Accounting AI content strategy: the pages answer engines actually need

Most accounting firm websites were built to convert a Google visitor who already knows what a "chartered accountant" is and just needs to compare three local names. That job hasn't gone away, but a second, faster-growing job has appeared alongside it: getting cited by ChatGPT, Claude, Perplexity, Gemini, and Google's AI Overviews when a prospect asks "who should I use for R&D tax credits in Manchester" or "what does a fractional CFO cost for a 20-person SaaS company." Answer engines don't browse your site the way a human does. They pull sentences, tables, and named entities out of context and re-serve them. If your content isn't built to survive that extraction, you're invisible in the answer even if you outrank everyone on the blue links.

This is a page-by-page build plan for the content an accounting firm needs to be citable, not just findable. It assumes you already have a brochure site — service pages, an about page, maybe a blog — and you want to know what to add, cut, and restructure.

What's actually different about writing for an answer engine?

Writing for an answer engine means writing extractable, self-contained claims that make sense with zero surrounding context, because that's exactly how they'll be lifted and reused. A Google searcher lands on your page and reads it top to bottom, with your navigation, your branding, and your calls to action all doing their job in sequence. An AI engine doesn't do that. It chunks your page into passages — often paragraph-sized or smaller — embeds them, and retrieves the passage that best matches the user's question. That passage then gets summarised or quoted directly, frequently without a click-through.

Three consequences follow from this.

First, every section needs to answer its own question without depending on the paragraph before it. If your pricing page says "the same applies to the tier below" and the tier below isn't in the same chunk, the model either drops the caveat or hallucinates one. Write each section as if it's the only thing anyone will ever read.

Second, specificity is what gets selected. A sentence like "we offer competitive pricing for small businesses" is not extractable because it contains no fact. A sentence like "our fixed-fee package for limited companies with under £500k turnover starts at £185/month and includes monthly bookkeeping, VAT returns, and year-end accounts" is extractable because it's a discrete, checkable claim. Answer engines are pattern-matching against user intent phrased as a question ("how much does a small business accountant cost"), so your content needs to contain the answer in a form that maps directly onto that question.

Third, structured data does real retrieval work here, not just SEO decoration. Organization, LocalBusiness (or the more specific AccountingService where supported), FAQPage, Person, Review, and Service schema give engines a machine-readable confirmation of facts that are also stated in your visible copy. Google has said AI Overviews draw on the same index and ranking signals as regular search, including structured data; Perplexity and the ChatGPT browsing tool both favour pages with clear entity markup because it reduces the ambiguity of extraction. Schema doesn't replace good prose — it corroborates it.

The practical shift: stop writing pages that build an argument across 1,200 words toward a conclusion at the bottom. Start writing pages that state the conclusion in the first two sentences of every section, then support it.

What's the minimum content map for an accounting firm?

The minimum viable content map for AEO is seven page types: core services, sectors, locations, FAQs, proof, pricing/process, and founder/entity pages — and most firms are missing at least three of these entirely. Here's what each one needs to do and why answer engines specifically look for it.

1. Core service pages — one per distinct service, not one page trying to cover everything. "Accounting Services" as a single catch-all page is close to useless for AEO because it can't answer a specific question specifically. Split it: Bookkeeping, VAT Returns, Year-End Accounts, Corporation Tax, Payroll, R&D Tax Credits, Company Formation, CFO/Advisory. Each page should open with a one-sentence definition of the service in your context (not a dictionary definition — what it means for the client), followed by who it's for, what's included, what it costs or how cost is determined, and how long it takes.

2. Sector pages — built for firms with genuine vertical specialism, not fabricated for SEO volume. If you actually do a lot of work with dental practices, e-commerce sellers, or contractors, a sector page lets you answer "accountant for [industry]" queries with specifics an answer engine can quote: VAT treatment quirks, common allowable expenses, sector-specific deadlines. A generic "Accountants for Small Business" page that could apply to any firm anywhere is not a sector page — it's filler.

3. Location pages — one per office or clearly defined service area, never duplicated by swapping the city name in a template. This is the single most common AEO failure mode in accounting: fifty near-identical "Accountants in [Town]" pages differentiated only by a find-and-replace. Answer engines detect this pattern (so does Google) and either suppress the whole set or cite none of them. A real location page names the office address, the team based there, the local clients you serve, and anything genuinely regional (a devolved tax quirk, a local business rates scheme, a Companies House filing office reference).

4. FAQ content — structured as actual question-and-answer pairs with FAQPage schema, addressing the literal phrasing prospects use in chat interfaces. This is the highest-leverage page type for answer engine citation because the format already matches how users query. "How much does it cost to file a confirmation statement?" "Do I need an accountant for a dormant company?" "What's the penalty for late VAT registration?" Pull these from your actual client email threads and sales calls — the questions people really ask, not the questions you wish they asked.

