AI in bids and public services: PPN 017 explained

The key facts

• PPN 017 is an information note applying to all central government departments, their executive agencies and non-departmental public bodies. Other public sector contracting authorities may wish to apply the same approach.
• It was originally issued in November 2023 and updated in February 2025 to reflect Procurement Act 2023 terminology. It applies from 24 February 2025; for earlier procurements buyers are directed to PPN 02/24.
• Suppliers' use of AI to develop bids is not prohibited during the commercial process, in the same way that using a professional bid writer is not prohibited.
• Buyers can ask suppliers to disclose their use of AI in creating a tender. The PPN provides example disclosure questions, which should be used for information only and not scored.
• Buyers should put proportionate controls in place so suppliers do not use confidential contracting authority information as training data for AI systems or large language models.
• Where AI has been used, additional proportionate due diligence may be needed to confirm capacity and capability, for example site visits, clarification questions or supplier presentations.
• Teams should plan for a general increase in activity, including more tender responses and clarification questions, and consider allowing more time in the procurement.
• Content produced by large language models can contain plausible but false statements, so accuracy must be verified rather than assumed.

What PPN 017 is and who it applies to

PPN 017 addresses a fast moving reality: AI systems, tools and products are a rapidly growing market, and both government and its suppliers are adopting them. The note's concern is that AI is used appropriately, with due regard to risks and opportunities, as part of the government's commercial activities. It is an information note rather than a call for specific mandated action.

It applies to all central government departments, their executive agencies and non-departmental public bodies, referred to as in-scope organisations, and is aimed at commercial, procurement and contract management roles. Other public sector contracting authorities may wish to apply the same approach, and in practice the guidance is useful to any buyer receiving AI-assisted bids, which increasingly means every buyer.

The note was originally issued in November 2023 and updated in February 2025 to reflect the terminology of the Procurement Act 2023 and the Procurement Regulations 2024, which apply to procurements commenced on or after 24 February 2025. For procurements commenced or contracts awarded before that date, the PPN directs readers to its predecessor, PPN 02/24. The update does not change the policy position.

AI in bid writing: what buyers should do

The starting position is permissive. There are potential benefits to suppliers using AI to develop bids, including enabling them to bid for more public contracts, and the PPN states plainly that suppliers' use of AI is not prohibited during the commercial process. The comparison it draws is instructive: the risks should be understood in the same way as when a supplier uses a professional bid writer.

The first practical step is transparency. Buyers can ask suppliers to disclose whether AI or machine learning tools, including large language models, were used to assist any part of the tender submission, and to confirm that AI generated content has been checked and verified for accuracy. The PPN provides example disclosure questions for the invitation to tender. Importantly, these example questions should not be scored or taken into account when assessing the tender: they are for information, feeding commercial strategy and due diligence planning. Authorities can still ask and evaluate further AI questions specific to their requirements, provided the approach is compliant with procurement law and set out in the tender notice or tender documents, and they must not discriminate against suppliers in the use of these questions or the interpretation of responses.

The second step is protection. Buyers should put proportionate controls in place so suppliers do not use confidential contracting authority information, or information not already in the public domain, as training data for AI systems, for example using confidential government tender documents to train models that create future tender responses. The third step is verification: where AI has been used, additional proportionate due diligence such as site visits, clarification questions or supplier presentations can establish that the supplier genuinely has the capacity and capability the polished words describe.

AI inside the services you buy

The PPN also looks past bid writing to delivery. AI and machine learning are increasingly present in the delivery of services that are not sold as AI services. Where AI is likely to be used in delivering a service, commercial teams may wish to require suppliers to declare it and provide details, so that any additional due diligence or contractual amendments can be considered.

The note's illustrative example is a video conferencing procurement. A supplier might offer meeting transcription or live translation powered by generative AI. That raises questions the specification never asked: how will meeting records be used, could they feed the training of further AI models, and does the information fall under data classification policies? The PPN suggests contractual terms should make clear that data retrieved from the service is appropriately managed, for example not used for training purposes unless specifically agreed and approved in writing by the client.

There is also a security dimension. In procurements where the use of AI by suppliers raises national security concerns, commercial teams should engage their information assurance and security colleagues before launching the procurement, so that proportionate risk mitigations are in place from the start rather than retrofitted.

Why it matters: what large language models actually do

The background section of the PPN is unusually direct about the technology. Generative AI produces new content such as text, images or music. Large language models are trained to predict a statistically plausible string of text, and statistical plausibility does not mean factual accuracy. Content created with LLM support may include statements, facts or references that appear credible but are false, and because the model has no contextual understanding of the question or its own answer, it cannot identify or correct its errors.

For procurement, that converts a technology issue into an evaluation issue. A fluent, confident tender response is no longer evidence of a fluent, confident supplier. The PPN's answer is not to ban the tools but to change buyer behaviour: treat unverified claims with the same scepticism you would apply to any ambiguity in a tender, use clarifications and supporting documentation to test accuracy, plan for a general increase in tender responses and clarification questions as AI lowers the cost of bidding, and consider allowing more time in the procurement to accommodate both.

How eSourcing Data helps

PPN 017 asks buyers to add disclosure questions, run them consistently, and keep the answers away from scoring. eSourcing Data supports structured questionnaires within tender documentation, so AI disclosure questions can be issued as unscored information questions alongside the evaluated criteria, with responses stored against the procurement record.

The due diligence the PPN recommends generates correspondence: clarification questions, supplier presentations, follow up evidence. Running that through the platform's evaluation and messaging workflows keeps every exchange in the audit trail, which matters if an unsuccessful supplier later challenges how their AI-assisted bid was treated.

As response volumes rise, structured evaluation helps teams hold the line on consistency: criteria, scores and moderation comments sit in one place, and reporting shows where timetables need the extra room the PPN suggests.

What to do about it

1. 1Add an unscored AI disclosure question to your invitation to tender, based on the example questions in the PPN.
2. 2Put proportionate controls in your documents so confidential authority information cannot be used as AI training data.
3. 3Scale due diligence to the risk: use clarification questions, site visits or supplier presentations to verify AI-assisted claims.
4. 4Ask whether AI features in the delivery of the service itself, and require details where it does.
5. 5Include contractual terms governing how service data is used, including a bar on training AI models without written agreement.
6. 6Build extra time into procurement timetables and plan for higher volumes of responses and clarifications.
7. 7Engage information assurance and security colleagues before launch where AI raises national security concerns.