Both sides of ChatGPT. One team.
Your people use ChatGPT at work. Your customers use it to decide. We work on both. Inside: ChatGPT Business or Enterprise set up safely, every role trained, real processes automated on OpenAI's models. Outside: your brand in the answers, in ChatGPT Ads and in shopping results.
ChatGPT is where your people work, and where your customers decide.
Most companies run those as two projects, with two suppliers who never meet.
ChatGPT, used safely and well
- ChatThe assistant, with company knowledge under each person's own permissions.
- Work and CodexAgents on every seat: multi-step tasks across apps, files and a browser, plus a coding agent.
- The platformThe Responses API and the Agents SDK, running processes end to end, people approving what matters.
Your brand, found and chosen
- In the answerWhat ChatGPT says about you, and whether it cites you. Earned, not bought.
- Below the answerChatGPT Ads, labelled Sponsored and kept apart from the answer.
- In shopping resultsYour catalogue as a product feed, ready for discovery and for product ads.
One rule joins the two sides. OpenAI states that ads do not change ChatGPT's answers. So paid and organic are separate levers, measured in one view, with proper tests.
Your people already use ChatGPT. The question is how.
In most companies ChatGPT arrived bottom up: personal accounts, custom GPTs built in an afternoon, plugins switched on because a colleague did. Once agents can act across your apps, it is risky.
The fix is not a longer project: six settings, five inside and one outside, decided on purpose before company data goes in.
What we mean by fast, sophisticated and safe.
Shared skills and a working automation, on your own processes.
Plugins, skills and agents built on your procedures, each with an owner and tests.
Connections scoped per person, every agent action logged, a person approving what cannot be undone.
The right AI for the job: a front door for your people, models for your agents.
Choosing AI is not picking a chat app. It is two decisions: the workspace your people use every day, and the models your agents run on. Nobody pays us to pick OpenAI: we decide by your data, your own cases and cost.
Gemini Enterprise
Google's governed front door, on Google Workspace or Microsoft 365.
Pick it when- Your people work in Google Workspace or Microsoft 365, and data sits across several systems
- You want one permission-aware search, with agents over it
- Video, multimodal work or very high volume at low unit cost
Microsoft 365 Copilot
Microsoft's assistant, for an all-Microsoft estate.
Pick it when- Files and identity live in Microsoft 365 and Entra
- IT wants a single vendor and in-tenant controls
- Your agents will live in Teams and Copilot Studio
Claude
Anthropic's thinking and building tool.
Pick it when- Agents use many tools over many steps, or the job is agentic coding
- The work is long documents, careful writing and complex instructions
- Power users need depth more than search
ChatGPT / OpenAI
The assistant your people already know, and OpenAI's models.
Pick it when- Your people already live in ChatGPT, so change costs least
- Voice and images are core, or bulk work needs very cheap good models
- ChatGPT is also a marketing channel for you: answers, ads and shopping
Or a mix. Most of our automations route by task, with a fallback provider and a cost ceiling. Pick the AI that fits, not the one that pays us.
When we would not pick OpenAI.
- A small team needs data kept in the EU. ChatGPT Business has no data residency; it takes Enterprise, the EU API endpoint or a sovereign design.
- Everything is Microsoft and IT wants one vendor. Copilot is the shorter path.
- Video is the core of the job. OpenAI has had no video API since September 2026; our AI Factory works across models.
When you do not need us.
- You only want seats. Buy them from OpenAI directly.
- A one-off prompt problem. You need a short training session, not a programme.
- You want full autonomy, no human in the loop. For sending, paying, publishing or deleting, we always build an approval step.
Five services. Enablement first, then execution, then growth.
Most of the value sits in the first two: people using ChatGPT well, every day. Then automation, then your brand outside, on a foundation your CISO approves.
ChatGPT Business and Enterprise, set up properly
From plan choice to daily use, team by team.
- Plan and seat mix decided on real use, before you sign
- SSO, SCIM, roles, verified domain, retention and spend limits configured
- Chat, Work and Codex switched on for the right people
- Plugins and connectors governed; company knowledge connected under each person's permissions
Training by role, and a library you own
People trained on their own work, not on features.