5. Proof pages — case studies, client outcomes, credentials, and verifiable third-party validation. This includes your ICAEW or ACCA membership number and registration status, your Companies House filing history if you're demonstrating longevity, and specific, attributable client results ("reduced a client's corporation tax liability by identifying £42,000 in unclaimed capital allowances" beats "we save clients money"). Answer engines increasingly weight pages that carry verifiable entities — professional body registration numbers, named reviewers, dated case studies — over pages making the same claim with no attribution.

6. Pricing and process pages — the page type accounting firms most often omit, and the one that answer engines are asked about constantly. "How much does an accountant cost for a limited company" is a top-volume conversational query across every engine we've tested. If you don't publish a number, a range, or a clear "it depends on X, Y, Z" framework, the engine will cite a competitor who did, or worse, cite a third-party average that has nothing to do with your market. Pair pricing with a process page: what happens in week one, what you need from the client, what the onboarding timeline looks like.

7. Founder and entity pages — a full profile for each named partner or director, not a two-line bio buried on a team page. Answer engines resolve entities (people, firms, credentials) against structured knowledge, and a thin bio gives them nothing to resolve against. Include full name, professional qualification (ACA, ACCA, ACMA CGMA, CTA), regulatory registration, years in practice, specific specialisms, and — where you have it — a Person schema block and a consistent presence across LinkedIn, Companies House officer filings, and professional body directories. Consistency across these sources is what builds what's sometimes called an entity signal — the model's confidence that "Sam Hoye ACMA CGMA" and "Sam Hoye, director, XYZ Accountants Ltd" are the same verified person.

If you have all seven page types built out with real specificity, you have a content base an answer engine can actually cite. Most brochure sites have services and an about page and nothing else on this list.

How should each page be structured for direct answers and citation?

Every page needs a lead-in answer, a scannable structure, and machine-readable schema — in that order — because that's the sequence in which an answer engine evaluates whether to use you as a source. Structure isn't a nice-to-have layer on top of good content; for AEO it's roughly half the work.

The lead-in answer. The first 40–60 words under any H1 or H2 should state the direct answer to the implicit question, in plain declarative sentences, before any scene-setting. If the page is "VAT Registration for Contractors," the first sentence should be something like: "Contractors must register for VAT once taxable turnover exceeds the current HMRC threshold, and many choose to register voluntarily earlier to reclaim input VAT on equipment and subcontractor costs." Not: "Navigating VAT as a contractor can feel overwhelming, but with the right guidance..." That sentence contains zero extractable fact and answer engines skip past it.

Heading structure as a query map. Use H2s that are themselves phrased as questions, matching how someone would type into ChatGPT or ask Siri or Alexa: "How much does year-end accounts preparation cost?" "Do sole traders need a separate business bank account?" "What happens if I miss the Companies House filing deadline?" This isn't keyword stuffing — it's building your page's table of contents to mirror the actual query space, which makes each section independently retrievable for its matching question.

Answer-first paragraphs, evidence second. Within each section: state the answer in sentence one, then the supporting detail, then the nuance or exception. Reverse that order — nuance first, answer buried at the end — and you're optimising for a human reading in order, at the cost of being unextractable to a model grabbing the top sentence of a chunk.

Tables for anything comparative or numeric. Pricing tiers, deadline calendars, allowance thresholds, service comparisons — put them in an actual table, not a paragraph describing a table. Tables are cleanly parseable and get pulled into AI Overviews and Perplexity's structured answer boxes disproportionately often relative to their share of on-page content.

Schema markup matched to page type. Service pages get Service schema. FAQ pages get FAQPage with each Q&A pair marked up individually. Founder pages get Person with alumniOf, hasCredential, and worksFor properties populated. Location pages get LocalBusiness with full address, geo, and openingHours. Reviews get Review or AggregateRating, only ever with real, attributable review data — never fabricated star ratings, which several firms have been caught doing and which both Google and the platforms hosting genuine reviews (Google Business Profile, Trustpilot) actively police.

A named source for every regulatory claim. Any statement about tax rates, thresholds, filing deadlines, or penalties should cite HMRC, Companies House, or the relevant professional body (ICAEW, ACCA) by name and, where practical, link to the primary source. This does two things: it gives the answer engine a verifiable anchor to check the claim against, and it protects you when rates change and someone finds your page months later with outdated numbers still stated as current.

Freshness signals that are actually true. A visible "last reviewed" date, updated when you genuinely review the page against current HMRC guidance — not a date field that auto-updates on every unrelated CMS save. Answer engines that weight recency (this matters most for tax rate and threshold content, since these change every tax year) will deprioritise pages showing stale dates next to time-sensitive claims.

Put together, a well-structured service page reads almost like a briefing document: direct answer, then a question-headed breakdown, then a pricing table, then a named-source callout, then an FAQ block, then the schema doing the same job invisibly underneath.

What should you not publish?

Don't publish thin blog posts, duplicated location pages, generic AI-written advice, or claims you can't support — these four categories actively damage your AEO position rather than merely failing to help it. This matters because "just publish more content" is the instinct most firms default to, and for answer engines specifically, volume without substance is a negative signal, not a neutral one.

