Most “AI best practices” are written for software teams. Proposal work has its own constraints: solicitations with disclosure clauses, data that belongs to someone else, evaluators who read every word. Some things stay off the table not because the tech can’t handle them, but because being wrong means a protest you can’t walk back.
Four things AI shouldn’t decide, no matter how capable it gets. Pricing strategyturns on variables only your team knows. Final sign-off belongs to a person; AI drafts, humans approve. Competitive-intel interpretation can be summarized by a model, but the recommendation is yours. And unsupervised output to the customer is how one hallucinated certification becomes one credibility disaster. These are governance calls, not capability calls.
Read the solicitation for AI clauses. The ground is moving fast. NSF asks proposers to disclose whether and how generative AI was used in preparing a proposal. GSA has issued a proposed clause, GSAR 552.239-7001, that would make contractors disclose the AI systems used in performance and keep NIST-aligned documentation. Check Sections L, M, and the general terms every time.
Keep controlled data out of consumer tiers.CUI, ITAR/EAR, source-selection-sensitive information, and competitor data under NDA don’t belong in a chatbot that logs prompts. Confirm your approved-tool list first. Be honest about AI imagery:if a graphic is AI-rendered, don’t let it pass as a photo of a real facility or real people. Evaluators catch it, and trust evaporates across the whole submission. Use NIST’s AI RMF as the shared vocabulary when legal, security, and program management need to govern AI across pursuits. And remember that agencies are using AI on the evaluation side too, which rewards unambiguous structure: clear sections, explicit answers, RFP terms verbatim.
Three questions about the work and the data:
Answer all three. Most “AI best practices” skip these — proposal work has its own rules.
On your current pursuit, read Sections L, M, and the general terms for any AI-disclosure clause, and confirm your team’s approved-tool list before any controlled data goes near a model.
For CUI and regulated bids we work inside your accredited boundary or stand up a compliant enclave. See the security page.
Book a call →