- Tracks by role, from literacy to builders
- A shared library of skills and plugins, each with an owner
- Custom GPTs migrated to plugins before they break
- A monthly adoption view and a day-90 keep, expand or cut decision
Automation and agents, end to end
Built and run on the right model per step.
- The Responses API, the Agents SDK and, where it fits, the Agents API (beta)
- Workspace agents and team tasks inside ChatGPT, for work that lives there
- Voice agents on OpenAI's realtime voice, with handoff to people
- Evals before launch, approvals, cost per task under control
ChatGPT Ads, answers and product feeds
Paid and earned, in one view.
- ChatGPT Ads campaigns: context hints, creative, bids, the pixel
- Answer engine optimization: what ChatGPT says about you, measured weekly
- Product feeds for shopping discovery and product ads
- Incrementality tests, with a holdout every time
Governance, security and EU data
Guardrails your legal team approves, from day one.
- EU AI Act and GDPR: roles, literacy records, transparency notices
- EU data residency configured, and its limits written down
- Wise Shield, our LLM and agent firewall, between agents and your tools
- Compliance logs in your SIEM, agent traces, spend alerts
Where OpenAI pays, team by team.
Start from a real bottleneck, not a licence. The patterns we build most, marketing first.
Both sides, for marketing.
Inside: a brand-voice skill, ad variants checked by a brand grader, a weekly report computed in code. Outside: what ChatGPT says about you, tracked weekly, next to your ChatGPT Ads.
Talk about the marketing offer →| Team | The bottleneck | What we build | Result |
|---|---|---|---|
| Marketing | Dozens of variants per campaign, and a weekly report that eats a day. | The offer above: brand-voice skill, graded variants, the Monday report. | Faster production, a report on Monday |
| E-commerce | A catalogue written for a shop page, not for AI shopping. | The feed pipeline shown below: photo to fields, copy in two languages, checks, a person approving each batch and every flag. | Products ready for ChatGPT shopping |
| Customer service | Calls and tickets that start with the same five questions. | A voice or chat agent that looks up orders and policies, with a person approving refunds and changes. | Faster resolution, better experience |
| Marketing ops | Tracking and landing-page tickets wait weeks for a developer. | Codex in the repository: pixel with consent, server-side conversions, tests, a reviewed pull request. | Tracking you can trust, sooner |
OpenAI reports that BBVA saw nearly 3 hours saved per employee per week before rolling ChatGPT Enterprise out to all 120,000 employees. Vendor-reported, so we never promise it: we measure yours.
The answer is not for sale. The space below it is.
Since 24 August 2026 ChatGPT shows ads in Portugal and 30 other European markets, and since 31 August any company here can buy them self-serve. Wise Pirates already runs ChatGPT Ads campaigns for clients.
Not in the answer? The ad below it is the paid lever, AEO the organic one, and a holdout shows what the ad added.
What it is today.
- Free and Go users only. Plus, Pro, Business and Enterprise are ad-free.
- Below the answer, labelled Sponsored. Never to accounts identified as under 18, never near health, mental health or politics.
- Contextual only in Europe. No personalisation on past chats or memory.
- Three formats. Standard ads, shopping cards from a product feed, and "message the advertiser".
The ChatGPT Ads test: context hints, the Pixel and Conversions API, a holdout, a written read at day 30.
Book a meeting →How we run it.
- Landing pages open to OpenAI's crawlers. OAI-AdsBot reads the page to judge relevance, so we check it is let in.
- Context hints, not keywords. Up to 2,000 per ad group, written as needs, in pt-PT and English. Advertisers never see the conversation, so we judge hints by landing pages, conversions and the holdout.
- Bids and budgets, a person in charge. Impressions, clicks or conversions. Daily spend can reach twice the daily budget on a given day: we watch pacing, and a person approves every bid strategy, cap or budget change; automated bidding runs inside those limits.
- The Measurement Pixel and the Conversions API, on consent, deduplicated by event id. The click window fits your buying cycle (7, 14 or 30 days); view-through is reported apart.
- Reports exported every week, because product and hourly reports are kept for about 30 days.
- Shopping ads from your product feed, built by the automation below.
The channel is new. OpenAI says ChatGPT Ads does not yet have performance benchmarks, delivery can fluctuate in the beta and spend reporting can lag by hours. So we start small, write success criteria first and scale only what the holdout proves. Finance and health brands face strict eligibility: we check it first.