Thin blog posts. A 400-word post titled "5 Tips for Managing Your Business Finances" that could have been written by any firm, about any business, in any country, adds nothing an answer engine will ever cite over a more specific competitor page — and it dilutes your site's overall topical authority signal. If you're going to blog, write posts that only your firm could have written: a breakdown of a specific HMRC consultation and what it means for your actual client base, a walkthrough of a real (anonymised) client scenario, an analysis of a sector-specific change. If it reads like it was assembled to hit a content calendar quota, don't publish it.

Duplicated city/location pages. Fifty pages reading "Looking for an accountant in [TOWN]? [FIRM NAME] provides expert accounting services throughout [TOWN] and the surrounding area" with the town name swapped are the clearest AEO own-goal in the industry. These pages don't just fail to earn citations — they can suppress your entire site's crawl priority and, in Google's case, have historically triggered manual actions for doorway pages. If you serve twelve towns from one office, you need one strong location page for that office listing all twelve service areas, not twelve thin pages pretending to be twelve offices.

Generic AI-written advice with no firm-specific detail. Content generated by dropping a keyword into a general-purpose model and publishing the output unedited is now extremely common, and answer engines are the first tools with a direct incentive to detect and deprioritise it — they're competing with it for the same query space and have no reason to cite an unattributed rehash of the same public information they already have access to. If your R&D tax credits page reads exactly like the fifteen other firms' R&D tax credits pages because everyone used the same prompt, you have zero differentiation and zero citation advantage. The fix isn't to avoid AI drafting tools — it's to never publish a draft without adding a specific, checkable, firm-only fact: a real number from real client work, a specific process detail, a named person's actual view.

Unsupported claims. "We're the leading accountancy firm in [region]" with no ranking source, "clients save an average of £X" with no methodology, "99% client satisfaction" with no linked survey — these are exactly the claims answer engines are least likely to repeat, because they can't verify or attribute them, and repeating an unverifiable claim carries reputational risk for the platform. Worse, if a competitor's page makes a similar claim with a linked source, theirs wins the citation. Every superlative or statistic on your site should have a visible, named source or be rewritten as a specific, attributable fact.

The through-line: answer engines reward pages that could only have been written by you, about your actual practice, supported by checkable facts. They penalise — actively, not just passively — anything that reads as templated, generic, or unverifiable.

What's the right publishing order if you're starting from a brochure site?

Start with pricing, FAQs, and founder pages before touching sectors or blog content, because those three page types close the biggest citation gaps fastest and require no new research. Sequencing matters because most firms have limited content production capacity and need to spend it on the pages with the highest citation return first, not the ones that feel most natural to write.

Phase one (weeks 1–4): fill the direct-answer gaps. Publish or rebuild your pricing/process page with real numbers or a clear framework. Build a genuine FAQ page (15–25 real questions, FAQPage schema) sourced from your actual sales calls and support emails. Expand your team page into individual founder/partner pages with full credentials and Person schema. These three moves address the queries prospects are asking answer engines *right now* — cost, process, and "who am I even dealing with" — and they're built from information you already have; no new research or client work required.

Phase two (weeks 5–8): rebuild core service pages one at a time. Take your existing service pages and rewrite each one against the structure above: direct-answer opener, question-headed H2s, comparison or pricing tables, named regulatory sources, service-specific schema. Don't try to do all of them simultaneously — pick your three highest-inquiry services first (for most firms: year-end accounts, VAT, and either payroll or R&D credits) and get those fully rebuilt before moving to the rest.

Phase three (weeks 9–12): sector pages, but only where you have real specialism. Audit your actual client base. If 30%+ of clients sit in one identifiable sector, build a real sector page for it with sector-specific detail. If your client base is genuinely generalist, skip this phase entirely — a fabricated sector page is worse than no sector page.

Phase four (ongoing): location pages, done properly. One page per real office, each with a full local profile, not a templated swap. If you're a single-office firm, spend this budget instead on proof content — case studies and outcomes — rather than manufacturing location pages for towns you don't have a physical presence in.

Phase five (ongoing): earned-authority blog content. Only after the structural pages exist. Blog posts at this stage should target the specific, narrow questions your FAQ page revealed prospects are asking but that don't warrant a standalone service page — regulatory change explainers, real client scenarios, sector commentary tied to your actual specialism.

What to measure at each phase. Track whether your firm is being surfaced (and correctly represented) when you or a colleague query ChatGPT, Perplexity, and Google AI Overviews directly with the phrases your FAQ page targets. This is manual and unscientific compared to rank tracking, but it's currently the most direct signal available, since none of the major answer engines publish a citation-tracking console equivalent to Google Search Console. Re-check monthly through Q3 2026 as engines update their retrieval indexes, and treat a citation you didn't have last month as the leading indicator that a specific page rebuild worked.

A firm that completes phases one and two properly — pricing, FAQs, founder pages, and three rebuilt service pages — will typically see more answer-engine visibility improvement in a single quarter than a firm that spends the same effort publishing twenty new generic blog posts. The gap isn't about volume. It's about whether each page contains a specific, attributable, structurally extractable answer to a question a real prospect is actually asking a real answer engine.