Accounting AI content strategy: the pages answer engines actually need

Accounting AI content strategy: the pages answer engines actually need

Most accounting firm websites are built to be browsed. Answer engines don't browse — they extract. ChatGPT, Claude, Perplexity, and Google's AI Overviews scan a page for a self-contained answer, lift the sentence or paragraph that satisfies the query, and cite (or paraphrase) it. If your page requires a reader to scroll through a company history and a stock photo of a handshake before reaching the actual answer, the engine skips it and cites a competitor instead.

This is a practical build order: what pages an accounting firm needs, how each one should be structured so an AI system can extract it cleanly, what to stop publishing, and what order to publish in if you're starting from a standard brochure site in Q3 2026.

What's the difference between writing for humans, Google, and answer engines?

Writing for humans optimises for persuasion and flow; writing for Google optimises for keyword coverage and backlink-worthy depth; writing for answer engines optimises for extractable, self-contained, verifiable statements. These are three different jobs, and most firm websites only do the first one — badly.

A human-first page opens with a story ("Founded in 1987 in a small office above a bakery...") and works toward the point. A Google-first page front-loads a keyword-rich H1 and builds topical depth to rank for a head term. An answer-engine page does something different: it states the direct answer in the first one to three sentences, in plain declarative language, with a number, a name, or a scoped claim the model can quote without needing the rest of the page for context.

Concretely, if a prospective client asks Perplexity "how much does a limited company accountant cost in Manchester," the winning page isn't the one with the best brand voice. It's the one with a sentence like: "Limited company accountancy packages in Manchester typically range from £120–£250 per month for a business with turnover under £250k, covering annual accounts, corporation tax, and payroll for up to two directors." That sentence is a complete, self-contained answer. It has a number, a scope, and a service list. An LLM can extract it verbatim and attribute it. A paragraph of brand narrative cannot be extracted the same way — there's nothing crisp to lift.

This doesn't mean abandoning brand voice or SEO structure. It means every page needs an answer-engine layer sitting on top of (or ahead of) the persuasive and keyword layers: a direct-answer opening, a scannable structure, and named specifics an engine can verify or attribute.

What is the minimum content map for an accounting firm?

The minimum viable content map for AI visibility has seven page types: core services, sectors served, locations, FAQs, proof (case studies and credentials), pricing/process, and a founder/entity page — and most firms are missing at least three of them.

Here's what each type does and why answer engines specifically need it.

1. Core service pages. One page per distinct service — self-assessment, corporation tax, VAT returns, payroll, R&D tax credits, management accounts, bookkeeping, company formation. Not a single "Services" page with six paragraphs stacked on top of each other. Answer engines match query intent to a specific page; a query about "R&D tax credit accountant" needs an R&D-specific URL to cite, not a subsection three-quarters down a generic services page.

2. Sector pages. Contractors, e-commerce sellers, hospitality, dentists, tech startups, property landlords — whichever sectors you actually serve, each gets a page addressing the tax and compliance issues specific to that sector (IR35 for contractors, VAT flat rate scheme for freelancers, CIS for construction). Generic advice doesn't get cited when a sector-specific query is asked; sector-specific advice does.

3. Location pages — but only where there's a real basis for them (see the "what not to publish" section below). A genuine location page answers "is there an accountant near me in [place]" with actual local detail: office address, local client examples, regional scheme knowledge (e.g., local authority business rates relief).

4. FAQ pages and FAQ blocks. This is the single highest-leverage format for answer engines because the question-answer structure mirrors exactly how users prompt ChatGPT and Perplexity. "Do I need an accountant for a limited company?" "How much does a sole trader accountant cost?" "Can I switch accountants mid-year?" Each question should be its own heading with a direct answer in the first sentence beneath it, marked up with FAQPage schema.

5. Proof pages — case studies, testimonials with attribution, credentials (ACCA, ICAEW, ATT, CTA membership numbers where applicable), and named client outcomes. Answer engines increasingly weight E-E-A-T signals (experience, expertise, authoritativeness, trust) when deciding which source to cite for anything touching financial advice — a YMYL (your-money-or-your-life) category Google treats with elevated scrutiny, and the same trust logic shows up in how LLMs weigh sources during retrieval-augmented generation.

