On the evidence I found, few standard proposal forms carry a dedicated AI section yet, but brokers say insurers are starting to gather AI governance information, and some carriers have filed AI exclusions. Prepare a one-page disclosure covering tools used, data entered, who reviews output, your policy and any incidents, kept consistent with the rest of the form.
Broker commentary from September 2026 says most professional liability policies are still silent on AI, which is why a clear disclosure helps both sides. Nothing here is cover advice: which wording you need, and whether to buy anything extra, is a conversation for your broker and insurer. Professional indemnity is called professional liability or errors and omissions cover in some markets, so match the terms on your own paperwork. I could not find a published standard form with AI questions in it, so treat the AI rows below as early practice and predictions, not as a fixed list.
What insurers ask now, and what they may ask next
A renewal has two layers. The first is long established: claims, circumstances, fee income and controls. The second is AI, where the evidence is thinner. A broker's renewal guide for accountancy firms, dated 25 March 2026, describes the proposal form as the evidence an insurer uses to price risk and decide terms, and says to keep answers consistent across the form, the renewal summary and any follow-up questionnaires. A September 2026 Insurance Business piece quotes two brokers: one says underwriters are not yet asking many questions about AI and governance, the other that insurers seem to be gathering governance information without yet knowing what to do with it, and the first advises clients to adopt a formal AI policy. The same piece says some affirmative AI cover comes with conditions such as documented governance, a risk assessment and evidence of monitoring. Those conditions shape the AI rows below; the vendor, data and contract rows are my own suggestions.
| Topic | What an underwriter may want | Evidence worth holding | Status of the evidence |
|---|---|---|---|
| Claims and circumstances | Date, plain-English summary of the allegation, status, costs paid or reserved, root cause and control changes made; near misses included | Claims register and notification records | Standard renewal practice |
| Fee income and services | Fee income reconciled to management accounts, split by service, with higher-risk work explained | Management accounts, service list | Standard renewal practice |
| AI governance | A written AI policy, a named accountable person, a way to approve new tools, and a register of approved tools and uses | Policy with date and approver, register | Brokers report insurers gathering it in 2026; a condition of some affirmative cover |
| Vendor checks | What you checked about AI vendors, including licence terms and the liability the vendor accepts | Vendor file with terms reviewed | My suggestion |
| Data and security | Where data goes, how it is protected, and whether you hold a separate cyber policy | Settings screens, cyber schedule | My suggestion |
| Human oversight | A documented review of AI output, who does it, and training on the tools' limits | Review checklist, training log | Monitoring is a condition of some affirmative cover |
| Contracts | Whether engagement terms mention AI, and how roles and liabilities are agreed with clients | Engagement letter wording | My suggestion |
The practical reading: prepare all seven rows, because the first two are certain and the rest cost little to assemble. If an insurer asks nothing about AI, you have lost a few hours. If it asks, you answer in minutes and with documents. The sibling tutorial for architects, what to tell your PI insurer about AI design work, covers the same ground for one profession.
Silent, affirmative or exclusionary: three ways a policy can treat AI
The Insurance Business article describes three ways carriers handle AI: silent, affirmative or exclusionary. It says most professional liability policies are silent. It also warns that affirmative wording does not guarantee better cover, that some affirmative wording is conditional on governance, risk assessments and monitoring, and that narrowing often arrives through revised definitions and carve-backs rather than one obvious AI exclusion. It notes that Verisk's general liability endorsements CG 40 47 and CG 40 48, in their January 2026 editions, exclude AI-related injury and damage claims, and that some carriers have filed broader exclusions. That is a different line of insurance from yours, but it shows the direction of travel.
| Treatment | In plain terms | Questions to put to your broker |
|---|---|---|
| Silent | The policy does not mention AI, so the insurer decides how a claim involving it is characterised under the existing wording | Is our policy silent? Will that change at renewal? |
| Affirmative | The policy says AI-related professional services are covered, sometimes with conditions | What conditions apply? What would we need to show, such as a policy or monitoring? |
| Exclusionary | The policy removes some or all AI-related claims, or narrows definitions so they fall outside | Is an exclusion or a narrower definition being added? Can you show me the wording? |
The article's advice to brokers is also useful to firms: read the full wording rather than the headline endorsement, and keep a record of these conversations because they matter if a coverage dispute arises. Preparation is as far as this goes. Whether to seek different wording is a decision for you and your broker.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA renewal timeline: start eight to twelve weeks out
The broker guide dated 25 March 2026 suggests starting six to eight weeks before renewal, earlier if you have had claims, changed services, merged, opened a new office or grown quickly, and warns that late submissions can mean fewer insurers quote. The AI paperwork adds work, because you may need to survey staff and write a policy, so I would start a little earlier. The schedule below is my suggestion, not a standard.
