AI readiness test for agencies
The five #Levels
The playbook takes an agency through five levels: understanding AI, choosing the right AI, protecting client data, building AI into workflows, and running AI agents. Each level ends with a checklist your team can tick off, so progress is visible and the training is documented.
AI basics for agencies
Frontier models, free vs paid, chatbot vs agent, MCP in plain words.
LEVEL 1Choosing your AI
ChatGPT, Claude, Gemini or Copilot, the criteria that matter and what it costs.
LEVEL 2Data, GDPR & security
The data traffic light, GDPR roles, the AI Act, local models and a contract clause.
LEVEL 3AI in agency workflows
Where AI fits, the 30-60-90 day plan, prompts and client house rules.
LEVEL 4Your first AI agent
Connect, instruct, schedule and keep humans in control.
Free #Tools inside the playbook
| Tool | What it gives you |
|---|---|
| AI readiness test | Your agency's level and where to start |
| AI chooser | Which AI fits your stack, needs and size |
| Client data traffic light | What data may go into which AI |
| AI clause generator | A draft clause for client contracts |
| One-page AI policy | Eight rules for the whole team |
| Hours calculator | How much time AI can return to your team |
| House rules generator | A reusable instruction block per client |
| Agent recipe builder | A ready-to-paste instruction for your first agent |
Why we wrote this #Playbook
Most agencies in Central Europe are somewhere between a free chatbot on a personal account and a vague plan to "do something with AI". Meanwhile the questions they ask us are always the same: which AI, is it legal with client data, where do we start, and what is this MCP thing. This playbook answers them in one place, in the order an agency actually needs them.
It is maintained by the team behind ZoomSphere, the social media management platform used by agencies and brands since 2011, and one of the first in the region to open its platform to AI agents through MCP. We run these workflows ourselves every week, so the advice here is tested on real client work, not written from a press release.
Edited by Jakub Mach, founder and CEO of ZoomSphere. This playbook is general guidance and not legal advice; for contracts and GDPR decisions, consult your lawyer or data protection officer.
Changelog
- October 28, 2026: Added quick answers on labelling AI-written posts and on proving human review, based on the European Commission's draft transparency guidelines published May 8, 2026. Noted the Digital Omnibus grace period: generative AI systems placed on the market before August 2, 2026 have until December 2, 2026 to meet the marking obligation.
- October 5, 2026 (Edition 2026.4): Playbook published with five levels and eight tools. Legal status reviewed: Digital Omnibus (Regulation (EU) 2026/1744) in force since July 27, 2026; AI Act Article 50 transparency rules apply since August 2, 2026; EU-US Data Privacy Framework in force with an appeal pending at the Court of Justice.
Quick #Answers
How should a social media agency start with AI?
Start by moving client work from free personal AI accounts to one business AI plan with no training on your data and a data processing agreement. Then pilot three workflows with three to five people for 30 days, write house rules for every client, and only then connect your tools and build the first agent.
Is it legal for an agency to use ChatGPT or Claude with client data?
Yes, on business plans with a data processing agreement, with the client's authorisation of the AI vendor as a sub-processor and with appropriate transfer safeguards. Putting client personal data into free consumer AI accounts is the risky part, because there is usually no DPA and chats may be used for training.
Which AI is best for a marketing agency?
The one that fits your stack and your work. Google Workspace agencies get Gemini with business protections, Microsoft 365 agencies should evaluate Copilot, and agencies that prioritise writing quality and connected agents usually choose Claude Team or ChatGPT Business. Use the AI chooser in Level 1.
What is MCP and why does it matter for agencies?
MCP (Model Context Protocol) is an open standard that lets AI assistants such as Claude or ChatGPT work directly in other software with the permissions you grant. For agencies it means the AI can read analytics, draft posts into the content calendar and prepare reports without copy-pasting. In ZoomSphere you create an account, add a Scheduler and connect ZoomSphere MCP to Claude or ChatGPT; the setup takes two minutes and the prompt library gives you ready instructions to start with.
What does the EU AI Act require from marketing agencies?
Mainly two things: taking measures to support AI literacy of staff (Article 4, applicable since February 2, 2025 and amended by the 2026 Digital Omnibus) and transparency for deepfakes and certain AI-generated text (Article 50, applicable since August 2, 2026). Typical agency use is not high-risk AI.
Do agencies have to label AI-written social media posts in the EU?
Not always. Article 50(4) of the AI Act requires a disclosure when AI-generated text is published to inform the public on matters of public interest, unless a human genuinely reviewed it and an identifiable person or organisation holds editorial responsibility for it. The Commission's draft guidelines (May 8, 2026) say spell-checking or a cursory sign-off is not enough. Realistic AI images and videos of people, places or events (deepfakes) used for commercial purposes should be labelled. Until the final guidelines are out, document who reviewed and who approved every AI-assisted post.
How can an agency prove that a human reviewed AI-drafted posts?
Keep the review inside a tool that records it. In ZoomSphere every AI draft goes through your own approval stages: your team edits it, the client approves the whole batch from one e-mail as a free client user, and the activity log records who changed the copy or the visual and who approved it, with a timestamp. That record is your evidence of genuine human review and editorial responsibility.
Do we need a developer to use AI agents?
No. Connecting an MCP connector in Claude or ChatGPT takes a few clicks with OAuth sign-in, and agents are instructed in plain language. Level 4 walks through the first agent step by step.
Can I download the playbook as a PDF?
Yes. Every page has a "Save as PDF" button in the side panel that creates a clean printable version, so you can share a level with your team or attach it to your training records.