6. Pricing and process pages. "How much does it cost" and "what happens after I sign up" are two of the most common pre-sale questions typed into AI chat interfaces because people find pricing pages evasive and want a fast, comparable answer. A firm willing to publish real price ranges and a step-by-step onboarding process gets cited over one that hides pricing behind a "contact us for a quote" form.

7. Founder/entity pages. A named partner or director page with credentials, specialisms, and a headshot does two jobs: it satisfies E-E-A-T by attaching real expertise to the content, and it gives answer engines an entity to anchor citations to ("according to [Name], [Firm], a Manchester-based ACCA practice..."). Firms with no named experts on the site are functionally anonymous to an AI system trying to assess trust.

Miss any of these seven categories and you've left a query type uncovered — which means a competitor's page fills that gap in the citation instead of yours.

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

Every page an answer engine should be able to cite needs four structural elements: a direct-answer opening, a scannable body with descriptive subheadings, at least one specific and verifiable data point, and schema markup that machine-tags the content type.

The direct-answer opening. The first 40–60 words under any H1 or H2 should answer the implied question without requiring the reader to scroll. Write it as if it will be read completely out of context — because on an AI Overview or a Perplexity citation card, it often is. Avoid throat-clearing ("When it comes to choosing an accountant, there are many factors to consider..."). State the answer, then the caveats.

Descriptive, question-shaped subheadings. Use H2s and H3s phrased the way a user would type a query: "How much does payroll outsourcing cost for a 10-person company?" not "Payroll Pricing." This does double duty — it improves featured snippet eligibility on Google and gives LLMs a clean semantic match against natural-language prompts.

Specific, verifiable data points. "We help lots of small businesses save money on tax" is unusable to an answer engine because it can't be verified or attributed with confidence. "We filed R&D tax credit claims for 34 clients in the 2025/26 tax year, with an average claim value of £28,400" is a fact an engine can cite, and it's checkable against your own records — which matters, because engines are increasingly cautious about amplifying unverifiable claims from financial services sites. If you don't have hard numbers yet, use a clearly flagged illustrative range rather than inventing a false precision.

Schema markup. FAQPage schema on FAQ content, Service schema on service pages, LocalBusiness and Accountant schema on location and homepage content, Person schema on founder pages, and Organization schema site-wide with sameAs links to your Companies House filing, ICAEW/ACCA directory listing, and LinkedIn. Schema doesn't guarantee a citation, but it removes ambiguity about what type of content the page is and who stands behind it — which is exactly the disambiguation step retrieval systems perform before deciding whether to trust a source.

A useful internal test: copy the first three sentences of a page's main content block into a blank document, with no heading and no URL attached. Can a stranger tell what the page answers and who wrote it? If not, restructure before publishing.

What shouldn't accounting firms publish?

Four categories actively damage AI visibility rather than helping it: thin generic blogs, duplicated city pages with swapped place names, unsupported or AI-boilerplate advice, and claims without attribution.

Thin blogs with no unique input. A 400-word post titled "5 Tax Tips for Small Businesses" that restates publicly available HMRC guidance in slightly different words adds no retrievable value — HMRC's own site, or a bigger competitor's more detailed version, will out-rank and out-cite it every time. If a blog post doesn't contain a number, a named case, a specific regional rule, or a professional judgment call that only your firm would make, don't publish it.

Duplicated location pages. The classic local-SEO trick — one template with "[City]" swapped 40 times — is now actively penalised by Google's spam policies on scaled content abuse, and it's useless to answer engines because there's no unique signal distinguishing the Leeds page from the Bradford page from the Wakefield page. Only build a location page where you have a genuine local presence: an office, a documented client base in that area, or specific knowledge of a local scheme.

Generic AI-written advice with no professional layer. Content that reads like it was generated from a prompt ("Managing your business finances effectively is crucial for success...") and never gets reviewed or annotated by a qualified accountant is now easy for both search engines and readers to detect, and it actively erodes the E-E-A-T signal the rest of your site is trying to build. If you use AI tools to draft, the published version needs a named professional's judgment, a specific example, or a caveat that only comes from practice experience — otherwise it's indistinguishable from the thousands of other AI-drafted accounting blogs competing for the same query.

