Using AI in small business grants applications while keeping accountability

An English business may be able to use artificial intelligence when preparing a grant application, but permission depends on the funder and opportunity. The applicant remains responsible for accuracy, originality, confidentiality, declarations and submission. Check the current policy before placing application material into any AI service.

What to take away

• The applicant stays responsible for accuracy, originality, confidentiality, declarations and submission even when AI helps prepare a grant application.
• Permission to use generative AI depends on the funder and the individual opportunity, so check the current policy before submitting material.
• A named person must approve every AI-assisted passage, calculation and citation, and fluency in the output is not verification.
• Do not put trade secrets, unpublished intellectual property, payroll or customer records into an AI service without approval.
• Open every cited source and recalculate budgets outside the language model, because AI links and figures can be wrong.

Read the funder's policy first

UKRI's generative AI policy applies across UKRI funding opportunities. The page was updated on 6 July 2026 and expects applicants to be transparent when they use generative AI to develop an application.

Individual opportunities can add instructions. A current UKRI funding record concerning explainable AI says generative AI use in application preparation is permitted with caution and links back to the policy. Another funder may prohibit, restrict or require disclosure of particular uses.

Save the policy version and opportunity guidance reviewed. Recheck them before submission rather than relying on an older application or an adviser's general statement.

Keep humans accountable

Assign a named person to approve every AI-assisted passage, calculation and citation. The reviewer needs access to the underlying business evidence. Fluency is not verification.

Do not ask a tool to invent market figures, customer quotes, technical results, partner commitments or project history. Remove an unsupported statement instead of softening it with phrases such as "industry experts believe".

Compare the final answer with the funder's question and scoring guidance. AI can produce a plausible response that avoids the specific evidence requested.

Protect confidential and personal information

Map what the tool receives, stores, uses for training or shares with subprocessors. Do not enter trade secrets, unpublished intellectual property, payroll, customer records or partner information until the business has approved the service and purpose.

The ICO guidance on AI and data protection explains UK data-protection issues for organisations using AI. The applicant should also check contractual confidentiality and funder restrictions that apply beyond personal data.

Use redacted or synthetic material for early tests. Restrict access, retain only the necessary record and follow the business's incident process if information is exposed incorrectly.

Verify sources and calculations

Open every cited source. Confirm publisher, date, geography, definition and relevance to the claim. AI-generated links can be wrong or lead to material that does not support the sentence.

Recalculate budgets outside the language model and reconcile them to quotations, accounts and cash flow. Check that totals, VAT, grant percentage and applicant contribution use the scheme's rules.

Record how AI was used

Keep the tool and version where available, date, user, purpose, material supplied, output retained, edits, verification and approver. Follow the funder's disclosure wording rather than creating a vague statement.

Do not claim that a human wrote the whole application if AI produced material that the relevant policy says should be disclosed. Equally, do not imply that the funder endorses a tool merely because its policy permits use.

Use AI for bounded tasks

Lower-risk uses may include reorganising applicant-supplied notes, testing whether an answer addresses every question, identifying undefined terms or suggesting a review checklist. The human must still verify the output and rewrite it in the applicant's accurate voice.

Higher-risk uses include generating evidence, interpreting legal eligibility, scoring the application as if the tool knew assessor decisions, or processing sensitive data without approval. Use qualified advice for tax, law and material financial decisions.

Test the review process on a small section before applying it across the submission.

This draft contains no live internal links. It remains on hold for a publication-day policy check, data-protection review and named human verification of every factual and technical statement.

Before you act

• Save the policy version and opportunity guidance you reviewed.
• Recheck the funder policy and guidance before submission.
• Assign a named person to approve each AI-assisted passage.
• Remove unsupported statements rather than softening them.
• Map what the tool receives, stores, uses for training or shares.
• Record tool, version, date, user, purpose, edits and approver.

Common questions

Who is accountable for the content of an AI-assisted grant application?

The applicant remains responsible for accuracy, originality, confidentiality, declarations and submission. Assign a named person to approve every AI-assisted passage, calculation and citation, and give that reviewer access to the underlying business evidence. Fluency in the generated text is not verification, so a human must still check the output.

What should I check before entering business information into an AI tool?

Map what the tool receives, stores, uses for training or shares with subprocessors. Do not enter trade secrets, unpublished intellectual property, payroll, customer records or partner information until the business has approved the service and purpose. Check the ICO guidance, plus contractual confidentiality and funder restrictions that apply beyond personal data.

How should I record and disclose the use of AI in an application?

Keep the tool and version where available, date, user, purpose, material supplied, output retained, edits, verification and approver. Follow the funder's disclosure wording rather than creating a vague statement. Do not claim a human wrote the whole application if AI produced material that the relevant policy says should be disclosed.