The GPT Agency: personality first

Vision and differentiation

A personality for every brand. A shared brand spirit; a relationship that belongs to the individual. ABI means Artificial Brand Intelligence.

We begin with the character and its relationship with people. Reusable capabilities then help it research, recommend and act when useful. A person may talk with Pre about motivation before there is any shopping need. If a suitable need emerges, the same character can help them understand a product and take an approved next step. This is our product position, not a blanket competitor claim or evidence of commercial lift.

The product

The persona provides researched identity, voice and points of view. Digital DNA supplies eight shared compounds. Capability modules provide research, recommendations, preparation and eligible transactions. CEM lets the brand shape expression and approve knowledge, tool access, destinations and limits. The Gauntlet and a named human govern release.

Pre and Arthur are independent demonstrations, not Nike/Guinness client deployments. Current prototypes demonstrate character and local voice; they do not connect membership, workouts, inventory, baskets, payments or social accounts.

Distribution and integration

The GPT Agency creation plugin would help brand teams build and manage personas. A consumer-facing brand plugin would let people access a persona and its approved tools. These are different products. A persona may also appear on owned websites and apps through tested host adapters.

MCP exposes authorised tools on an integration server; the client’s existing systems remain authoritative for identity, products, run data and orders. Plugin installation does not guarantee the host replaces its own personality, supplies memory/voice or grants Sponsored Agent capabilities.

A Nike journey could invite official membership, open an NRC download route, connect approved run records and prepare a relevant product choice. The customer completes terms, app installation and social follows in the responsible provider’s environment. Account linking and basket access require agreed Nike interfaces; public pages do not establish that permission.

Purchasing responsibilities

The agency builds the persona experience, scoped adapters, CEM controls and evaluation. The client supplies approved sources and access, authoritative prices/inventory and membership/commerce interfaces. Its merchant systems and payment provider execute payment, orders, fulfilment and returns.

The first purchase route is product advice plus an approved merchant checkout handoff. A basket can be prepared when the client supports it. Later confirmed transactions need exact order consent, revalidated totals, server-enforced limits, idempotency and authoritative receipts. OpenAI currently recommends external checkout for most plugins; embedded payment-sheet access is limited. Do not equate MCP tools with unrestricted checkout permission.

The proposition

The GPT Agency develops distinctive brand personalities, prepares them for supported conversational environments, and gives brands a CEM—Conversation Engine Management—platform to manage quality, knowledge, priorities and performance.

ABI: Artificial Brand Intelligence. The category proposition: a brand’s identity becomes a conversational presence, with a distinctive character and reviewed evolution. This is vision language, not a claim of artificial general intelligence.

The same Nike spirit. A relationship that's yours.

Pre is the illustrative Nike concept: athlete-to-athlete conversation with imagination, sporting depth and empathy. A private member companion and a sponsored product-discovery agent share character while having different permissions and objectives. Personal vulnerability is never a sales-targeting signal.

What a brand buys

• Design and build: researched identity, character, sourced knowledge, voice direction and approved actions.
• Deployment: integration with brand developers and tested host adapters, with limitations documented.
• Management: continuously maintained sources, evaluation, reviewed improvements, operation and reporting.
• Performance: an agreed persona CPC for verified outward handoff clicks from the sponsored-agent conversation. The platform keeps the entry click.

Build, deployment and management charges remain. There is no extra agency performance charge merely for entering a conversation. A handoff click is not a sale or necessarily a qualified lead; those outcomes are measured separately.

The economic case

A recurring relationship can make the brand's existing digital audience more useful and valuable. In paid conversational media, an interactive brand experience can help an interested person make an informed next step. Neither claim proves better return than existing channels; controlled pilots must establish incremental contribution and total acquisition cost.

Our competitive advantage must be demonstrated: faster delivery than an equivalent internal build, stronger character and grounded answers, repeatable cross-channel deployment, lower equivalent ownership cost and better measured outcomes. The software and accumulated Gauntlet methodology should make each deployment more repeatable. We have not proved a unique market position merely by naming four evaluators.

Enterprise recommendation: brand-owned cloud and provider accounts, brand-controlled records and approvals, specialist agency development and maintenance, with a documented exit/handover. The brand pays underlying usage directly where applicable; our fees and media spending remain visible. Scale and costs depend on minutes and provider quotas, not on storing thousands of complete personality copies. See capacity and economics.

CEM and continuous improvement

Shared Digital DNA plus brand-specific character. A persona equaliser with presets, market-specific experiment proposals, authorised signal-driven adjustment proposals and reviewed controls for expression, priorities, retrieval budgets and approved destinations. Evaluate duration, useful exchanges, satisfaction and permitted voluntary return as well as handoffs; recommendations create testable improvement proposals. Privacy-aware performance records feed proposed changes through our Gauntlet; humans approve releases. Low-cost profiles retrieve less context without deleting the corpus or weakening essential rules. Larger context is not a guarantee of accuracy.

Proprietary value is intended to reside in the implementation, authored evaluation cases and rubrics, research process and accumulated operating expertise. We should demonstrate those assets rather than assert exclusive ownership of generic testing ideas.

Distribution and discovery

OpenAI is testing Sponsored Agents, but public documentation reviewed does not establish import, voice, updates or outward-click export specifications. Our aim is compatibility; seamless deployment is not yet verified. Agency accreditation, access and revenue sharing are not secured. See published specifications and gaps.

Website/app and other host adapters extend the same internal personality where supported. MCP connects tools; it does not guarantee persona reproduction or audience access. Publish helpful sourced public content with sound technical SEO; monitor organic discovery and AI citations rather than promise rankings. CEM is a management platform; calling it a media platform additionally requires audience or inventory we control.

Current proof and next milestone

Local Guinness and Pre voice prototypes, structured brand data, documented research, software checks and a local CEM review dashboard. Dashboard numbers/transcripts are synthetic. No Sponsored Agent deployment, private production logging, campaign billing or autonomous Gauntlet runner operates today.

Next proof: assess Sponsored Agent access/specifications, compare Pre against a baseline, measure quality and delivery cost, implement authorised handoff receipts in an owned-channel pilot, and approve the exact release. Successful creative and operating evidence should precede any claim of scale, media superiority or seamless compatibility.

Proposed differentiator: brand-owned athlete evidence

Public pages explain a shoe. Approved internal evidence could let Pre explain who wore that exact model, at which event, and with what result. Nike could contribute documented appearances and permitted testing/training records across college, elite and everyday athletes. We would establish what records actually exist during discovery, rather than assume a complete Nike database.

The agency turns the approved records into structured, retrievable knowledge, with exact model identification, linked results, publication rights, CEM visibility controls and Gauntlet evaluation. The brand retains control of confidential athlete information. No Nike internal connection is live. Discovery and a small reviewed pilot precede integration.