| Weeks before renewal | Task | Output |
|---|---|---|
| 12 to 10 | Run the staff survey on AI use; collect every tool and account; start an AI policy if you have none | AI-use register, first version |
| 10 to 8 | Write the claims and circumstances summary; reconcile fee income to the accounts | Claims summary, income reconciliation |
| 8 to 6 | List controls you can evidence; draft the AI-use disclosure | Controls list, draft disclosure |
| 6 to 4 | Meet your broker; fix gaps; check consistency against last year's answers | Final pack |
| 4 to 2 | Submit; answer follow-up questions within two working days | Submission and answers log |
| 2 to 0 | Read the terms; ask your broker to explain any new exclusion or condition in writing | Signed-off terms and a written explanation |
Put the dates in a shared calendar with a named owner for each step. The tutorial on gathering the figures before your broker calls shows the same discipline for a trades business.
Claims and circumstances: a filled-in summary and the AI draft
The same broker guide says to record, for each claim, notification or circumstance, the date, a plain-English summary of the allegation, the current status, costs paid or reserved, the root cause and the control changes you made, and to include near misses. Keep it short and factual, but cover what changed after the event. Here is a filled-in example for a six-adviser financial advice firm. The events and figures are invented for illustration.
| Type and date | Summary | Status and cost | Root cause and fix |
|---|---|---|---|
| Notified claim, 14 March | A client says a suitability report left out an ongoing charge on one product | Settled; $18,400 paid including defence costs | The report template had no line for that charge. Template corrected; a second-person check added to every report. |
| Circumstance, 2 June | A client letter complains that an annual review was missed; no claim made | Open; $5,000 reserved | Review dates sat in a spreadsheet nobody owned. Dates moved into the practice system with a named owner and a 30-day reminder. |
| Near miss, 9 September | An AI-drafted client letter quoted a fund charge that did not match the factsheet; caught at the second-person check before sending | No loss; nothing notified | The draft was not checked against the source. Rule added: every figure in an AI draft is matched to its source document before approval. |
The third row matters for your AI disclosure as well. It shows a control that failed once and was then tightened, which is a better story than "no incidents" when your staff use AI daily. To get from rough notes to a table like this, use a prompt that forbids opinions.
Below is my firm's claims and circumstances register in rough notes.
Turn it into a table with these columns: type and date; summary of the
allegation in plain English (two sentences at most); status; costs paid
or reserved; root cause; control changes made.
Rules: use only what I pasted. Write "not stated" for anything missing.
Do not name any person or client. Do not give any opinion on liability.
Keep the wording factual and neutral.
[paste notes, anonymised]What you would fix in the first draft: models like to explain. A typical reply writes "staff shortage" as the root cause of the missed review because the notes mentioned a busy month, rounds $18,400 to "about $18,000", and adds "the client was at fault" to the first row. Delete any cause the register does not state, restore exact figures, and remove every opinion on liability. Then check each row against your insurer's notification records, and have the person responsible for risk approve the final table. Do not paste claim files into a consumer chat tool; use a business plan, anonymise, and delete the working copies when finished.
Building the AI-use register from a short staff survey
You cannot disclose what you do not know. Run a five-minute survey, with an assurance that the goal is a register, not a telling-off. Ask which AI tools each person uses for work, including personal accounts; what for; what kinds of information they type or upload; whether the account is personal free, personal paid or a company plan; who checks the output before it reaches a client; and whether they have had a mistake or near miss. The tutorial on shadow AI explains why the personal accounts are the part most likely to surprise you.
Here is a register for the same six-adviser firm, again illustrative.
| Tool and account | Used for | Data allowed | Review before it reaches a client | Vendor terms checked |
|---|---|---|---|---|
| ChatGPT Business, company workspace, 6 seats | Meeting-note summaries and first drafts of client letters | Initials only; no account numbers or full names | Adviser edits every draft; second person checks suitability letters | Business plans do not train on business content by default; OpenAI's output indemnity does not cover Business plans (checked November 2026; re-check before renewal) |
| Company meeting transcription tool | Transcripts of client meetings | Client conversation audio | Adviser confirms every figure against the file | Retention and deletion terms to be confirmed |
| Free chatbot on two advisers' personal accounts, found in the survey | Rewriting emails | Client email text had been pasted | None | Not reviewed; to be stopped and moved to the company workspace |
The last row is why the survey comes first. Once found, it becomes a documented fix with a date, which you can show an insurer. If you leave it unfound, it becomes a statement that does not match reality. The "vendor terms" column records the facts an underwriter would ask about, and each one carries a date, because terms change.
