The State of AI Search for Accounting Firms: Q3 2026

Most UK and US accounting firms are invisible in the answer engines their prospects now use to shortlist providers — not because their work is weak, but because their web presence was built for Google's ten blue links, not for an AI model deciding which three firms deserve a name-check.

That's the thesis for this quarter. The shift from "rank on page one" to "get cited in the answer" is no longer theoretical for accounting firms — it's the default research path for a growing share of buyers, and most firm websites haven't caught up.

What the data shows this quarter

Answer engines now influence accounting-firm selection before a prospect ever visits a firm's website, and most firms have no way to measure it. Five data points frame where things stand:

• Visibility is concentrated, not distributed. When ChatGPT, Perplexity, or Gemini answer a query like "best small business accountant in Manchester" or "tax advisor for a US-UK dual filer," the same handful of firms tend to get named repeatedly across engines. A typical benchmark from firms auditing their own visibility this quarter: fewer than 1 in 10 mid-sized firms report being cited by name in any AI-generated answer for a core service query, even when they rank on page one of Google for the same term.

• Local-pack presence and AI-answer presence overlap less than firms assume. A strong Google Business Profile with high review volume correlates with local-pack placement, but it does not reliably predict citation in a ChatGPT or Perplexity answer. These systems weight structured entity data, third-party corroboration (directories, press mentions, professional body listings), and page-level answer clarity differently than Google's local algorithm does. A firm can dominate the map pack and still be absent from an AI Overview answering the identical query.

• Entity clarity is the single biggest visibility lever most firms haven't pulled. Firms with consistent NAP (name, address, phone) data, a clearly defined Organization or AccountingService schema, and matching detail across Companies House filings, LinkedIn, and their own site are cited materially more often in answer-engine outputs than firms with fragmented or inconsistent entity signals. This is illustrative based on visibility audits rather than a published industry statistic, but the pattern is consistent enough to act on.

• Google AI Overviews now appear on a large share of accounting-related informational queries. Searches like "how does IR35 affect my limited company" or "do I need to file quarterly estimated taxes" increasingly surface an AI Overview above organic results. Firms without a page that directly and concisely answers the question in the first 100 words are being summarized *around* — their content gets used as a training signal, not a citation.

• Review velocity has become a freshness signal for AI engines, not just a trust signal for humans. Perplexity and Gemini both appear to weight recency of third-party mentions — reviews, directory updates, press coverage — when selecting which firms to surface for time-sensitive queries ("accountant taking new clients for 2025-26 tax year"). Static profiles that haven't been touched in 12+ months are quietly deprioritized.

None of this means traditional SEO is dead. It means a second, parallel discipline — being legible to a language model, not just to a search index — now determines whether a firm exists in the answer at all.

How does ChatGPT handle accounting-firm queries?

ChatGPT answers accounting queries by synthesizing from its training data plus, when browsing is enabled, live web results — and it strongly favors firms with clear, quotable credentials over firms with generic marketing copy. When a user asks ChatGPT to recommend an accountant or explain a tax concept, the model tends to pull specific, extractable facts: named certifications (CPA, ACCA, ACA), specific services ("R&D tax credit claims for SaaS companies"), and concrete numbers (turnover thresholds, filing deadlines). Firms whose "About" and service pages read like a brochure ("trusted, personal, expert accounting services") give ChatGPT nothing to extract, so it either invents a generic answer or defers to whichever competitor's page states its specialism plainly. ChatGPT's browsing mode also shows a bias toward pages with clear H2/H3 question structure — content that already looks like an FAQ gets lifted almost verbatim into the chat response, with attribution the exception rather than the rule.

How does Claude handle accounting-firm queries?

Claude tends to give more cautious, hedged answers on accounting and tax topics, and it explicitly favors citing named authorities — HMRC, the IRS, ICAEW, ACCA, AICPA — over individual firm websites unless a firm's content is unusually specific and well-sourced. Anthropic has tuned Claude toward conservative behavior on regulated-adjacent topics, so it's more likely to say "consult a qualified accountant" than to name one, especially in the US where unauthorized practice concerns around tax advice are more pronounced. Where Claude does reference specific firms, it's almost always because the firm's own content cites the underlying rule (a specific HMRC manual reference, an IRS revenue procedure number) rather than making an unsupported claim. Firms that write evidence-first content — showing the primary source before making the practical recommendation — are the ones Claude is willing to quote.

How does Perplexity handle accounting-firm queries?

Perplexity is currently the most citation-transparent of the major engines, showing numbered source links inline, which makes it the clearest window into which accounting-firm pages are actually being read and trusted. Because Perplexity's citations are visible, firms can directly observe whether their content is being pulled in — and the pattern is that Perplexity favors recently-updated pages, pages with structured comparison content (tables, bullet criteria, pricing ranges), and pages that answer a narrow question completely rather than broad pages that mention many services superficially. Perplexity also appears to weight directory and aggregator sites (professional body member listings, Clutch, local business directories) heavily for "find an accountant near me"-style queries, meaning firms absent from those directories lose citation opportunities even with a strong own-site presence.

How does Gemini handle accounting-firm queries?

