Use the client's wording, including the wording that makes you wince.
3. One question, one page. Stop bundling six topics into a services page. An answer engine cannot cite a page that is about four things.
4. Put a 40 to 60 word answer at the top of every page. Complete sentences, no dependence on the paragraph before, no throat-clearing. This is the block that gets lifted.
5. Fix the entity record. Reconcile name, address, phone, ABN and service description across the website, Google Business Profile, LinkedIn, ABN Lookup and every directory listing you can find.
6. Add AccountingService and Organization markup. Not "Accountant", which does not exist. Include author markup naming the practitioner and their registration.
7. Check robots.txt admits the assistants. Confirm OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended are not blocked, then decide deliberately, rather than discovering a default.
8. Publish one thing only your firm could write. A dated position, an anonymised pattern from your client book, a worked example with real numbers. One is enough to start.
9. Turn on the measurement. Open the Search Console generative AI report, segment assistant referrals in analytics, and diarise the monthly prompt panel.
10. Review in ninety days. Compare the panel against the baseline you recorded in step nine, then rewrite the three weakest answers rather than adding three more pages.

One compliance note, because this is an accounting practice and not a marketing agency. Any answer published under a registered practitioner's name is still professional communication. If AI drafted it, the Tax Practitioners Board's expectations about verification and accuracy apply before it goes live, which is covered in the guidance on what tax and BAS agents must disclose. Publishing a wrong threshold at speed is worse than publishing nothing.

The firms that get cited are the ones that answered

There is no mechanism here a competent Australian accounting practice cannot run itself. Answer the real questions directly, keep the firm's facts consistent, say something only you can say, let the crawlers in, and measure on a schedule. The acronyms are doing far more work in sales decks than in search results.

What stops most practices is capacity, not knowledge, which is where a dedicated AI changes the arithmetic: Agentive runs single-tenant on AWS Sydney with all inference inside Australia and client data never used to train a model, and the same AI Operation Engine that reconciles a ledger will draft twenty answers for a principal to correct in an hour. The same pattern is set out for accounting practices elsewhere on this site, and for pipeline work in the HubSpot CRM playbook. Start with step one. The owner's name is the whole exercise.

This post is general information for Australian practices and is not professional or legal advice. Search and AI features change frequently; verify current behaviour against the primary sources linked above before acting.