How to Run an AI Visibility Audit for an Accounting Firm

An AI visibility audit tells you exactly where your accounting firm appears — and where it doesn't — when prospective clients ask ChatGPT, Perplexity, Gemini, Claude, or Google AI Overviews a question your firm should be answering. Unlike a traditional SEO audit, the output isn't a ranking position; it's a citation record: who gets named, what facts are stated, and whether your competitors are filling the space you're leaving empty.

This guide walks through the full process: the queries to run, the scoring framework, the page-level checks, and a monthly reporting template you can maintain with a spreadsheet and two hours a month.

---

Why Does AI Visibility Matter for Accounting Firms Right Now?

AI engines are now the first stop for a significant share of professional-services research. A prospect typing "best accountant for R&D tax credits in Manchester" or "does my US LLC need a UK accountant" is increasingly reading an AI-generated answer — not ten blue links. If your firm isn't cited in that answer, you weren't considered. The prospect may never reach your website at all.

The audit below establishes your current citation baseline so you can measure improvement quarter over quarter, starting from Q2 2026.

---

Step 1: Build Your Query Set

The right query set mirrors how real prospects phrase questions — not how your firm describes its services internally.

Construct queries across four intent types:

Service + location queries — The most commercially valuable. Examples: - "Best accountant for small businesses in [city]" - "VAT specialist for e-commerce UK" - "Outsourced CFO services for startups London" - "CPA for UK expats living in the US"

Problem-led queries — Prospects describing their situation, not searching a service category: - "My company received an HMRC compliance check, who can help?" - "I'm behind on my Self Assessment tax return, what should I do?" - "How do I set up payroll for my first UK employee?"

Comparison and credibility queries — Used later in the buyer journey: - "How do I choose an accountant for a property portfolio?" - "What qualifications should my accountant have — ICAEW or ACCA?" - "Questions to ask an accountant before hiring them"

Firm-name queries — Direct reputation tests: - "[Your firm name] reviews" - "Is [your firm name] good for [service]?" - "Who are the partners at [your firm name]?"

Aim for 20–30 queries in total. Keep them in a spreadsheet with columns: Query Text, Intent Type, Engine Tested, Date Run.

---

Step 2: Run Queries Across All Five Engines

Test each query in ChatGPT (GPT-4o), Perplexity, Gemini (via Google), Claude (Sonnet or Opus), and Google AI Overviews (sign into a Google account in an incognito window to trigger AI Overview where available).

Practical rules for consistent results:

• Use a fresh conversation thread for each query — never build on prior context in the same session. Prior context contaminates results.
• Run queries in incognito or a clean browser profile to reduce personalisation.
• Screenshot the full response before it disappears or regenerates.
• Note the date and engine version where visible (e.g. "Perplexity — Pro, 30 June 2026").

For Google AI Overviews specifically, note whether an Overview appears at all — many commercial queries still trigger traditional results. Record "No AI Overview triggered" as a data point; it tells you something about how Google classifies the intent.

---

Step 3: Record the Four Citation Outcomes

For every query-engine combination, log one of four outcomes:

1. Cited — accurate. Your firm is named and the facts stated (location, specialisms, qualifications, contact details) are correct. This is the target state.

2. Cited — inaccurate (hallucinated facts). Your firm is named but something is wrong: wrong city, wrong partner name, incorrect specialism, outdated fee claim. This is an active reputational risk and should be treated as the highest-urgency fix. Log the exact false claim verbatim.

3. Not cited — competitor named. Your firm doesn't appear, but a named competitor does. Log which competitor and which engine. A pattern here tells you where to focus optimisation effort.

4. Not cited — generic answer. No firm is named. The engine produces a generic how-to response or a list of factors to consider. This is neutral — it means the engine hasn't confidently attributed authority to any firm for this query yet, which is an opportunity.

Your spreadsheet columns for this step: Query, Engine, Outcome (1–4), Firm Cited (Y/N), Hallucinated Claim (verbatim), Competitor Named, Screenshot Reference.

---

Step 4: Score Your Citation Rate

Once you've run all queries across all engines, calculate a citation rate per engine:

Citation Rate = (Cited — accurate) ÷ Total queries tested × 100

A typical benchmark for a firm that has done no AI-specific optimisation is a citation rate below 10% on service + location queries and 20–30% on firm-name queries. If your accurate citation rate is below these figures, the subsequent fixes in this audit will move the needle meaningfully.

Also calculate:

• Hallucination rate: (Cited — inaccurate) ÷ Total queries × 100. Any figure above zero is urgent.
• Competitor displacement rate: (Not cited — competitor named) ÷ Total queries × 100. High displacement on service queries is the clearest signal you need entity and content fixes.

---

Step 5: Run the Page-Level Checks

A low citation rate is a symptom. The following page-level checks identify the causes.