The controls list an insurer can verify
List only the controls you operate and can evidence. An insurer may not check them at renewal, but a claim is when they get tested, and a control that exists only on paper turns a disclosure into a liability. Each row below needs an owner and a date, and the evidence column is what you would hand over if asked.
| Control | Evidence to attach | Owner |
|---|---|---|
| A written AI policy saying what is allowed and what is not | Policy with version, date and approver | Managing partner |
| An approved tool list and a new-tool approval step | The register above, with approval dates | Operations manager |
| No client personal data in personal or free accounts | Settings screenshots and staff acknowledgements | Operations manager |
| Human review before any AI output reaches a client | Review checklist and a sample of signed-off items | Each adviser |
| A second-person check on advice letters or reports that used AI | Initials and date in the file | Compliance lead |
| Engagement letter wording on AI use | The template clause and the date it was introduced | Managing partner |
| Staff training on the tools' limits | Attendance log with dates | Compliance lead |
| An incident and near-miss log | Log entries with the fix for each | Compliance lead |
| A vendor terms review | Vendor file with training default, retention and liability noted and dated | Operations manager |
| Cyber and data-breach arrangements | Policy schedule or incident plan | Managing partner |
The policy item can start small. The tutorials on an AI acceptable use policy for a small professional firm and an AI usage log for client work cover the policy and the record that supports the human-review row. An underwriter may also ask whether you hold a separate cyber policy, so confirm yours is in place and note its limits; what insurers ask about AI incidents under cyber cover goes into that side.
Drafting the AI-use disclosure with a prompt that cannot embellish
With the register and controls list ready, a model can draft the one-page disclosure in seconds. The danger is that it will make you sound better than you are, so write the prompt to forbid it.
Using ONLY the register and controls list pasted below, draft a one-page
AI-use disclosure for our professional indemnity renewal. Sections: tools
and account types; what each is used for; data we enter and data we never
enter; review before output reaches a client; policy and training;
incidents in the last 24 months; planned changes.
Rules: plain words, first person plural. Do not add any control that is
not listed. Where a control is missing, write "not in place". Flag any
sentence that depends on staff behaviour I have not verified, with the
word CHECK in capitals.
REGISTER: [paste]
CONTROLS: [paste]
INCIDENTS: [paste]An illustrative reply would open like this.
We use three AI tools. A company ChatGPT Business workspace (6 seats) is
used for meeting-note summaries and first drafts of client letters. A
company transcription tool records client meetings. Two advisers used free
personal chatbot accounts for email rewriting until September; this has
been stopped and the accounts are being closed. CHECK: confirm both
advisers have closed their accounts.
We enter initials only into ChatGPT. We do not enter account numbers or
full client names. Every AI draft is edited by the adviser, and suitability
letters get a second-person check before sending. In the last 24 months we
logged one AI-related near miss (a fund charge that did not match the
source); it was caught at review, and every figure in an AI draft is now
matched to its source document.What you would fix: the CHECK flags are the model doing its job, so resolve each one and delete the word. Confirm every sentence against a document, and remove anything that reads as a promise you cannot keep, such as "all staff always" or "never". Then compare the draft with last year's answers and the rest of this year's form, because the broker guide's central advice is consistency. A mismatch between a polished AI paragraph and a thin claims answer elsewhere is what draws a follow-up question.
The most common failure is overclaiming. Compare these two statements.
BEFORE: "We never use AI for client advice."
AFTER: "AI drafts first versions of some client letters and meeting
summaries. An adviser edits each one, and a second person checks
suitability letters before they are sent."The first sentence sounds reassuring and is false the moment anyone asks about meeting summaries. The second is longer, accurate and shows controls. The accurate version is the one that holds up under a follow-up question.
Eight questions for the broker conversation
Take the pack to your broker and ask these, then record the answers. They are questions, not recommendations: what to do with the answers is the broker's advice to give.
- Does the proposal form, or any supplementary questionnaire, ask about AI use this year?
- Is our current policy silent, affirmative or exclusionary on AI?
- Will any AI exclusion, condition or narrower definition be added at renewal, and can you show me the wording?
- If affirmative AI wording is offered, what conditions come with it, and what would we need to show?
- Which of our AI uses would the insurer treat as professional services?
- What evidence do you want from us: the policy, the register, the controls list, the incident log?
- Does our separate cyber policy respond to AI-related incidents, and where does it stop?