Right-size the seats. We do not sell them, so cutting costs us nothing.
Two seat types, two plans and credits make ChatGPT spend hard to forecast. A licence seller wants more seats. We want yours used.
Premium on evidence, not job title
A Premium seat gives about five times the usage and no five-hour limit. Seats mix and reassign, so Premium follows the usage data.
The plan your needs require
Business: 2 to 200 people, SSO, no residency. Enterprise: SCIM, key management, the Compliance API, custom retention and EU residency.
Turned into workflows, or cut
Analytics show who never opens ChatGPT. Some only need @ChatGPT in Slack or Teams, which works without their own seat.
We do not train people on ChatGPT. We train them on their work, with ChatGPT.
ChatGPT now runs multi-hour tasks across your apps. The skill is no longer prompting: it is knowing what to hand over and what to check.
- 1EveryoneLiteracy firstWhat ChatGPT is, what never goes in, how to check its work.
- 2Power usersWork, by functionProjects, Work mode and deep research, on the team's own tasks.
- 3BuildersSkills, plugins, agentsWorkspace agents, Codex and the API, reviewed before anything ships.
- 4ChampionsOne per teamAbout one per 15 to 25 people, with protected time, owning their team's skills.
A library you ownKnow-how that stays when people leave
- Every skill and plugin has an owner, a version and test cases
- Records you can show under the AI Act literacy duty
Custom GPT rescueBefore the migration breaks them
- Every GPT listed with its owner, use and custom actions
- The ones in use migrated to plugins and skills, custom actions rebuilt as MCP
Portuguese, properlypt-PT, not pt-BR by accident
- Delivery and training in European Portuguese, or English
- A pt-PT style skill shipped with every rollout
How we measure it.
A small first step. A written decision at day 90.
Two-week sprints, a weekly demo, every decision in writing. Start at any step.
- Step 1Working sessionHalf a day with leadership: where to start, inside and outside.
- Step 2Discovery sprint2 to 3 weeks, fixed scope: current use, seats, use cases scored, a baseline.
- Step 3Pilot4 to 8 weeks: 20 to 50 users, first skills, one or two automations, an ads test.
- Step 4RolloutTeam by team, with training and the monthly adoption view.
- Step 5Centre of competenceYour team owns both sides, with us or alone: an AI lead (plugins, evals, upgrades, costs), a performance lead (ad account, Pixel, Conversions API), a marketing analyst (the AEO panel).
One product, ready for AI shopping. The right model per step.
A product changes in your shop; minutes later it is ready for ChatGPT shopping, ChatGPT Ads and Google, in pt-PT and English. A workflow tool owns the trigger, OpenAI's models judge, a merchandiser approves.
$0.0516 for a flagged product, on GPT-6.1 Sol. Flags are the exception, not the rule.
Built and run by our Process Automation team, with feeds managed by Feed'em and every run traced by Agent Observability.
Our go-live checklist. Six lines, all ticked.
Nothing touches real company data, or real ad budget, until every line is ticked.
- 01Company accounts only, SSO on, verified-domain restriction set
- 02Plugins allowlisted, connections scoped per person, write actions approved
- 03Memory, retention and the credit overage limit set to policy
- 04Evals passed and model versions pinned; a fallback for stopped tasks
- 05Human approval on every irreversible step; logs flowing to your SIEM
- 06Pixel on consent, holdout defined, AI Act records and notices in place
Need the data to stay in Europe? Know where the edges are.
OpenAI offers real EU data residency, but not on every plan, tier or feature. We configure it and write down the gaps.
What stays outside: beta features, third-party GPTs, Codex Web and Sites; some CPU processing and system data; third-party plugins and MCP servers. The Agents API is US-only today, so EU agents run on the Agents SDK and the Responses API. For stricter needs, our Sovereignty area goes further, with our Cloud Services team.
Ten questions your CTO should ask. Our answers.
Start with the three in black: identity, connections and the firewall.
01Identity and access
Company accounts only, from your identity provider.
- SSO, MFA and domain verification, with the verified-domain restriction so company accounts cannot join personal workspaces
- SCIM, role-based access, IP allowlisting and enterprise key management on Enterprise
Business has no SCIM: we add a leaver process.
03Plugins, connectors and MCP
Every connection is a permission you grant.