Unsupported claims. "Award-winning," "trusted by hundreds of businesses," "the UK's leading accountancy firm for startups" — these phrases carry no evidentiary weight and answer engines generally won't cite them because there's nothing to verify. Replace every unsupported superlative with either a number you can stand behind or a specific, checkable fact (a named award, a client count with a date range, a Companies House-verifiable years-trading figure).

The pattern across all four: answer engines are built to route around ambiguity and reward specificity. Anything generic, duplicated, or unverifiable is filtered out before it gets anywhere near a citation.

What order should a firm publish in, starting from a brochure site?

Publish in five phases over roughly two quarters: foundation and schema first, then core service pages, then FAQ and proof content, then sector and location pages, then a maintenance and refresh cycle — because entity and structural fixes compound in value with everything published after them.

Phase 1 — Foundation (weeks 1–3). Fix the entity layer before writing anything new. Add Organization schema site-wide, create or update the founder/partner page with named credentials, verify your Companies House and ICAEW/ACCA/ATT/CTA directory listings match your site exactly (name, address, registration numbers), and set up a Google Business Profile if one doesn't exist. This is unglamorous and produces no immediate content, but every page published afterward inherits this trust layer. Skipping this step and going straight to content is the most common mistake — firms publish 20 excellent service pages on top of an unverifiable, anonymous entity and wonder why nothing gets cited.

Phase 2 — Core service pages (weeks 3–8). Rebuild or split your services into one dedicated page per service, each with a direct-answer opening, pricing ranges, and Service schema. Prioritise by query volume and margin: the services clients most often search for and the ones most profitable to win. For most small-to-mid UK and US firms that's self-assessment, limited company/corporation tax accounting, VAT, payroll, and bookkeeping, with R&D tax credits or fractional CFO services added if that's a real specialism.

Phase 3 — FAQ and proof (weeks 8–12). Build a comprehensive FAQ page with FAQPage schema, drawing questions directly from what clients actually ask in discovery calls and email threads — not guessed keywords. In parallel, publish two to three detailed case studies with named (or anonymised-but-specific) outcomes and numbers, and a pricing/process page that states real ranges and real steps. This phase is where most of the citation-worthy content actually lives, because FAQs and proof pages map most directly onto how people prompt AI assistants.

Phase 4 — Sector and location pages (weeks 12–20). Only now, once the foundation and service layer exist, add sector pages for genuine specialisms and location pages for genuine physical or client presence. Each sector page should cross-link to the relevant service pages and cite sector-specific rules (IR35, CIS, flat rate VAT scheme thresholds) by name.

Phase 5 — Maintenance and refresh (ongoing from week 20). Answer engines favour freshness signals — updated dates, current tax-year figures, current thresholds. Set a quarterly review cycle to update every page with numbers tied to a tax year (allowances, rates, thresholds) as soon as HMRC or the IRS publishes updated figures each year. A page citing the wrong tax year's personal allowance is worse than no page at all, because it damages trust in everything else on the domain.

This order matters because answer engines and AI crawlers assess a domain holistically, not page by page in isolation — a strong entity and service foundation raises the retrieval confidence of every page published afterward, while a content-first, foundation-last approach means good pages get built on an unverifiable base and underperform relative to their actual quality.

What should firms measure to know if this is working?

Track citation appearances in AI answers, not just organic rankings — because a page can rank on page one of Google and still never get pulled into a ChatGPT or Perplexity answer, and the two need to be measured separately.

Run a monthly manual check: take your ten highest-priority queries (the ones a prospective client would realistically type into ChatGPT or ask Perplexity — "best accountant for e-commerce sellers in [region]," "how much does self-assessment cost with an accountant") and check whether your firm is cited, and in what form (direct link, named mention without link, or absent). Track this alongside standard Google Search Console data on impressions and click-through for the same queries. A firm that shows up in Google's top three but never in an AI Overview or Perplexity citation for the same query has a structural extraction problem, not a rankings problem — which almost always traces back to missing schema, a buried direct answer, or an unverifiable claim on the page.

The firms that will be cited by answer engines through the rest of 2026 aren't necessarily the ones with the most content. They're the ones whose content is structured so a machine can lift a clean, specific, verifiable answer without having to guess who wrote it or whether to trust it.