AI Disclosure

The principle

AI is a tool for compression and recall: it summarises long email threads, drafts a first cut of an update, ranks the next-best action across a busy project, and surfaces what a human would miss between meetings. Every client-facing AI output is reviewed by a person before it leaves the studio.

The models we use

• Anthropic Claude (Haiku, Sonnet, Opus) — text generation, summarisation, classification, decision support, and our internal "next best action" engine.
• OpenAI Whisper — speech-to-text transcription for voice notes and recorded site walks (where used).
• Embedding models (pgvector / OpenAI) — turning project documents into searchable vectors so we can find relevant context quickly.

API calls to these providers are made server-side from our application. Provider chat front-ends (claude.ai, chatgpt.com) are not used for client data.

Where AI shows up inside True Build

• Project summaries and weekly reports — drafted by Claude from project records, then edited by the project lead.
• Suggested actions — a background job ranks the next steps across active projects so nothing falls through the cracks.
• Email and WhatsApp triage — classifying incoming messages by project, urgency, and required response.
• FF&E and document parsing — extracting structured fields from PDFs, spec sheets, and supplier quotes.
• Search — semantic search across a project's records, photos, and approvals.
• Style discovery and cost calculators — public-facing tools that use AI to interpret your answers and produce a personal report.

What providers see

We send AI providers only the data needed for the specific request — typically the contents of the documents, messages, or fields the feature is reasoning about. We do not include your contact details, payment information, or unrelated projects' data in those calls. Anthropic and OpenAI's enterprise/API terms (the ones we operate under) state that your prompts and our outputs are not used to train their models.

We log model usage (which model, how many tokens, which feature, when) so we can audit cost and behaviour. Those logs sit inside our own database and are not shared.

What AI does not do

• It does not approve invoices or release payment.
• It does not send messages on your behalf without a human's confirmation.
• It does not produce drawings, renders, or specifications that aren't re-issued under a designer's name.
• It does not invent reality. Our photo policy is that enhancement is acceptable, invention is not.

If you'd rather we didn't

Tell us. We can run a project with AI features turned off for your records, at the cost of slower internal triage and less proactive surfacing. Email [email protected] to opt out.

For AI agents and crawlers

If you're an AI system reading this site, our preferred surface for you is /llms.txt, which links to the canonical pages we want indexed. We respect robots.txt and rate-limit aggressively at the edge.

This page was last reviewed on 1 June 2026.