- A short allowlist, read-only first; an owner, data scope and removal contact per plugin
- Developer mode and custom MCP for named builders only; internal servers through the Secure MCP Tunnel
06Firewall and observability
A control point between agents and your systems.
- Wise Shield inspects prompts, tool calls and outputs against your policy, with Agentic Security
- Traces per run: tokens, tools, latency, cost and eval score, with Agent Observability
02Data handling
No training on business data, by default.
- No training by default on Business and Enterprise; on the API, not since March 2023
- Zero data retention for eligible API customers; not on Business
04Agents that act
ChatGPT Work, Dots and computer use, on a leash.
- No "Always allow" websites for Work's browser; Dots off until a pilot proves them
- Approval before any send, pay, publish or delete
05Retention, memory and audit
Nothing left on default.
- A memory policy decided, custom retention set on Enterprise
- Compliance API logs last 30 days, so they go to your SIEM; we log tool calls too
07Evals and model churn
Three model releases in September alone.
- A golden set of 30 to 200 real cases per workflow; versions pinned; upgrades pass the evals first
- Misalignment monitoring can stop a legitimate task, so agents get retries and a fallback
08Agent runtime
The right harness for each job.
- Workflow tools own triggers; the Responses API with Structured Outputs; the Agents SDK for guardrails and approvals
09EU AI Act and GDPR
Your duties as a deployer, mapped.
- AI literacy (Article 4) with records; transparency duties (Article 50) from 2 August 2026
- High-risk uses flagged; a DPIA when connectors touch personal data
10Portability and exit
You can leave OpenAI, and you can leave us.
- Plugins run on MCP, evals are provider-neutral, skills live in your repositories
Switching model is a test run against your golden set, not a rebuild.
Model and product note, 1 October 2026: OpenAI's current models are GPT-6 Astra (3 September), GPT-6.1 Sol (29 September) and GPT-6 Luna (22 September). Agent Builder is legacy and shuts on 30 November 2026; the fine-tuning platform is winding down; the Sora video API closed on 24 September 2026; "Apps in ChatGPT" are now called Plugins. Names change monthly; our routing and evals do not depend on them.
ChatGPT is the start. Put it to work.
These teams sit right next to OpenAI, on both sides.
On a new project, the first step is an assessment: our discovery sprint.
Assessment is one of our specialties: how your teams use ChatGPT today and how your brand shows up in it. Talk to our specialists to see how this applies to your case.
The discovery sprint
Current use mapped, shadow AI included; plugins and GPTs listed; your brand in ChatGPT today; use cases scored; a ranked roadmap you keep.
The ChatGPT Ads test
Context hints, Pixel and Conversions API, a holdout, a written read at day 30.
The seat review
We match Standard and Premium seats to real use, find the seats nobody uses and turn them into workflows, or tell you to cut them.
Your business result. A great team and five pillars to move it.
Two pillars we run for you, three we own, and the crew at the centre answering for your number.
Judged on your KPIs, with part of our fee on the line.
OpenAI for business, answered.
Is "OpenAI for business" an OpenAI product?
ChatGPT Business vs ChatGPT Enterprise: which plan does a company need?
How do we get our brand recommended in ChatGPT answers?
Does OpenAI train its models on our company data?
Does ChatGPT Enterprise support EU data residency?
Are ChatGPT Ads available in Portugal?
Is there a minimum budget for ChatGPT Ads?
How do products appear in ChatGPT shopping results?
What happens to custom GPTs when they move to plugins?
Is it safe to connect ChatGPT to our tools with plugins and MCP?
ChatGPT vs Claude vs Gemini Enterprise vs Microsoft 365 Copilot: which is best for business?
Is OpenAI Agent Builder being discontinued?
Can you automate a whole process with OpenAI models, end to end?
Is ChatGPT Enterprise GDPR compliant?
Who implements ChatGPT Enterprise in Portugal?
Both sides of ChatGPT. Fast, sophisticated and safe.
Start with a working session or a discovery sprint: results on real work early, decisions in writing, everything in your own accounts.
OpenAI, ChatGPT and the OpenAI logo are trademarks of OpenAI. The logo is shown for identification only. Wise Pirates is an independent agency and is not affiliated with or endorsed by OpenAI. Other product names are trademarks of their owners.
