Human-led.
AI-assisted.
Results-obsessed.
Our standard, version
The standard
Read the standard in full
A named human is accountable for everything that leaves GTU.
The person whose name is on a deliverable is accountable for it regardless of how it was produced. “The AI wrote it” is not an explanation for an error; it leaves an admission that we skipped the requisite human review step in our QA workflow.
No AI output reaches a client or production without a delivery lead reviewing it first.
Every piece of work that involves AI assistance, in copy, research, analysis, code, or design, requires human review before it is used, shared, or presented. AI output is a first draft. An unreviewed AI deliverable is not a GTU deliverable; it is an unfinished one, and not ready for prime time.
AI never makes the call.
All consequential decisions and all system architecture are made by people. Human oversight applies at every iteration of the work, not as a final check.
Humans write client communication.
Emails, status reports, recommendations, and anything else a client reads is written by the person who owns the relationship. AI may check that writing for gaps, clarity, and completeness. AI does not author client communication. One logged exception: business development. Prospecting communication may begin from an AI-drafted model, but still requires human editing, polish, and review by the sender before delivery.
We verify before we publish.
AI is not used to generate client-specific data, statistics, or claims without independent verification. Research is cross-referenced against authoritative sources before it is treated as fact. Quoted material includes its source.
Client information stays confidential.
Confidential client data is never entered into public AI tools and is treated with the same discretion as any proprietary business information.
Platform automation acts inside defined limits and is reviewed on a cadence.
Ad platforms run AI-powered deployment as a function of the system. We define where that automation is allowed to act and what it may not change without a person, and a named specialist reviews its decisions on an agreed cadence rather than per change. The cadence is set per service and per engagement: reporting tends to run monthly, while paid media review may follow the campaign rather than the calendar.
No agentic action on a live client system without a human at the controls.
Automated agents do not act on production sites, live client interfaces, or client accounts without direct human control.
What AI does in our workflow
We disclose our AI usage in task-level detail, because that is what a sophisticated buyer wants to see and what our leadership vote asked for.
Development
Where AI assists
Code drafting under architect review; staging crawls and parity checks
Who reviews
Development lead and architect
Search (SEO, GEO, AEO)
Where AI assists
Title and metadata rewrites at scale; research synthesis; technical audits
Who reviews
Strategist who owns the account
Paid media
Where AI assists
Platform automation within defined bid, budget, and creative limits; reporting drafts
Who reviews
Named paid media specialist, on the agreed cadence
Design
Where AI assists
Design concepting, wireframing, design decision documentation, design-system token mapping, competitive research, QA, developer handoff prep, design-to-code
Who reviews
Designer who owns the work
Content
Where AI assists
Research, outlines, first drafts, copywriting at scale, proofreading and completeness checks
Who reviews
Writer and delivery lead
Operations
Where AI assists
Invoicing and scope checks; QA against checklists
Who reviews
Project manager
Business development
Where AI assists
Prospecting drafts, as the logged exception to rule 4
Who reviews
The sender edits, polishes, reviews, and sends personally
The rows describe practice as of November 2026. When our practice changes (as we know AI most certainly will), we update the tasks along with it.
Every engagement starts with a diagnosis
Every engagement starts with a diagnosis, because what a client asks for and what they need are often two different things.