- If we add a tool or a new AI use during the year, what should we tell the insurer, and when?
Ask for key answers in an email after the call. If a point is unclear, say so and ask the broker to explain it in plain words; the renewal guide's advice that an insurer prefers a transparent submission to a vague one applies to the conversation with your broker as well.
Using AI to prepare, and what never goes in
AI earns its place in three jobs here: turning rough notes into tidy tables, drafting the disclosure from a register, and checking for inconsistencies. It should never be the author of the declaration.
- Use a business plan and anonymise. Strip names, client identifiers and any insurer reference numbers, and delete working copies when you finish.
- Paste facts, not memories. Every prompt should say "use only what I pasted", so gaps appear as "not stated" instead of invented details.
- Name a checker. A named person verifies each answer against a document, and signs off the final pack.
- Keep the trail. File the prompts, drafts and final answers with the renewal papers so you can show how the submission was built.
Two further prompts are worth keeping. The first compares last year's answers with this year's draft and lists only the differences.
Below are our answers from last year's renewal and our draft answers
for this year. List only the differences, grouped by topic, and mark each
as: explained by a change we listed, or UNEXPLAINED. Do not rewrite
anything.
LAST YEAR: [paste]
THIS YEAR: [paste]The second plays an underwriter reading your disclosure, so you can rehearse the follow-ups.
Act as an underwriter reading the AI-use disclosure below for a
professional indemnity renewal. List the five follow-up questions you
would ask and the document you would ask to see for each. Do not invent
any facts about the firm.
[paste disclosure]Treat the output as a rehearsal. If the model asks a question you cannot answer from your documents, you have found a gap before the insurer did. The tutorial on whether business insurance covers AI mistakes gives background on how these claims are characterised.
The form that said "we do not use AI"
A six-person consultancy submitted its renewal with a line saying the firm did not use AI. Three weeks later, a tidy-up of browser extensions turned up four staff using personal chatbot accounts for drafting, one of them with client reports. The owner had to tell the broker, correct the form, and explain what controls existed, which in practice meant writing the policy in a hurry.
A short survey before the submission would have produced a register, a one-page policy and an accurate disclosure, for roughly the same effort. This is an illustration, not a reported case, but it shows an easy way for a thin AI answer to go wrong: the form was signed before anyone had looked at what staff were doing. Asking first, documenting what you find and fixing what you can is a simpler position to defend than discovering it afterwards.
Questions firms ask before a professional indemnity renewal
Must I mention AI use if the proposal form does not ask?
I cannot give you the rule, because disclosure duties differ by policy and place. Many forms end with a declaration that your answers are complete and accurate. Ask your broker directly whether AI use is something to tell the insurer, and keep their answer in writing alongside your register.
Should we stop using AI before the renewal?
Not because of the renewal. Stop unapproved use, such as client data in personal accounts, because that is a policy question. Document what you do use, the checks you run and the incidents you have had. Whether any use affects your cover is a question for your broker and insurer, not a reason to hide it.
Can AI fill in the proposal form for us?
It can draft summaries from records you paste, and it can check last year's answers against this year's draft. A named person must verify every answer against the source, because the declaration is yours. Do not paste claim details into a consumer chat tool; use a business plan and anonymise.
How long should the AI-use disclosure be?
One page, with the register and controls list as attachments. It should cover the tools and plans, what they are used for, the data you do and do not enter, who reviews output before it reaches a client, your policy and training, and any incidents. Longer disclosures are harder to keep consistent with the rest of the form.
Further reads
- Does Agency E&O Cover AI Mistakes? What to Ask Your Carrier — Questions to put to a carrier about AI mistakes, from an agency angle.
- Shadow AI: How to Stop Staff Pasting Client Data Into Free Tools — Stopping staff pasting client data into free tools.
- How Brokers Use AI to Compare Policy Wordings at Renewal — How brokers compare policy wordings, which helps you ask better questions.
- AI for Chartered Accountants: What to Tell Clients About Your AI Use — What to tell clients about your AI use.
- How to Write an AI Usage Policy for Your Small Business — Writing the policy document an insurer may ask to see.
- AI Readiness Checklist for Accountants, Solicitors, Consultants — A readiness check for accountants, solicitors and consultants.
- Structural Engineer Fee Proposal: Scope and Assumptions Drafted With AI — A worked fee proposal with stages, assumptions, exclusions and variation rates, plus prompts that draft and stress-test each section.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: a broker's renewal guide for accountancy firms (25 March 2026); Insurance Business reporting on AI and professional liability policies (21 September 2026); OpenAI business pricing page and service terms.