Do You Have Answer Blocks on Key Service Pages?

AI engines extract direct-answer sentences — short, declarative statements that resolve a specific question — and surface them verbatim or near-verbatim in responses. If your service pages are written in marketing prose ("We pride ourselves on delivering exceptional…"), they don't contain extractable answers.

Check: Open each core service page. Can you find a sentence that directly answers "What does [service] involve?" or "Who is this service for?" within the first 100 words of the page body? If not, the page is not AI-optimised.

Fix signal: Draft one direct-answer sentence per service and place it at the top of the body copy. Example: "Our R&D tax credit service handles the technical narrative, financial calculations, and HMRC submission for UK SMEs claiming under the RDEC or SME schemes."

Is Your Schema Markup Complete?

Schema markup tells AI crawlers and Google's Knowledge Graph what your firm is, where it operates, and what it does. For accounting firms, the minimum viable schema set is:

• LocalBusiness (or the more specific AccountingService type) with name, address, telephone, url, openingHours, and geo populated.
• Person schema for each named partner or director, linked to the firm's Organization entity.
• FAQPage schema on any page structured as questions and answers.
• Review or AggregateRating schema if you display client testimonials.

Check: Run your homepage and key service pages through Google's Rich Results Test and a schema validator such as Schema.org's validator. Look for missing required fields, particularly address and sameAs (links to your Google Business Profile, Companies House entry, ICAEW/ACCA directory listing, and LinkedIn).

The sameAs property is the single most under-used schema field for accounting firms. It creates explicit entity links that AI knowledge graphs use to verify and consolidate information about your firm.

Are Your Author and Entity Signals Strong Enough?

AI engines weight content from identifiable, credentialled humans more heavily than anonymous pages. For accounting firms, this means:

• Every article, guide, or insight on your site should have a named author with a linked author bio page.
• The bio page should state qualifications (ACMA, CGMA, ACA, ACCA, CTA — use the full credential name at least once, alongside the abbreviation).
• The bio page should link out to the author's LinkedIn profile and any professional directory entries (ICAEW Find a Chartered Accountant, ACCA's member directory).
• The firm's About page should list all partners and directors by name, with their credentials and areas of specialism stated explicitly.

Check: Search "[partner name] accountant [city]" in each AI engine. Does the engine return accurate information about that person? If the engine draws a blank or fabricates a role, the entity signals for that individual are insufficient.

Do You Have Niche Proof Content?

Niche proof is the term for content that demonstrates deep, specific expertise in a service area — the kind of content a generalist firm cannot produce. AI engines use niche proof to calibrate whether a firm is genuinely authoritative on a topic or merely mentions it in passing.

Examples of niche proof for accounting firms: - A detailed guide to claiming HMRC's Enterprise Investment Scheme relief for first-time investors, with worked examples. - A comparison of US GAAP and UK GAAP treatment for a specific asset class, written for a named client sector. - A step-by-step walkthrough of Making Tax Digital for Income Tax Self Assessment (MTD for ITSA), covering the April 2026 mandation timeline.

Check: For each service in your query set, does your site have at least one piece of content that goes beyond surface-level description? Does it name specific HMRC forms, IRS publications, Companies House filing requirements, or ICAEW guidance documents by reference number or title?

If not, that service is unlikely to generate citations on expertise-led queries.

Are Your Reviews Visible and Schema-Tagged?

Reviews directly influence AI engine responses on credibility queries ("[firm name] reviews", "is [firm name] good for…"). AI engines pull review signals from Google Business Profile, Trustpilot, and Yelp (for US-facing firms), as well as on-site testimonials marked up with Review schema.

Check: - Does your Google Business Profile have a minimum of 15 reviews with an average of 4.5 or above (a typical benchmark for citation in local queries)? - Are recent reviews (within the last six months) present? Recency signals that the firm is active. - Do any reviews mention specific services by name? Named-service reviews carry more weight than generic praise. - Are on-site testimonials marked up with Review schema including reviewBody, author, and datePublished?

Is Your NAP Consistent Across All Directories?

NAP — Name, Address, Phone number — consistency is a foundational local entity signal. Inconsistencies (different trading names, old phone numbers, outdated postcodes) create conflicting entity data that AI engines resolve by de-weighting or ignoring the source entirely.

Check: Search your firm name across Google Business Profile, Companies House, your ICAEW or ACCA directory entry, Yell, Thomson Local, Yelp, and any other directories where you have a listing. Every listing should show identical Name, Address, and Phone. Trading name variations (e.g. "Smith & Co" vs "Smith and Co Accountants Ltd") must be standardised.