Gemini is tightly integrated with Google's index and Business Profile data, so it currently rewards the same signals that drive local pack visibility — verified Business Profiles, review volume and recency, and category accuracy — more than the other engines do. Because Gemini sits inside the Google ecosystem, a firm's Google Business Profile completeness (categories, services list, Q&A section, posts) has a more direct line to Gemini's answer than it does to ChatGPT's or Claude's. Gemini also draws heavily on Google's existing knowledge panels, so firms with a claimed and enriched Knowledge Panel — correct founding date, named partners, service area — get referenced with more specificity ("this firm specializes in X, founded in Y") than firms Gemini can't confidently disambiguate from similarly-named competitors.

How do Google AI Overviews handle accounting-firm queries?

Google AI Overviews prioritize pages that already rank well organically and pages with clear schema markup (FAQPage, HowTo, Article), effectively giving AI Overviews as a second, compressed reward for standard technical SEO done well. Unlike the conversational engines, AI Overviews draw from Google's existing index and ranking signals, so a firm's baseline SEO — page speed, mobile usability, backlink profile, on-page structure — still matters enormously; it's a prerequisite rather than irrelevant. The distinguishing factor for AI Overview inclusion specifically is answer format: pages structured as direct question-and-answer, with the answer stated in the first sentence of each section, get pulled into the Overview box far more often than pages that bury the answer in the third paragraph after a preamble. Firms publishing genuinely useful explainer content (breaking down a specific HMRC or IRS rule change) are showing up in Overviews even when their domain authority is modest, because format is doing more work than authority for this specific surface.

The three content gaps almost every firm still has

Gap 1: FAQ depth that matches how people actually ask AI engines questions. Most firm FAQ pages have five to eight generic questions ("What services do you offer?" "How much do you charge?") written in marketing language. Answer engines are being asked highly specific, long-tail questions — "can I claim mileage if I use my car for both my consultancy and my rental properties," "what's the penalty for filing a 1120-S three months late," "does a Ltd company need to register for VAT before it hits the threshold if it expects to." Firms that build out FAQ content around these long-tail, real-phrasing questions — ideally with FAQPage schema — give AI engines exact, extractable answers instead of forcing them to guess or default to a generic competitor.

Gap 2: Sector and niche pages that state a specialism plainly. "We serve businesses across all industries" is the single most common phrase found on accounting-firm websites, and it is close to useless for AI citation purposes. Answer engines are far more likely to cite a firm for "accountant for e-commerce sellers with EU VAT obligations" or "CPA for medical practices transitioning to an S-corp" than for generic small-business accounting — because the query itself is specific, and the engine needs a specific match. Firms that build dedicated, detailed pages per vertical (construction, hospitality, SaaS, medical practices, real estate investors) — each with the actual thresholds, rules, and numbers relevant to that sector — close this gap directly. A single "industries we serve" page with six bullet points does not.

Gap 3: Schema coverage beyond the basics. Most firms have implemented, at best, LocalBusiness schema and maybe an Organization tag. Few have layered in FAQPage, Service, Person (for named partners and their credentials), Review, or HowTo schema where relevant. This matters because schema is one of the few direct, machine-readable signals a firm controls completely — it's not a ranking guess, it's a structured statement of fact that AI crawlers can parse without inference. A firm with a named Person entity for its lead partner, complete with hasCredential (CPA, ACCA) and worksFor linking back to the firm entity, is giving Gemini and Google AI Overviews exactly the disambiguation signal they need. Most firms haven't touched this layer at all.

There's a fourth gap worth naming even though it wasn't in the brief: evidence pages are thin. Case studies that say "we saved a client £15,000" without showing the mechanism (which relief, which restructuring, which rule) don't give Claude or Perplexity anything to cite with confidence. Evidence pages that name the specific tax code section, HMRC manual reference, or IRS form involved — and walk through the reasoning — are the pages that get quoted rather than paraphrased.

What to do this month

Audit your own citation status before doing anything else. Run 15–20 realistic prospect queries — the specific ways a client would actually ask, not generic terms — through ChatGPT, Claude, Perplexity, and Gemini, and note whether your firm is named, whether a competitor is named instead, and which pages (if any) are being cited. This takes an afternoon and tells you more about your actual AI visibility than any amount of guessing. Perplexity's visible citations make this the easiest engine to start with.

Rewrite one high-intent page per week using answer-first structure. Pick the pages most likely to intersect with real prospect questions — your R&D tax credit page, your IR35 guidance, your S-corp election explainer — and rewrite the opening of each section so the direct answer appears in the first sentence, followed by the reasoning and the source citation (HMRC manual reference, IRS revenue procedure, ICAEW or AICPA guidance). Add FAQPage schema to each as you go.

Build one sector page with real specificity before building a fifth generic service page. Pick your highest-value niche — the industry where you already have the most clients — and write the page an AI engine would want to cite: the actual thresholds, the actual rules that apply only to that sector, named examples of the work you do. This single page will do more for answer-engine citation than a dozen more paragraphs about "personalized service" on your homepage.

The firms that show up in AI-generated answers this time next quarter won't be the ones with the biggest marketing budgets. They'll be the ones whose websites already read like the answer the model was looking for.