Most teams I meet already use AI, just not together. One person has a great prompt in a notes app, another pastes documents into a chat every morning, and the results depend on who did the work. The change I help companies make is moving from individual AI use to a shared system.
Shared workspace: the Claude Team plan
A Team plan gives every person a seat and gives the team shared projects with company knowledge. When I manage one, I set up who sees what, which projects exist, the usage rules, and the onboarding for new people, so everyone starts from the same context instead of a blank chat.
Skills: your best process, in one step
Skills are reusable instructions and resources that teach Claude how your company does a task. I build them from the way your best people work: brand voice, content briefs, SEO audits, reports, proposals. On my own site, a brand voice skill checks every page before it goes live. Quality stops depending on memory.
Connectors and agents
- Connectors let AI work with your documents, email, calendar and tools, with the permissions you decide.
- Agents take whole tasks: research, first drafts, reporting, follow-ups, connected through n8n, Make.com or Zapier.
- People keep the judgment: strategy, client relationships, final decisions and the fact-check before anything is published.
The new roles inside the team
Writers become editors and subject experts. Someone owns the skills and keeps them current. Someone reviews and signs off facts, which Google now explicitly asks for. None of this needs a bigger team. It needs a team that works from the same playbook.
Start small: one shared project, two skills for the tasks your team repeats every week, and one automation. Then measure and grow from what works.
A week in a team that works this way
Here's an example of what a week can look like once the system is in place. On Monday, an agent compiles a research brief from the sources the team trusts and the questions sales heard last week. On Tuesday, a writer spends twenty minutes interviewing the expert, and Claude drafts the piece with the content-brief skill. On Wednesday, an editor checks every fact with the review skill and logs what was corrected. On Thursday, an SEO check covers title, description, headings and alt texts before publishing. On Friday, a reporting agent pulls the numbers from Search Console and Google Analytics into a one-page summary.
The first skills I build
- Brand voice: how the company talks, with real examples of good and bad copy.
- Content brief: turns real customer questions into a structured brief with proprietary input.
- Review: the fact-check checklist, including title, meta description, structured data and alt text.
- SEO check: headings, internal links, duplicates and snippet settings before anything goes live.
- Monthly report: the same metrics, in the same format, every month.
Connectors, with care
Connectors are powerful because they give AI your real context, and that's also why they need rules. I give access by role, decide together with the team which documents each connector can reach, keep sensitive information out where it doesn't belong and use the admin controls of the business plan. Then I train people on what should and shouldn't be shared with AI.
Start agents with one boring process
The best first agent is usually the most boring one: the weekly report, the summary of new leads, the reminder to follow up. It's repetitive, easy to check and saves time from the first week. I connect it with n8n, Make.com or Zapier, and anything that goes out to a customer still gets approved by a person.
Training that sticks
Generic AI workshops are forgotten in a week. What sticks is practicing with real tasks from each person's job, clear rules about what to do and what to avoid, and a few habits repeated until they're automatic. I pick the tool for the task: ChatGPT and Claude for writing and analysis, Cursor for anyone who touches code or the website, and Grok Bot when the team works on X.
What I never automate
- The final fact-check before publishing.
- Conversations with clients that need judgment and trust.
- Strategy decisions.
- Claims about health, money or legal matters without an expert signing them.
How to know it is working
I look at four things: the time each recurring task takes before and after, the corrections in the review log going down over time, the publishing pace compared with how quickly new pages get indexed, and the leads that come from the content. If those move in the right direction, the system is working.
Clear ownership
- One person owns the skills and keeps them current when processes or tools change.
- A named reviewer signs off facts before anything is published.
- Someone decides which data each connector can reach.
- Someone watches the numbers and brings them to the team every month.
Mistakes I see again and again
- Company data pasted into personal accounts instead of a managed workspace.
- The best prompts living in one person’s notes app.
- AI drafts published without anyone reading them in full.
- Automating a broken process, which only makes it break faster. Fix the process first.
- Starting without measuring how long things take today, so nobody can tell what improved.
I train teams on ChatGPT, Claude, Cursor and Grok Bot with tasks from their own work, because that's where habits stick. If you want your marketing team working this way, let's build it together.
Sources
- Google Search's guidance on using generative AI content on your website · Google Search Central, updated October 1, 2026
- Creating helpful, reliable, people-first content · Google Search Central, updated October 5, 2026
How I made this article: I researched and drafted it with the help of Claude, then checked every fact myself against the official documentation listed in the sources.