---

Step 6: Prioritise Fixes by Impact and Difficulty

Not every fix delivers equal return. Use a simple 2×2 prioritisation:

High impact, low difficulty — do immediately: - Correct hallucinated facts (contact the engine's feedback mechanism; update your website so the correct information is crawlable and schema-tagged). - Add sameAs schema links to your homepage and About page. - Standardise NAP across all directories. - Add a direct-answer sentence to the top of each core service page.

High impact, higher difficulty — schedule within 30 days: - Write niche proof content for your top three service areas. - Build or rebuild author bio pages with full credential listings and external links. - Add FAQPage schema to your FAQ and service pages.

Lower impact, low difficulty — batch monthly: - Request reviews from recent clients, asking them to mention the specific service used. - Add datePublished and dateModified to all article schema. - Update any stale content (anything not reviewed in the past 12 months) with a new dateModified.

Lower impact, higher difficulty — defer or deprioritise: - Full site redesign to accommodate answer-block layouts. - Building out entity pages for every team member.

---

Step 7: Monthly Reporting Template

Maintaining a monthly audit log lets you track citation rate changes, catch new hallucinations quickly, and demonstrate progress to firm partners. The following template requires no specialist tools — a shared Google Sheet or Excel workbook is sufficient.

Tab 1: Query Log

Columns: Query Text | Intent Type | Engine | Date | Outcome (1–4) | Firm Cited | Hallucinated Claim | Competitor Named | Screenshot Ref

Run 10–15 priority queries each month (a rotating subset of your full query set). Log every result.

Tab 2: Citation Rate Dashboard

Rows: ChatGPT | Perplexity | Gemini | Claude | Google AI Overviews | Total Columns: Month | Queries Tested | Cited Accurate | Hallucinated | Not Cited (Competitor) | Not Cited (Generic) | Citation Rate %

Calculate citation rate automatically with a formula: =Cited Accurate / Queries Tested.

Tab 3: Hallucination Tracker

Columns: Date Discovered | Engine | Query | False Claim (verbatim) | Correct Information | Fix Applied | Date Fixed | Re-tested (Y/N) | Resolved (Y/N)

Any open row in this tab is an active risk. Review it at the start of every monthly session.

Tab 4: Fix Backlog

Columns: Fix Description | Page / Asset Affected | Priority (High/Med/Low) | Owner | Due Date | Status

Pull from Step 6 prioritisation. Mark items complete only after re-testing in at least two engines.

Tab 5: Competitor Displacement Map

Rows: Each competitor firm named in your audit results Columns: Engine | Query | Times Cited (this month) | Times Cited (last month) | Trend

This tab tells you whether a competitor's AI visibility is growing or static. If one firm is displacing you consistently on a specific query type, analyse their content and schema against yours.

---

How Long Does an Audit Take?

A first-pass audit — building the query set, running 25 queries across five engines, logging results, and completing the page-level checks — takes a typical firm between four and six hours. Subsequent monthly maintenance runs take 90 minutes to two hours once the spreadsheet infrastructure is in place.

The highest-value investment is the first audit. It establishes your baseline citation rate and surfaces hallucinated facts that could be actively damaging your reputation with AI-generated responses right now, in Q2 2026, before a prospect ever reaches your website.

---

What to Do When an AI Engine Has the Wrong Facts

Hallucinated or outdated firm facts require action on two fronts simultaneously.

On your own site: Ensure the correct information is present, prominently placed, and schema-tagged. The engine's next crawl will index the corrected data. This is the most reliable long-term fix — AI engines ultimately ground their responses in crawlable sources.

Direct feedback: ChatGPT, Perplexity, and Gemini all offer response feedback mechanisms (thumbs down, "Report an issue"). Use them. Flag the specific incorrect claim and provide the correct information. This doesn't guarantee immediate correction but contributes to training and moderation queues.

Third-party corroboration: If the wrong city or wrong specialism is being cited, check whether any third-party source (an old directory listing, a press mention, a stale partner bio on a previous employer's site) is the source of the confusion. Remove or correct those upstream sources.

---

The Audit as an Ongoing Practice, Not a One-Off Project

A single audit gives you a snapshot. Monthly tracking gives you a trend. AI engines update their knowledge continuously — new crawls, updated training data, retrieval-augmented generation pulling from live sources — which means your citation status can change without any action on your part.

Firms that establish a regular audit cadence starting in Q2 2026 will have six months of baseline data by year-end, which is enough to demonstrate measurable citation rate improvement to firm partners and to identify which content and schema investments are producing the clearest results.

The playbook above is deliberately engine-agnostic. The query types, the citation scoring, the page-level checks, and the reporting template apply equally whether you're optimising for ChatGPT today or for whatever engine leads the market in twelve months. The underlying principle doesn't change: AI engines cite sources they can identify, verify, and trust. The audit tells you where that trust is missing.